CN116187514A - Method, device, terminal and storage medium for predicting building energy efficiency - Google Patents
Method, device, terminal and storage medium for predicting building energy efficiency Download PDFInfo
- Publication number
- CN116187514A CN116187514A CN202211586076.8A CN202211586076A CN116187514A CN 116187514 A CN116187514 A CN 116187514A CN 202211586076 A CN202211586076 A CN 202211586076A CN 116187514 A CN116187514 A CN 116187514A
- Authority
- CN
- China
- Prior art keywords
- building
- model
- energy efficiency
- local
- federal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 122
- 230000005012 migration Effects 0.000 claims abstract description 191
- 238000013508 migration Methods 0.000 claims abstract description 191
- 238000012549 training Methods 0.000 claims abstract description 178
- 238000005265 energy consumption Methods 0.000 claims abstract description 115
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 58
- 230000002776 aggregation Effects 0.000 claims description 65
- 238000004220 aggregation Methods 0.000 claims description 65
- 230000008569 process Effects 0.000 claims description 47
- 238000012423 maintenance Methods 0.000 claims description 8
- 239000000463 material Substances 0.000 claims description 8
- 230000004927 fusion Effects 0.000 claims description 5
- 230000006870 function Effects 0.000 description 14
- 230000035945 sensitivity Effects 0.000 description 14
- 238000000605 extraction Methods 0.000 description 13
- 238000007726 management method Methods 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 12
- 238000010801 machine learning Methods 0.000 description 12
- 238000010276 construction Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 8
- 230000008901 benefit Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000010438 heat treatment Methods 0.000 description 5
- 238000004378 air conditioning Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000010206 sensitivity analysis Methods 0.000 description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000007477 logistic regression Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000001816 cooling Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000004134 energy conservation Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000004807 localization Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000000704 physical effect Effects 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000012550 audit Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000013076 uncertainty analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/08—Construction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Primary Health Care (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Operations Research (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Educational Administration (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
Abstract
The invention discloses a method, a device, a terminal and a storage medium for predicting building energy efficiency, wherein the method comprises the following steps: acquiring energy consumption parameters and operation data of a target building; selecting other buildings similar to the energy consumption parameters of the target building according to the energy consumption parameters of the target building; constructing a similar building migration network based on the target building and other buildings; training all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model by utilizing a federal learning algorithm to obtain a federal model which can be shared by all the buildings in the similar building migration network; and optimizing the federation model by utilizing the operation data of the target building to obtain a local federation model of the target building, so that the local federation model is utilized to predict the building energy efficiency of the target building. According to the scheme, a shared and movable federal model is constructed by establishing a migration network between similar buildings based on a federal learning algorithm, so that the energy efficiency prediction of the buildings with limited operation data is realized.
Description
Technical Field
The invention belongs to the technical field of building energy conservation, in particular relates to a method, a device, a terminal and a storage medium for predicting building energy efficiency, and especially relates to a method, a device, a terminal and a storage medium for predicting building energy efficiency based on federal learning.
Background
The energy consumption prediction is a foundation for improving the energy efficiency level of the building and realizing energy conservation and emission reduction, and is an important support for energy management tasks such as building optimization design, optimization control, demand side response, energy audit and the like. In the building design stage, the annual cold, heat and electric loads of the building can be rapidly predicted, so that the selection of building design parameters can be guided; in the building operation stage, according to the building cold load demand in a future period, the chilled water supply temperature set point of the air conditioning system can be optimized, and even the chilled water supply temperature set point can participate in the power grid demand side response more flexibly; in the building energy auditing stage, the energy consumption prediction model can provide building reference energy consumption, so that the energy saving amount brought by energy saving reconstruction measures can be estimated more accurately.
In the related scheme, the data driving method is used for predicting the building energy efficiency, but the data driving method has higher requirements on the quantity and quality of data. However, in the practical project of energy efficiency prediction of a building, for a new building or an old building with an imperfect energy management system, there is often limited operation data, so that the data driving method is difficult to apply. How to solve the problem of building energy consumption prediction in a small sample scene is one of important research directions in the field of building energy consumption prediction.
The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present invention and is not intended to represent an admission that the foregoing is prior art.
Disclosure of Invention
The invention aims to provide a method, a device, a terminal and a storage medium for predicting building energy efficiency, which are used for solving the problems that in a related scheme, the data driving method is used for predicting the building energy efficiency, but the data driving method has higher requirements on the quantity and quality of building operation data, so that for the building with limited building operation data such as new building, old building with imperfect energy management system and the like, the data driving method is difficult to use for predicting the building energy efficiency, the effects of building energy efficiency prediction for the building with limited building operation data are realized by establishing a migration network between similar buildings based on a federal learning algorithm, obtaining a sharable and mobilizable federal model based on the common global training of all building main bodies in the migration network and locally fine-tuning the federal model.
The invention provides a method for predicting building energy efficiency, which comprises the following steps: acquiring energy consumption parameters of a target building, and acquiring operation data of the target building; according to the energy consumption parameters of the target building, selecting a building similar to the energy consumption parameters of the target building, and marking the building as other buildings; constructing a migration network based on the target building and the other buildings, and recording the migration network as a similar building migration network; training all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model by utilizing a federal learning algorithm to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model; optimizing the federation model by utilizing the operation data of the target building to obtain an optimized federation model, wherein the optimized federation model is used as a local federation model of the target building; and predicting the building energy efficiency of the target building by using the local federal model of the target building.
In some embodiments, selecting a building similar to the energy usage parameter of the target building, denoted as another building, according to the energy usage parameter of the target building, comprising: according to the energy consumption parameters of the target building, selecting a building with the similarity degree of the energy consumption parameters of the target building within a preset building range, and marking the building as other buildings; the target building and the other buildings have the similarity in at least one of building type, climate zone where the building is located and building structure characteristics within a set similarity range; the building structure characteristics include: at least one of a window wall ratio, a maintenance structural material, a building area, and a building orientation.
In some embodiments, the pre-constructed basic energy efficiency prediction model is a time sequence prediction method based on RNNs and BPNNs, and the pre-constructed mixed model fusing the RNNs and the BPNNs is obtained.
In some embodiments, training all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model by using a federal learning algorithm to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model, including: constructing a federal learning framework based on a pre-constructed basic energy efficiency prediction model by utilizing a federal learning algorithm; aiming at each building in the similar building migration network, enabling the building to perform local training by utilizing the federal learning framework based on the running data of the building to obtain training parameters of a local model of the building; aiming at training parameters of local models of all buildings in the similar building migration network, carrying out safe aggregation treatment to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model; under the condition that a new building is added into the similar building migration network, aiming at the new building, enabling the new building to perform local training by utilizing the federal learning framework based on the running data of the building to obtain training parameters of a local model of the new building; and carrying out safe aggregation treatment on the training parameters of the local model of the new building and the training parameters of the local model of all the buildings in the similar building migration network together to obtain a shared energy efficiency prediction model which can be shared by the new building and all the buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
In some embodiments, for each building in the similar building migration network, the building is made to perform local training by using the federal learning framework based on the operation data of the building, to obtain training parameters of a local model of the building, including: aiming at each building in the similar building migration network, on the local side of the building, enabling the building to perform local training based on the running data of the building, and using the basic energy efficiency prediction model in the federal learning frame to obtain training gradient of the local model of the building, and marking the training gradient as gradient information; encrypting the gradient information to obtain an encryption result, and taking the encryption result as a training parameter of a local model of the building; correspondingly, aiming at training parameters of local models of all buildings in the similar building migration network, carrying out security aggregation treatment to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model, wherein the method comprises the following steps of: aiming at the training parameters of the local models of all the buildings in the similar building migration network, carrying out safe aggregation treatment on the training parameters of the local models of all the buildings at a server side to obtain aggregation gradient information; and updating training parameters of the basic energy efficiency prediction model in the federal learning framework based on the aggregation gradient information to obtain a shared energy efficiency prediction model which can be shared by all buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
In some embodiments, wherein, for each building in the similar building migration network, at the local side of the building, the energy consumption characteristic parameter of the building is selected as the operational data of the building itself; the energy consumption characteristic parameters of the building comprise: at least one of time characteristics, weather information, historical energy consumption, and building structure information.
In some embodiments, optimizing the federation model using the operational data of the target building to obtain an optimized federation model as a local federation model of the target building includes: aiming at the target building, enabling the target building to perform local training by utilizing the federal model based on the running data of the target building to obtain training parameters of the local model of the target building; and optimizing the training parameters of the federal model by utilizing the training parameters of the local model built by the target, so as to obtain an optimized federal model which is used as the local federal model of the target building.
In accordance with another aspect of the present invention, there is provided an apparatus for predicting energy efficiency of a building, comprising: the system comprises an acquisition unit, a control unit and a control unit, wherein the acquisition unit is configured to acquire energy consumption parameters of a target building and acquire operation data of the target building; a control unit configured to select a building similar to the energy usage parameter of the target building, as the other building, according to the energy usage parameter of the target building; the control unit is further configured to construct a migration network based on the target building and the other buildings, and the migration network is recorded as a similar building migration network; the control unit is further configured to train all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model by utilizing a federal learning algorithm to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and record the shared energy efficiency prediction model as a federal model; the control unit is further configured to optimize the federation model by using the operation data of the target building to obtain an optimized federation model, and the optimized federation model is used as a local federation model of the target building; the control unit is further configured to predict a building energy efficiency for the target building using a local federal model of the target building.
In some embodiments, the control unit selects a building similar to the energy usage parameter of the target building, denoted as another building, according to the energy usage parameter of the target building, including: according to the energy consumption parameters of the target building, selecting a building with the similarity degree of the energy consumption parameters of the target building within a preset building range, and marking the building as other buildings; the target building and the other buildings have the similarity in at least one of building type, climate zone where the building is located and building structure characteristics within a set similarity range; the building structure characteristics include: at least one of a window wall ratio, a maintenance structural material, a building area, and a building orientation.
In some embodiments, the control unit pre-constructs a basic energy efficiency prediction model, which is a time sequence prediction method based on the RNN and the BPNN, and pre-constructs a hybrid model fusing the RNN and the BPNN.
In some embodiments, the control unit trains all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model by using a federal learning algorithm to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and the shared energy efficiency prediction model is recorded as a federal model, and includes: constructing a federal learning framework based on a pre-constructed basic energy efficiency prediction model by utilizing a federal learning algorithm; aiming at each building in the similar building migration network, enabling the building to perform local training by utilizing the federal learning framework based on the running data of the building to obtain training parameters of a local model of the building; aiming at training parameters of local models of all buildings in the similar building migration network, carrying out safe aggregation treatment to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model; under the condition that a new building is added into the similar building migration network, aiming at the new building, enabling the new building to perform local training by utilizing the federal learning framework based on the running data of the building to obtain training parameters of a local model of the new building; and carrying out safe aggregation treatment on the training parameters of the local model of the new building and the training parameters of the local model of all the buildings in the similar building migration network together to obtain a shared energy efficiency prediction model which can be shared by the new building and all the buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
In some embodiments, the control unit, for each building in the similar building migration network, makes the building perform local training by using the federal learning framework based on the operation data of the building, to obtain training parameters of a local model of the building, including: aiming at each building in the similar building migration network, on the local side of the building, enabling the building to perform local training based on the running data of the building, and using the basic energy efficiency prediction model in the federal learning frame to obtain training gradient of the local model of the building, and marking the training gradient as gradient information; encrypting the gradient information to obtain an encryption result, and taking the encryption result as a training parameter of a local model of the building; correspondingly, the control unit performs a security aggregation process on training parameters of local models of all buildings in the similar building migration network to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and the shared energy efficiency prediction model is recorded as a federal model, and comprises the following steps: aiming at the training parameters of the local models of all the buildings in the similar building migration network, carrying out safe aggregation treatment on the training parameters of the local models of all the buildings at a server side to obtain aggregation gradient information; and updating training parameters of the basic energy efficiency prediction model in the federal learning framework based on the aggregation gradient information to obtain a shared energy efficiency prediction model which can be shared by all buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
In some embodiments, the control unit selects, for each building in the similar building migration network, on a local side of the building, an energy consumption characteristic parameter of the building as the operation data of the building itself; the energy consumption characteristic parameters of the building comprise: at least one of time characteristics, weather information, historical energy consumption, and building structure information.
In some embodiments, the control unit optimizes the federal model using the operational data of the target building to obtain an optimized federal model as a local federal model of the target building, including: aiming at the target building, enabling the target building to perform local training by utilizing the federal model based on the running data of the target building to obtain training parameters of the local model of the target building; and optimizing the training parameters of the federal model by utilizing the training parameters of the local model built by the target, so as to obtain an optimized federal model which is used as the local federal model of the target building.
In accordance with another aspect of the present invention, there is provided a terminal comprising: the device for predicting the energy efficiency of the building.
In accordance with a further aspect of the present invention, there is provided a storage medium comprising a stored program, wherein the program, when run, controls a device in which the storage medium is located to perform the method of building energy efficiency prediction described above.
According to the scheme, a similar building migration network is obtained by establishing a migration network between similar buildings based on a federal learning algorithm, each building main body in the similar building network trains a local prediction model based on own building operation data and shares parameters of the local prediction model, and the parameters of all the local prediction models are aggregated through a safe aggregation algorithm to obtain a globally trained federal model; when a new building is added, the parameters of the local prediction model of the new building are only needed to be included in the process of parameter aggregation by the parameter noble safety aggregation algorithm of all the local prediction models; building energy efficiency prediction of the building with limited building operation data is achieved by carrying out local fine adjustment on the federal model, so that a migration network between similar buildings is established based on a federal learning algorithm, a sharable and movable federal model is obtained based on common global training of all building main bodies in the migration network, and the building energy efficiency prediction of the building with limited building operation data is achieved by carrying out local fine adjustment on the federal model.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
FIG. 1 is a flow chart of a method for predicting building energy efficiency according to an embodiment of the present invention;
FIG. 2 is a flow chart of an embodiment of training all buildings in the similar building migration network in the method of the present invention;
FIG. 3 is a schematic structural view of an embodiment of a device for predicting building energy efficiency according to the present invention;
FIG. 4 is a schematic flow diagram of a federal model training process;
FIG. 5 is a flow diagram of one embodiment of a method of building migration network construction based on federal learning;
fig. 6 is a schematic diagram of a training process of a building energy efficiency system model (i.e., a building energy consumption prediction model).
In the embodiment of the present invention, reference numerals are as follows, in combination with the accompanying drawings:
102-an acquisition unit; 104-a control unit.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to specific embodiments of the present invention and corresponding drawings. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In consideration of the fact that the related scheme utilizes a data driving method to predict the energy efficiency of the building, the data driving method can be specifically classified into a statistical regression method and a machine learning method, namely, various data driving methods are proposed based on different statistical regression or machine learning algorithms, and a great deal of research is conducted on different types of buildings (such as office buildings, residential buildings, commercial buildings, teaching buildings and the like) or different prediction objects (such as loads of cold loads, heat loads, electric loads and the like). When the data driving methods are used for predicting the building energy efficiency, firstly building operation data of a building to be predicted for the energy efficiency are obtained, the obtained building operation data are preprocessed, then the preprocessed building operation data are input into a data driving model, the data driving model comprises a Support Vector Regression (SVR) model, a Multiple Linear Regression (MLR) model, a gradient lifting tree (GBDT) model, a Random Forest (RF) model and the like, and finally output values of the corresponding models are used as predicted values of the building energy efficiency.
For example: taking a building air conditioning system as an example, fuzzy PID (i.e. proportional-integral-derivative) control based on load prediction can reduce the overshoot and the oscillation time of control actions, improve the control precision of the building air conditioning system, reduce the energy consumption of a chilled water system water pump, and save energy compared with a common PID control method; based on weather forecast data, controlling the running time of the floor radiant heating system by using a predictive control optimizer; according to the predicted value of the cooling load, searching the optimal operation parameters of the central air conditioning cold station system meeting the load demand through a particle swarm algorithm, wherein the optimal operation parameters are all optimized algorithm modes.
However, in the related scheme, the data driving method is used for predicting the building energy efficiency, and the time dimension characteristic and the space dimension characteristic of the building operation data are not fully considered. In the time dimension, the time sequence is regarded as unordered data, the importance of the data at each time point is the same, and a prediction model (namely, a data driving model) is input at the same time, so that the simplified processing mode is too rough, and effective information is lost. In the space dimension, only a large amount of data of a single building is used for training a prediction model only suitable for the building, and the data of a plurality of buildings cannot be effectively and comprehensively utilized, so that the building energy efficiency prediction is carried out by utilizing a data driving method in a related scheme in the face of the building with limited building operation data, and the prediction accuracy is not applicable or low.
Therefore, in order to cope with the shortfall in the time dimension and the space dimension of the building energy efficiency prediction using the data driving method in the related scheme. The scheme of the invention provides a building energy efficiency prediction method, in particular to a federal learning-based building energy efficiency prediction method, which is used for establishing a migration network between similar buildings based on a federal learning algorithm, obtaining a sharable and movable federal model based on common global training of all building main bodies in the migration network, and realizing the building energy efficiency prediction of buildings with limited building operation data by locally fine-tuning the federal model.
According to an embodiment of the present invention, a method for predicting energy efficiency of a building is provided, and a flowchart of an embodiment of the method of the present invention is shown in fig. 1. The method for predicting the energy efficiency of the building can comprise the following steps: step S110 to step S160.
At step S110, energy usage parameters of a target building (e.g., energy usage pattern of the target building) are acquired, and operation data of the target building is acquired if the target building has operation data.
At step S120, a building similar to the energy usage parameter of the target building is selected according to the energy usage parameter of the target building, and is recorded as another building.
In some embodiments, in step S120, a building similar to the energy consumption parameter of the target building is selected according to the energy consumption parameter of the target building, and denoted as another building, including: and selecting the building with the similarity degree of the energy consumption parameters of the target building within the preset building range within the set similarity degree range according to the energy consumption parameters of the target building, and marking the building as other buildings.
The target building and the other buildings are in a set similarity range in at least one of building type, climate zone where the building is located and building structure characteristics. The building structure characteristics include: at least one of a window wall ratio, a maintenance structural material, a building area, and a building orientation.
FIG. 5 is a flow diagram of one embodiment of a method of building migration network construction based on federal learning. As shown in fig. 5, the building migration network construction method based on federal learning includes: step 21, similar building selection. Similar building selection is intended to select a building that is similar to the energy usage pattern of the target building (e.g., new building or existing building with limited data).
In step 21, when similar building selection is performed, when the characteristics of two buildings are different, there is a significant difference in their energy usage patterns. When building a building migration network, the operation data of similar buildings in energy consumption mode is beneficial to federal model training. Wherein similar buildings in terms of building type, physical characteristics, geographical position and the like often have similar energy usage patterns, and the generated operation time sequence data have similar change rules. Such as: taking office buildings as an example, the energy consumption is generally higher in the daytime on the workday, and lower in the nighttime and the rest days on the workday.
When a similar building is specifically selected, three main factors influencing the energy consumption of the building are mainly considered: building type, climate zone in which the building is located, and building structural characteristics. The first factor is the type of building (e.g., different buildings such as commercial buildings, commercial residential areas, villas, etc.), and the energy patterns for different types of buildings are significantly different. The second factor is that the building in the climatic zone is high in energy consumption for heating in winter in cold climates, while the building in the hot climates does not need heating in winter. The third factor is building structural characteristics such as window wall ratio, maintenance structural materials, building area and building orientation, etc., which all affect building energy consumption. Two buildings are similar in type, in the same climate zone, and similar in structural characteristics.
At step S130, a migration network is constructed based on the target building and the other buildings, denoted as a similar building migration network.
At step S140, training all buildings in the similar building migration network based on the pre-constructed basic energy efficiency prediction model by using a federal learning algorithm to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model.
The pre-constructed basic energy efficiency prediction model in step S140 is a time sequence prediction method based on RNN and BPNN, and the pre-constructed hybrid model fusing RNN and BPNN is obtained.
Specifically, since the building energy system has characteristics of various kinds of equipment, complex topology, strong nonlinearity, large time lag, strong multi-system coupling, and the like, the building energy consumption is taken as typical time series data, and the change thereof is affected by various factors such as physical properties of the building, outdoor weather conditions, personnel use behaviors, equipment operation conditions, socioeconomic changes, geographical locations, and the like. How to comprehensively consider the internal and external influence factors, effectively utilize new technologies such as big data, artificial intelligence and the like, and construct an accurate and reliable energy consumption prediction model is an important research content in the building field. Therefore, in order to construct a new, non-fully connected form of the feature engineering model, the order of the time series is maintained throughout the feature extraction process. In the scheme of the invention, a time sequence prediction method based on a neural network such as an RNN (recurrent neural network) and a BPNN (BP neural network) is provided. The time sequence prediction method constructs a mixed model fused with RNN and BPNN so as to couple the feature extraction and the energy consumption prediction into the same model. RNNs have unique timing structures for extracting features from the timing data of building operational data. The BPNN has strong nonlinear fitting capability and is used for establishing a mapping relation between the extracted features and the energy consumption.
The feature extraction and the energy consumption prediction are coupled in the same model, the model fusion can be performed by combining the advantages of two algorithms, and the advantages of the two algorithms are written later. In the model training process, the minimum prediction error is taken as a target, and the feature extraction process and the energy consumption prediction process are monitored at the same time, so that the two processes are simultaneously optimized. The feature extraction stage adopts RNN to extract features, the energy consumption prediction stage adopts BPNN to output predicted values, meanwhile, the feature extraction process and the energy consumption prediction process are tightly coupled, feature extraction and energy consumption prediction are guided by taking the minimum prediction error as a target, and finally effective features are extracted from building operation data with time characteristics, so that the prediction accuracy is improved.
In some embodiments, in step S140, training all buildings in the similar building migration network based on the pre-constructed basic energy efficiency prediction model by using a federal learning algorithm to obtain a shared energy efficiency prediction model that can be shared by all buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a specific process of the federal model, see the following exemplary description.
The following is a schematic flow chart of an embodiment of training all the buildings in the similar building migration network in the method of the present invention in connection with fig. 2, which further describes a specific process of training all the buildings in the similar building migration network in step S140, including: step S210 to step S230.
Step S210, constructing a federal learning framework based on a pre-constructed basic energy efficiency prediction model by using a federal learning algorithm.
Step S220, aiming at each building in the similar building migration network, enabling the building to perform local training by utilizing the federal learning framework based on the running data of the building, and obtaining training parameters of a local model of the building.
Step S230, aiming at training parameters of local models of all buildings in the similar building migration network, carrying out safety aggregation treatment to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model.
Under the condition that a new building is added into the similar building migration network, aiming at the new building, the new building is enabled to perform local training by utilizing the federal learning framework based on the running data of the building, and training parameters of the local model of the new building are obtained. And carrying out safe aggregation treatment on the training parameters of the local model of the new building and the training parameters of the local model of all the buildings in the similar building migration network together to obtain a shared energy efficiency prediction model which can be shared by the new building and all the buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
In the scheme of the invention, in order to couple the feature extraction and the energy consumption prediction of the mixed model fusing the RNN and the BPNN, the feature extraction and the energy consumption prediction are performed in a supervision mode, so that the two are simultaneously optimized. According to the scheme, a similar building migration network is built based on federal learning (Federated Learning), each building main body in the similar building migration network trains a local prediction model based on own building operation data and shares parameters of the local prediction model, and after the parameters of all the local prediction models are aggregated through a secure aggregation algorithm, the parameters of all the local prediction models are used for updating the shared and movable federal model.
When a new building is added, only parameters of a local prediction model of the new building are needed to be included in the parameter aggregation process, all the buildings are not a unidirectional and fixed migration chain, but a multidirectional, flexible and extensible similar building migration network is formed, and the problem that the building energy efficiency prediction is difficult to carry out by using a data driving method for buildings with limited building operation data such as old buildings with new buildings and imperfect energy management systems is solved.
Preferably, in step S220, for each building in the similar building migration network, the building is locally trained by using the federal learning framework based on the operation data of the building itself, to obtain training parameters of a local model of the building itself, including: aiming at each building in the similar building migration network, on the local side of the building, the building is enabled to perform local training based on the running data of the building, the basis energy efficiency prediction model in the federal learning framework is utilized to obtain training gradient of the local model of the building, and the training gradient is recorded as gradient information. And encrypting the gradient information to obtain an encryption result, and taking the encryption result as a training parameter of the local model of the building.
Accordingly, in step S230, for training parameters of the local models of all buildings in the similar building migration network, a secure aggregation process is performed to obtain a shared energy efficiency prediction model that all buildings in the similar building migration network can share, which is denoted as a federal model, and the method includes: aiming at the training parameters of the local models of all the buildings in the similar building migration network, carrying out safe aggregation processing on the training parameters of the local models of all the buildings at a server side to obtain aggregation gradient information. And updating training parameters of the basic energy efficiency prediction model in the federal learning framework based on the aggregation gradient information to obtain a shared energy efficiency prediction model which can be shared by all buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
In particular, in the framework of federal learning, data from different ownersThe data of the system can be safely used, and privacy disclosure can be effectively prevented. FIG. 4 is a flow chart of a federal model training process in which the federal learning framework includes K participants (p 1 ,p 2 ,Kp k ) The goal of the participants is to utilize the respective data (D) securely by some means 1 ,D 2 K,D K ) To train better machine learning model to complete the task of prediction or classification, K is a positive integer. The federal learning framework is essentially a model training paradigm in which all participants co-train a federal model with the aid of the cloud or a third party (the federal model performs better than a model trained using the participants' own data) without exposing their original data to other participants or third parties. Here, all participants are virtually independent of each other, each having a target building that is not identical.
In the example shown in fig. 4, (1), (2), (3) and (4) respectively represent: (1) representing calculating training gradient of the local model, (2) representing safe aggregation and updating the federal model, (3) representing issuing the federal model, and (4) representing updating the local model.
In the example shown in fig. 4, there are four levels: a participant layer, a local model layer, a communication layer, and a cloud layer. At the participant level, each participant has data stored at its local server and does not send any data containing sensitive information to the cloud or other participant's server. At the local model layer, each participant stores a respective local model at its local server. And at the communication layer, the participants and the cloud end transmit information. At the cloud level, the cloud server stores the federal model. The interaction between the four hierarchies includes the following four steps:
And 11, each participant calculates training gradient of the local model based on own data, encrypts gradient information through homomorphic encryption, differential privacy or secret sharing and other technologies, and uploads the encryption result of the participant to the cloud.
And step 12, the cloud server safely aggregates the encryption results of all participants, and updates the parameters of the federation model based on the aggregation gradient to obtain new parameters of the federation model.
And 13, the cloud transmits new parameters of the federal model to each participant. Wherein the new parameters of the federal model issued by the cloud to different participants are the same.
Step 14, each participant receives new parameters of the federal model issued by the cloud and updates the parameters of the local model of the participant.
Thus, the training process of the federal model continuously repeats the four steps until the parameters of the federal model converge. The federal model training process is applicable to various machine learning models, such as LR (Logistic Regression ), BPNN (BP neural network), DNN (deep neural network), etc., and all participants share the final federal model.
And selecting the energy consumption characteristic parameters of each building in the similar building migration network as the operation data of the building on the local side of the building. The energy consumption characteristic parameters of the building comprise: at least one of time characteristics, weather information, historical energy consumption, and building structure information. The energy consumption characteristic parameter of the building is a characteristic parameter related to the energy consumption of the building among the characteristic parameters of the building.
Specifically, as shown in fig. 5, the building migration network construction method based on federal learning further includes:
step 22, feature selection. Feature selection, aimed at selecting the appropriate feature as the input of the predictive model.
In step 22, the feature selection is performed by selecting the feature related to the building energy consumption as the input of the prediction model. Features related to building energy consumption, such as time features, weather information, historical energy consumption, building structure information and the like, are features commonly used in energy consumption prediction models. The time profile reflects the impact of the number of people and activities inside the building on energy consumption. Weather information is a major driver of the cold/heat load of a building.
Step 23, training and optimizing the federal model. Federal model training, which aims to train a mobilizable federal model using data of similar buildings securely.
In step 23, based on the constructed similar building migration network, all buildings cooperate to train a federal model, and the training process comprises the following three steps:
step 231, selecting an appropriate data-driven algorithm (i.e., data-driven model) as the base prediction algorithm (i.e., base prediction model).
And 232, training the basic prediction model by adopting a safe aggregation algorithm to obtain a federal model (namely a building energy efficiency prediction model).
Step 233, optimizing super parameters of the federal model.
Fig. 6 is a schematic diagram of a training process of a building energy efficiency system model (i.e., a building energy consumption prediction model). As shown in fig. 6, the process of training the basic prediction model to obtain the federal model (i.e., the building energy efficiency prediction model) in step 232 by using a security aggregation algorithm includes:
in the training process, data processing and normalization are performed first.
Step 32, determining the number of nodes, network weights, basis function widths and central parameters required by the neural network.
And 33, judging whether the error meets the requirement according to the prediction model of the building energy efficiency system.
And step 34, outputting the final model as a federal model (namely, a building energy efficiency prediction model) if the requirements are met.
In the scheme of the invention, in order to construct an extensible and flexible similar building migration network, any building in the similar building migration network can share a migration model (namely a shared and movable federal model), so that the migration effect is ensured, and repeated training is avoided. According to the scheme, the energy consumption prediction model (namely the shared and movable federal model) is adopted for carrying out federal learning-based building energy efficiency prediction, and the sensitivity index input by the energy consumption prediction model is calculated firstly based on a sensitivity analysis method so as to measure the influence of the sensitivity index input by the energy consumption prediction model on the energy consumption prediction value. And then calculating a weighted Manhattan distance based on the sensitivity index to quantify the difference between different working conditions and obtain a prediction error. And finally, fitting a relation between the prediction error and the weighted Manhattan distance to infer a change rule of the prediction error of the energy consumption prediction model and evaluate the reliability of the energy consumption prediction model.
The sensitivity analysis method is an uncertainty analysis method for finding out the sensitivity factors which have important influence on the economic benefit index of the investment project from a plurality of uncertainty factors, analyzing and measuring the influence degree and the sensitivity degree of the sensitivity factors on the economic benefit index of the project, and further judging the risk bearing capacity of the project. Here, the sensitivity index of the predictive model input can be better found by using a sensitivity analysis method.
The weighted manhattan distance is calculated based on the sensitivity index, and may specifically be: the input features x of the predictive model and the building energy consumption y are commonly characterized, and the model input and the energy consumption are simultaneously considered when the weighted Manhattan distance (the following formula 1) is calculated. The weighted manhattan distance calculation formula in the prediction field is as follows in formula 2. The weight of each model input feature is the ratio of the sensitivity index of that feature to the sum of the sensitivity indexes of all features. The distance D (p) between an unknown predicted condition p and a known training condition is the minimum of the weighted manhattan distances for the predicted condition and all training conditions, as shown in equation 3.
D(p)=min(D([x p ,y p ],[x m ,y m ]))(m=1,2,…M) (3)。
Wherein d (x 1 ,x 2 ) Is x 1 And x 2 Manhattan distance between two points, point x 1 Can be expressed as an n-dimensional vector (x 11 ,x 1,2 ,Λ,x 1,n ) Point x 2 Can be expressed as an n-dimensional vector (x 2,1 ,x 2,2 ,Λ,x 2,n ),D([x 1 ,y 1 ],[x 2 ,y 2 ]Is the working condition [ x ] 1 ,y 1 ]And operating mode [ x ] 2 ,y 2 ]Is a weighted Manhattan distance, I (x i ) For inputting characteristic x i Sensitivity index, w y The energy consumption weight is that y is the energy consumption under a certain working condition, and D (p) is the minimum weighted Manhattan distance between the predicted working condition p and all M training working conditions, which can be abbreviated as D. The weighted manhattan distance D from the training condition represents the difference between the two conditions, and the larger the distance is, the larger the difference between the two conditions is, the larger the error of the prediction model is likely to be, and the reliability of the prediction result is likely to be reduced.
At step S150, the federation model is optimized using the operation data of the target building, to obtain an optimized federation model, which is used as a local federation model of the target building.
In some embodiments, in step S150, the optimizing the federal model using the operation data of the target building to obtain an optimized federal model, which is a local federal model of the target building, includes: aiming at the target building, the target building is enabled to perform local training by utilizing the federal model based on the running data of the target building, and training parameters of the local model of the target building are obtained. And optimizing the training parameters of the federal model by utilizing the training parameters of the local model built by the target, so as to obtain an optimized federal model which is used as the local federal model of the target building.
Specifically, as shown in fig. 5, the building migration network construction method based on federal learning further includes: step 24, fine tuning of the federal model. Federal model fine tuning refers to further fine tuning the federal model based on operational data of the target building to improve prediction accuracy.
In step 24, if the target building has operation data, the optimal federal model can be finely tuned by using the operation data to become a localized federal model, so as to further adapt to the operation condition of the target building and obtain higher prediction accuracy. Specifically, referring to the example shown in fig. 5, the weights and activation thresholds of the optimal federation model are first migrated to the local as initial parameters of the localization federation model. New gradients are then calculated using the local data and the parameters of these federal models are adjusted (i.e., federal model fine-tuning). After the fine tuning is finished, the localized federal model can be used for the energy consumption prediction task of the target building. The operation data of the target building may be history data of previous operation, such as some history data: parameters affecting building energy consumption, such as outdoor temperature, outdoor humidity, and cooling load.
According to the scheme, a migration network between similar buildings, namely the similar building migration network, is established based on a federal learning algorithm, all building main bodies in the similar building migration network jointly train a sharable and movable federal model, and through global training and local fine adjustment of the federal model, the common energy usage rules of multiple buildings and the individual energy characteristics of single buildings are learned, and the similar building migration network can realize safe and effective utilization of multiparty data, so that small sample scenes with limited part of building operation data are effectively utilized, and the problems that buildings with limited building operation data such as new buildings and old buildings with imperfect energy management systems are solved, and the prediction of building energy efficiency is difficult to use a data driving method are solved.
At step S160, a building energy efficiency prediction is performed for the target building using the local federal model of the target building.
In the scheme of the invention, in order to more finely process time sequence data in building operation data and extract more effective features from the time sequence data in the building operation data in the time dimension, the scheme of the invention provides a feature engineering model (such as a mixed model fusing RNN and BPNN) based on deep learning. In the space dimension, in order to effectively utilize building operation data of other buildings to cope with a small sample scene with limited building operation data of a target building, the scheme of the invention provides a migration learning strategy, such as building a similar building migration network based on federal learning (Federated Learning). Wherein, small sample scenes such as: some buildings have low data density per unit time due to sensor missing or failure, low data acquisition frequency, data storage errors and the like, so that the accumulated historical operation data is limited even if the building is operated for a long time. The federal learning is a distributed machine learning technology, and is implemented by performing distributed model training among a plurality of data sources with local data, and constructing a global model based on virtual fusion data only by exchanging model parameters or intermediate results on the premise of not exchanging local individual or sample data, so that balance between data privacy protection and data sharing calculation is realized, namely, a new application paradigm of 'data available invisible' and 'data motionless model'.
In this way, the scheme of the invention provides a building energy efficiency prediction method based on federal learning, in particular to a similar building migration network construction method based on federal learning, wherein similar building migration networks are constructed based on federal learning, each building main body in the similar building networks trains a local prediction model based on own building operation data and shares parameters of the local prediction model, and after the parameters of all the local prediction models are aggregated through a safe aggregation algorithm, the parameters of all the local prediction models are used for updating the shared and movable federal model, so that the problem that building energy efficiency prediction is difficult to be carried out by using a data driving method for new buildings, old buildings with imperfect energy management systems and other buildings with limited building operation data is solved. Meanwhile, in the scheme of the invention, as the federal learning technology can build a similar building migration network for all participants, multiparty data can be safely utilized in the similar building network, and a sharable and migratable data driving model (namely a federal model) can be built, so that the problem of data island under privacy protection can be solved.
When an enterprise develops to a certain stage, each department stores data, and the data between departments cannot be shared, so that the data are in a lack of relevance like individual islands. The isolated information system can not provide comprehensive information of cross departments and cross systems, various data can not form valuable information, local information can not be promoted to management knowledge, enterprise operation data can not be provided on the whole, and scientific decisions of an enterprise management layer can not be supported. The main reason for the problem of data island is that the data collected by each department has confidentiality, the shared data needs to be subjected to complicated procedures and high transmission cost, and the leakage of private data of the user and even the damage to a data system of the user can be caused, so that the data is difficult to use in a large-scale cross-region manner, and the problem of data island in the system is urgently solved. The federal learning defines a new distributed machine learning framework, under the framework, the problem that different data owners cooperate under the condition of not exchanging data is solved by designing a virtual model, a plurality of institutions can be effectively helped to perform data use and machine learning modeling under the condition that the requirements of user privacy protection, data security and government regulations are met, and the problem of data island under privacy protection can be solved.
By adopting the technical scheme of the embodiment, the similar building migration network is obtained by establishing the migration network between similar buildings based on the federal learning algorithm, each building main body in the similar building network trains the local prediction model based on own building operation data and shares the parameters of the local prediction model, and the federal model of global training is obtained after the parameters of all the local prediction models are aggregated through the secure aggregation algorithm. When a new building is added, the parameters of the local prediction model of the new building are only needed to be included in the process of parameter aggregation by the parameter noble safety aggregation algorithm of all the local prediction models. Building energy efficiency prediction of the building with limited building operation data is achieved by carrying out local fine adjustment on the federal model, so that a migration network between similar buildings is established based on a federal learning algorithm, a sharable and movable federal model is obtained based on common global training of all building main bodies in the migration network, and the building energy efficiency prediction of the building with limited building operation data is achieved by carrying out local fine adjustment on the federal model.
There is also provided, in accordance with an embodiment of the present invention, an apparatus for building energy efficiency prediction corresponding to the method for building energy efficiency prediction. Referring to fig. 3, a schematic diagram of an embodiment of the apparatus of the present invention is shown. The apparatus for building energy efficiency prediction may include: an acquisition unit 102 and a control unit 104.
Wherein, the obtaining unit 102 is configured to obtain the energy consumption parameter of the target building (such as obtaining the energy consumption mode of the target building), and obtain the operation data of the target building when the target building has the operation data. The specific function and process of the acquisition unit 102 refer to step S110.
The control unit 104 is configured to select a building similar to the energy usage parameter of the target building, and record the selected building as another building according to the energy usage parameter of the target building. The specific function and process of the control unit 104 refer to step S120.
In some embodiments, the control unit 104 selects a building similar to the energy usage parameter of the target building, denoted as another building, according to the energy usage parameter of the target building, including: the control unit 104 is specifically further configured to select, according to the energy consumption parameter of the target building, a building with a similarity degree with the energy consumption parameter of the target building within a preset building range, and record the building as another building.
The target building and the other buildings are in a set similarity range in at least one of building type, climate zone where the building is located and building structure characteristics. The building structure characteristics include: at least one of a window wall ratio, a maintenance structural material, a building area, and a building orientation.
FIG. 5 is a flow diagram of one embodiment of a method of building migration network construction based on federal learning. As shown in fig. 5, the building migration network construction method based on federal learning includes: step 21, similar building selection. Similar building selection is intended to select a building that is similar to the energy usage pattern of the target building (e.g., new building or existing building with limited data).
In step 21, when similar building selection is performed, when the characteristics of two buildings are different, there is a significant difference in their energy usage patterns. When building a building migration network, the operation data of similar buildings in energy consumption mode is beneficial to federal model training.
When a similar building is specifically selected, three main factors influencing the energy consumption of the building are mainly considered: building type, climate zone in which the building is located, and building structural characteristics. The first factor is the type of building, and the different types of building energy patterns differ significantly. The second factor is that the building in the climatic zone is high in energy consumption for heating in winter in cold climates, while the building in the hot climates does not need heating in winter. The third factor is building structural characteristics such as window wall ratio, maintenance structural materials, building area and building orientation, etc., which all affect building energy consumption. Two buildings are similar in type, in the same climate zone, and similar in structural characteristics.
The control unit 104 is further configured to construct a migration network, denoted as a similar building migration network, based on the target building and the other buildings. The specific function and processing of the control unit 104 is also referred to in step S130.
The control unit 104 is further configured to train all the buildings in the similar building migration network based on the pre-constructed basic energy efficiency prediction model by using a federal learning algorithm, so as to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and record the shared energy efficiency prediction model as a federal model. The specific function and process of the control unit 104 also refer to step S140.
In some embodiments, the pre-constructed basic energy efficiency prediction model of the control unit 104 is a time sequence prediction method based on RNN and BPNN, and the pre-constructed hybrid model of fusion RNN and BPNN is obtained.
Specifically, since the building energy system has characteristics of various kinds of equipment, complex topology, strong nonlinearity, large time lag, strong multi-system coupling, and the like, the building energy consumption is taken as typical time series data, and the change thereof is affected by various factors such as physical properties of the building, outdoor weather conditions, personnel use behaviors, equipment operation conditions, socioeconomic changes, geographical locations, and the like. How to comprehensively consider the internal and external influence factors, effectively utilize new technologies such as big data, artificial intelligence and the like, and construct an accurate and reliable energy consumption prediction model is an important research content in the building field. Therefore, in order to construct a new, non-fully connected form of the feature engineering model, the order of the time series is maintained throughout the feature extraction process. In the scheme of the invention, a time sequence prediction method based on a neural network such as an RNN (recurrent neural network) and a BPNN (BP neural network) is provided. The time sequence prediction method constructs a mixed model fused with RNN and BPNN so as to couple the feature extraction and the energy consumption prediction into the same model. RNNs have unique timing structures for extracting features from the timing data of building operational data. The BPNN has strong nonlinear fitting capability and is used for establishing a mapping relation between the extracted features and the energy consumption.
In some embodiments, the control unit 104 trains all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model by using a federal learning algorithm to obtain a shared energy efficiency prediction model that can be shared by all buildings in the similar building migration network, which is denoted as a federal model, and includes:
the control unit 104 is specifically further configured to build a federal learning framework based on a pre-built basic energy efficiency prediction model using a federal learning algorithm. The specific function and process of the control unit 104 also refer to step S210.
The control unit 104 is specifically further configured to, for each building in the similar building migration network, make the building perform local training by using the federal learning framework based on the operation data of the building itself, so as to obtain training parameters of a local model of the building itself. The specific function and process of the control unit 104 is also referred to as step S220.
The control unit 104 is specifically further configured to perform a secure aggregation process on training parameters of local models of all buildings in the similar building migration network, so as to obtain a shared energy efficiency prediction model that all buildings in the similar building migration network can share, and record the shared energy efficiency prediction model as a federal model. The specific function and process of the control unit 104 is also referred to as step S230.
The control unit 104 is specifically further configured to, in a case where a new building joins the similar building migration network, make the new building perform local training by using the federal learning framework based on the operation data of the building itself for the new building, so as to obtain training parameters of the local model of the new building itself. And carrying out safe aggregation treatment on the training parameters of the local model of the new building and the training parameters of the local model of all the buildings in the similar building migration network together to obtain a shared energy efficiency prediction model which can be shared by the new building and all the buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
In the scheme of the invention, in order to couple the feature extraction and the energy consumption prediction of the mixed model fusing the RNN and the BPNN, the feature extraction and the energy consumption prediction are performed in a supervision mode, so that the two are simultaneously optimized. According to the scheme, a similar building migration network is built based on federal learning (Federated Learning), each building main body in the similar building migration network trains a local prediction model based on own building operation data and shares parameters of the local prediction model, and after the parameters of all the local prediction models are aggregated through a secure aggregation algorithm, the parameters of all the local prediction models are used for updating the shared and movable federal model.
When a new building is added, only parameters of a local prediction model of the new building are needed to be included in the parameter aggregation process, all the buildings are not a unidirectional and fixed migration chain, but a multidirectional, flexible and extensible similar building migration network is formed, and the problem that the building energy efficiency prediction is difficult to carry out by using a data driving method for buildings with limited building operation data such as old buildings with new buildings and imperfect energy management systems is solved.
Preferably, the control unit 104, for each building in the similar building migration network, makes the building perform local training by using the federal learning framework based on the operation data of the building itself, to obtain training parameters of a local model of the building itself, including: the control unit 104 is specifically further configured to, for each building in the similar building migration network, locally train the building on the local side of the building based on the running data of the building itself by using the basic energy efficiency prediction model in the federal learning framework, and obtain a training gradient of the local model of the building itself, which is recorded as gradient information. And encrypting the gradient information to obtain an encryption result, and taking the encryption result as a training parameter of the local model of the building.
Accordingly, the control unit 104 performs a secure aggregation process on training parameters of local models of all buildings in the similar building migration network to obtain a shared energy efficiency prediction model that all buildings in the similar building migration network can share, which is denoted as a federal model, and includes: the control unit 104 is specifically further configured to perform, on the server side, secure aggregation processing on the training parameters of the local models of all buildings in the similar building migration network, to obtain aggregation gradient information. And updating training parameters of the basic energy efficiency prediction model in the federal learning framework based on the aggregation gradient information to obtain a shared energy efficiency prediction model which can be shared by all buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
In particular, in the framework of federal learning, data from different data owners can be safely used, and privacy disclosure can be effectively prevented. FIG. 4 is a flow chart of a federal model training process in which the federal learning framework includes K participants (p 1 ,p 2 ,Kp k ) The goal of the participants is to utilize the respective data (D) securely by some means 1 ,D 2 K,D K ) To train better machine learning model to complete the task of prediction or classification, K is a positive integer. The federal learning framework is essentially a model training paradigm in which all participants co-train a federal model with the aid of the cloud or a third party (the federal model performs better than a model trained using the participants' own data) without exposing their original data to other participants or third parties. Here, all participants are virtually independent of each other, each having a target building that is not identical.
In the example shown in fig. 4, (1), (2), (3) and (4) respectively represent: (1) representing calculating training gradient of the local model, (2) representing safe aggregation and updating the federal model, (3) representing issuing the federal model, and (4) representing updating the local model.
In the example shown in fig. 4, there are four levels: a participant layer, a local model layer, a communication layer, and a cloud layer. At the participant level, each participant has data stored at its local server and does not send any data containing sensitive information to the cloud or other participant's server. At the local model layer, each participant stores a respective local model at its local server. And at the communication layer, the participants and the cloud end transmit information. At the cloud level, the cloud server stores the federal model. The interaction between the four hierarchies includes the following four steps:
And 11, each participant calculates training gradient of the local model based on own data, encrypts gradient information through homomorphic encryption, differential privacy or secret sharing and other technologies, and uploads the encryption result of the participant to the cloud.
And step 12, the cloud server safely aggregates the encryption results of all participants, and updates the parameters of the federation model based on the aggregation gradient to obtain new parameters of the federation model.
And 13, the cloud transmits new parameters of the federal model to each participant. Wherein the new parameters of the federal model issued by the cloud to different participants are the same.
Step 14, each participant receives new parameters of the federal model issued by the cloud and updates the parameters of the local model of the participant.
Thus, the training process of the federal model continuously repeats the four steps until the parameters of the federal model converge. The training process of the federal model is applicable to various machine learning models, such as LR, BPNN (BP neural network), DNN (deep neural network) and the like, and all participants share the final federal model.
Wherein, the control unit 104 selects, for each building in the similar building migration network, an energy consumption characteristic parameter of the building as operation data of the building itself on a local side of the building. The energy consumption characteristic parameters of the building comprise: at least one of time characteristics, weather information, historical energy consumption, and building structure information. The energy consumption characteristic parameter of the building is a characteristic parameter related to the energy consumption of the building among the characteristic parameters of the building.
Specifically, as shown in fig. 5, the building migration network construction method based on federal learning further includes:
step 22, feature selection. Feature selection, aimed at selecting the appropriate feature as the input of the predictive model.
In step 22, the feature selection is performed by selecting the feature related to the building energy consumption as the input of the prediction model. Features related to building energy consumption, such as time features, weather information, historical energy consumption, building structure information and the like, are features commonly used in energy consumption prediction models. The time profile reflects the impact of the number of people and activities inside the building on energy consumption. Weather information is a major driver of the cold/heat load of a building.
Step 23, training and optimizing the federal model. Federal model training, which aims to train a mobilizable federal model using data of similar buildings securely.
In step 23, based on the constructed similar building migration network, all buildings cooperate to train a federal model, and the training process comprises the following three steps:
step 231, selecting an appropriate data-driven algorithm (i.e., data-driven model) as the base prediction algorithm (i.e., base prediction model).
And 232, training the basic prediction model by adopting a safe aggregation algorithm to obtain a federal model (namely a building energy efficiency prediction model).
Step 233, optimizing super parameters of the federal model.
Fig. 6 is a schematic diagram of a training process of a building energy efficiency system model (i.e., a building energy consumption prediction model). As shown in fig. 6, the process of training the basic prediction model to obtain the federal model (i.e., the building energy efficiency prediction model) in step 232 by using a security aggregation algorithm includes:
in the training process, data processing and normalization are performed first.
Step 32, determining the number of nodes, network weights, basis function widths and central parameters required by the neural network.
And 33, judging whether the error meets the requirement according to the prediction model of the building energy efficiency system.
And step 34, outputting the final model as a federal model (namely, a building energy efficiency prediction model) if the requirements are met.
In the scheme of the invention, in order to construct an extensible and flexible similar building migration network, any building in the similar building migration network can share a migration model (namely a shared and movable federal model), so that the migration effect is ensured, and repeated training is avoided. According to the scheme, the energy consumption prediction model (namely the shared and movable federal model) is adopted for carrying out federal learning-based building energy efficiency prediction, and the sensitivity index input by the energy consumption prediction model is calculated firstly based on a sensitivity analysis method so as to measure the influence of the sensitivity index input by the energy consumption prediction model on the energy consumption prediction value. And then calculating a weighted Manhattan distance based on the sensitivity index to quantify the difference between different working conditions and obtain a prediction error. And finally, fitting a relation between the prediction error and the weighted Manhattan distance to infer a change rule of the prediction error of the energy consumption prediction model and evaluate the reliability of the energy consumption prediction model.
The control unit 104 is further configured to optimize the federal model by using the operation data of the target building, to obtain an optimized federal model, which is used as a local federal model of the target building. The specific function and process of the control unit 104 also refer to step S150.
In some embodiments, the control unit 104 optimizes the federal model using the operational data of the target building to obtain an optimized federal model as a local federal model of the target building, including: the control unit 104 is specifically further configured to make the target building perform local training by using the federal model based on the operation data of the target building to obtain training parameters of the local model of the target building. And optimizing the training parameters of the federal model by utilizing the training parameters of the local model built by the target, so as to obtain an optimized federal model which is used as the local federal model of the target building.
Specifically, as shown in fig. 5, the building migration network construction method based on federal learning further includes: step 24, fine tuning of the federal model. Federal model fine tuning refers to further fine tuning the federal model based on operational data of the target building to improve prediction accuracy.
In step 24, if the target building has operation data, the optimal federal model can be finely tuned by using the operation data to become a localized federal model, so as to further adapt to the operation condition of the target building and obtain higher prediction accuracy. Specifically, referring to the example shown in fig. 5, the weights and activation thresholds of the optimal federation model are first migrated to the local as initial parameters of the localization federation model. New gradients are then calculated using the local data and the parameters of these federal models are adjusted (i.e., federal model fine-tuning). After the fine tuning is finished, the localized federal model can be used for the energy consumption prediction task of the target building.
According to the scheme, a migration network between similar buildings, namely the similar building migration network, is established based on a federal learning algorithm, all building main bodies in the similar building migration network jointly train a sharable and movable federal model, and through global training and local fine adjustment of the federal model, the common energy usage rules of multiple buildings and the individual energy characteristics of single buildings are learned, and the similar building migration network can realize safe and effective utilization of multiparty data, so that small sample scenes with limited part of building operation data are effectively utilized, and the problems that buildings with limited building operation data such as new buildings and old buildings with imperfect energy management systems are solved, and the prediction of building energy efficiency is difficult to use a data driving method are solved.
The control unit 104 is further configured to predict a building energy efficiency of the target building using a local federal model of the target building. The specific function and process of the control unit 104 is also referred to as step S160.
In the scheme of the invention, in order to more finely process time sequence data in building operation data and extract more effective features from the time sequence data in the building operation data in the time dimension, the scheme of the invention provides a feature engineering model (such as a mixed model fusing RNN and BPNN) based on deep learning. In the space dimension, in order to effectively utilize building operation data of other buildings to cope with a small sample scene with limited building operation data of a target building, the scheme of the invention provides a migration learning strategy, such as building a similar building migration network based on federal learning (Federated Learning). Wherein, small sample scenes such as: some buildings have low data density per unit time due to sensor missing or failure, low data acquisition frequency, data storage errors and the like, so that the accumulated historical operation data is limited even if the building is operated for a long time. The federal learning is a distributed machine learning technology, and is implemented by performing distributed model training among a plurality of data sources with local data, and constructing a global model based on virtual fusion data only by exchanging model parameters or intermediate results on the premise of not exchanging local individual or sample data, so that balance between data privacy protection and data sharing calculation is realized, namely, a new application paradigm of 'data available invisible' and 'data motionless model'.
In this way, the scheme of the invention provides a building energy efficiency prediction method based on federal learning, in particular to a similar building migration network construction method based on federal learning, wherein similar building migration networks are constructed based on federal learning, each building main body in the similar building networks trains a local prediction model based on own building operation data and shares parameters of the local prediction model, and after the parameters of all the local prediction models are aggregated through a safe aggregation algorithm, the parameters of all the local prediction models are used for updating the shared and movable federal model, so that the problem that building energy efficiency prediction is difficult to be carried out by using a data driving method for new buildings, old buildings with imperfect energy management systems and other buildings with limited building operation data is solved. Meanwhile, in the scheme of the invention, as the federal learning technology can build a similar building migration network for all participants, multiparty data can be safely utilized in the similar building network, and a sharable and migratable data driving model (namely a federal model) can be built, so that the problem of data island under privacy protection can be solved.
When an enterprise develops to a certain stage, each department stores data, and the data between departments cannot be shared, so that the data are in a lack of relevance like individual islands. The isolated information system can not provide comprehensive information of cross departments and cross systems, various data can not form valuable information, local information can not be promoted to management knowledge, enterprise operation data can not be provided on the whole, and scientific decisions of an enterprise management layer can not be supported. The main reason for the problem of data island is that the data collected by each department has confidentiality, the shared data needs to be subjected to complicated procedures and high transmission cost, and the leakage of private data of the user and even the damage to a data system of the user can be caused, so that the data is difficult to use in a large-scale cross-region manner, and the problem of data island in the system is urgently solved. The federal learning defines a new distributed machine learning framework, under the framework, the problem that different data owners cooperate under the condition of not exchanging data is solved by designing a virtual model, a plurality of institutions can be effectively helped to perform data use and machine learning modeling under the condition that the requirements of user privacy protection, data security and government regulations are met, and the problem of data island under privacy protection can be solved.
Since the processes and functions implemented by the apparatus of the present embodiment substantially correspond to the embodiments, principles and examples of the foregoing methods, the descriptions of the embodiments are not exhaustive, and reference may be made to the descriptions of the foregoing embodiments and their descriptions are omitted herein.
By adopting the technical scheme, a similar building migration network is obtained by establishing a migration network between similar buildings based on a federal learning algorithm, each building main body in the similar building network trains a local prediction model based on own building operation data and shares the parameters of the local prediction model, and the parameters of all the local prediction models are aggregated by a secure aggregation algorithm to obtain a globally trained federal model; when a new building is added, the parameters of the local prediction model of the new building are only needed to be included in the process of parameter aggregation by the parameter noble safety aggregation algorithm of all the local prediction models; and for the buildings with limited building operation data, the federal model is subjected to local fine adjustment, so that the building operation data of other buildings are effectively utilized to cope with a small sample scene with limited building operation data of a target building, and the building energy efficiency prediction of the buildings with limited building operation data is realized.
There is also provided, in accordance with an embodiment of the present invention, a terminal corresponding to an apparatus for building energy efficiency prediction. The terminal may include: the device for predicting the energy efficiency of the building.
Since the processes and functions implemented by the terminal of the present embodiment basically correspond to the embodiments, principles and examples of the foregoing apparatus, the description of the present embodiment is not exhaustive, and reference may be made to the related descriptions of the foregoing embodiments, which are not repeated herein.
By adopting the technical scheme, a similar building migration network is obtained by establishing a migration network between similar buildings based on a federal learning algorithm, each building main body in the similar building network trains a local prediction model based on own building operation data and shares the parameters of the local prediction model, and the parameters of all the local prediction models are aggregated by a secure aggregation algorithm to obtain a globally trained federal model; when a new building is added, the parameters of the local prediction model of the new building are only needed to be included in the process of parameter aggregation by the parameter noble safety aggregation algorithm of all the local prediction models; the building energy efficiency prediction of the building with limited building operation data is realized by carrying out local fine adjustment on the federal model, and a similar building migration network which is unidirectional, fixed, multidirectional, flexible and extensible is not formed between all the buildings.
According to an embodiment of the present invention, there is also provided a storage medium corresponding to a method of building energy efficiency prediction, the storage medium including a stored program, wherein the device in which the storage medium is controlled to perform the above-described method of building energy efficiency prediction when the program is run.
Since the processes and functions implemented by the storage medium of the present embodiment substantially correspond to the embodiments, principles and examples of the foregoing methods, the descriptions of the present embodiment are not exhaustive, and reference may be made to the related descriptions of the foregoing embodiments, which are not repeated herein.
By adopting the technical scheme, a similar building migration network is obtained by establishing a migration network between similar buildings based on a federal learning algorithm, each building main body in the similar building network trains a local prediction model based on own building operation data and shares the parameters of the local prediction model, and the parameters of all the local prediction models are aggregated by a secure aggregation algorithm to obtain a globally trained federal model; when a new building is added, the parameters of the local prediction model of the new building are only needed to be included in the process of parameter aggregation by the parameter noble safety aggregation algorithm of all the local prediction models; the building energy efficiency prediction of the building with limited building operation data is realized by carrying out local fine adjustment on the federal model, multiparty data can be safely utilized in a similar building network, and the problem of data island under privacy protection can be solved.
In summary, it is readily understood by those skilled in the art that the above-described advantageous ways can be freely combined and superimposed without conflict.
The above description is only an example of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.
Claims (16)
1. A method of building energy efficiency prediction, comprising:
acquiring energy consumption parameters of a target building, and acquiring operation data of the target building;
according to the energy consumption parameters of the target building, selecting a building similar to the energy consumption parameters of the target building, and marking the building as other buildings;
constructing a migration network based on the target building and the other buildings, and recording the migration network as a similar building migration network;
training all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model by utilizing a federal learning algorithm to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model;
Optimizing the federation model by utilizing the operation data of the target building to obtain an optimized federation model, wherein the optimized federation model is used as a local federation model of the target building;
and predicting the building energy efficiency of the target building by using the local federal model of the target building.
2. The method of building energy efficiency prediction according to claim 1, wherein selecting a building similar to the energy usage parameter of the target building, denoted as another building, based on the energy usage parameter of the target building, comprises:
according to the energy consumption parameters of the target building, selecting a building with the similarity degree of the energy consumption parameters of the target building within a preset building range, and marking the building as other buildings;
the target building and the other buildings have the similarity in at least one of building type, climate zone where the building is located and building structure characteristics within a set similarity range; the building structure characteristics include: at least one of a window wall ratio, a maintenance structural material, a building area, and a building orientation.
3. The method for predicting building energy efficiency according to claim 1 or 2, wherein the pre-constructed basic energy efficiency prediction model is a time sequence prediction method based on RNN and BPNN, and the pre-constructed hybrid model is a hybrid model of fusion RNN and BPNN.
4. A method of energy efficiency prediction for buildings according to any of claims 1 to 3, wherein training all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model using a federal learning algorithm to obtain a shared energy efficiency prediction model that all buildings in the similar building migration network can share, denoted as a federal model, comprises:
constructing a federal learning framework based on a pre-constructed basic energy efficiency prediction model by utilizing a federal learning algorithm;
aiming at each building in the similar building migration network, enabling the building to perform local training by utilizing the federal learning framework based on the running data of the building to obtain training parameters of a local model of the building;
aiming at training parameters of local models of all buildings in the similar building migration network, carrying out safe aggregation treatment to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model;
under the condition that a new building is added into the similar building migration network, aiming at the new building, enabling the new building to perform local training by utilizing the federal learning framework based on the running data of the building to obtain training parameters of a local model of the new building; and carrying out safe aggregation treatment on the training parameters of the local model of the new building and the training parameters of the local model of all the buildings in the similar building migration network together to obtain a shared energy efficiency prediction model which can be shared by the new building and all the buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
5. The method of building energy efficiency prediction according to claim 4, wherein,
aiming at each building in the similar building migration network, the building is locally trained by utilizing the federal learning framework based on the running data of the building, so as to obtain training parameters of a local model of the building, comprising the following steps:
aiming at each building in the similar building migration network, on the local side of the building, enabling the building to perform local training based on the running data of the building, and using the basic energy efficiency prediction model in the federal learning frame to obtain training gradient of the local model of the building, and marking the training gradient as gradient information; encrypting the gradient information to obtain an encryption result, and taking the encryption result as a training parameter of a local model of the building;
correspondingly, aiming at training parameters of local models of all buildings in the similar building migration network, carrying out security aggregation treatment to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model, wherein the method comprises the following steps of:
aiming at the training parameters of the local models of all the buildings in the similar building migration network, carrying out safe aggregation treatment on the training parameters of the local models of all the buildings at a server side to obtain aggregation gradient information; and updating training parameters of the basic energy efficiency prediction model in the federal learning framework based on the aggregation gradient information to obtain a shared energy efficiency prediction model which can be shared by all buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
6. The method of building energy efficiency prediction according to claim 5, wherein for each building in the similar building migration network, at the local side of the building, the energy consumption characteristic parameter of the building is selected as the operational data of the building itself; the energy consumption characteristic parameters of the building comprise: at least one of time characteristics, weather information, historical energy consumption, and building structure information.
7. The method of claim 1 to 6, wherein optimizing the federal model using the operational data of the target building to obtain an optimized federal model is a local federal model of the target building, comprising:
aiming at the target building, enabling the target building to perform local training by utilizing the federal model based on the running data of the target building to obtain training parameters of the local model of the target building; and optimizing the training parameters of the federal model by utilizing the training parameters of the local model built by the target, so as to obtain an optimized federal model which is used as the local federal model of the target building.
8. An apparatus for building energy efficiency prediction, comprising:
The system comprises an acquisition unit, a control unit and a control unit, wherein the acquisition unit is configured to acquire energy consumption parameters of a target building and acquire operation data of the target building;
a control unit configured to select a building similar to the energy usage parameter of the target building, as the other building, according to the energy usage parameter of the target building;
the control unit is further configured to construct a migration network based on the target building and the other buildings, and the migration network is recorded as a similar building migration network;
the control unit is further configured to train all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model by utilizing a federal learning algorithm to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and record the shared energy efficiency prediction model as a federal model;
the control unit is further configured to optimize the federation model by using the operation data of the target building to obtain an optimized federation model, and the optimized federation model is used as a local federation model of the target building;
the control unit is further configured to predict a building energy efficiency for the target building using a local federal model of the target building.
9. The apparatus for predicting energy efficiency of a building according to claim 8, wherein the control unit selects a building similar to the energy consumption parameter of the target building as the other building according to the energy consumption parameter of the target building, comprising:
According to the energy consumption parameters of the target building, selecting a building with the similarity degree of the energy consumption parameters of the target building within a preset building range, and marking the building as other buildings;
the target building and the other buildings have the similarity in at least one of building type, climate zone where the building is located and building structure characteristics within a set similarity range; the building structure characteristics include: at least one of a window wall ratio, a maintenance structural material, a building area, and a building orientation.
10. The apparatus according to claim 8 or 9, wherein the control unit pre-constructs a basic energy efficiency prediction model, which is a time-series prediction apparatus based on RNN and BPNN, and pre-constructs a hybrid model that fuses RNN and BPNN.
11. The apparatus according to any one of claims 8 to 10, wherein the control unit trains all buildings in the similar building migration network based on a pre-constructed basic energy efficiency prediction model using a federal learning algorithm to obtain a shared energy efficiency prediction model that all buildings in the similar building migration network can share, denoted as a federal model, comprising:
Constructing a federal learning framework based on a pre-constructed basic energy efficiency prediction model by utilizing a federal learning algorithm;
aiming at each building in the similar building migration network, enabling the building to perform local training by utilizing the federal learning framework based on the running data of the building to obtain training parameters of a local model of the building;
aiming at training parameters of local models of all buildings in the similar building migration network, carrying out safe aggregation treatment to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and marking the shared energy efficiency prediction model as a federal model;
under the condition that a new building is added into the similar building migration network, aiming at the new building, enabling the new building to perform local training by utilizing the federal learning framework based on the running data of the building to obtain training parameters of a local model of the new building; and carrying out safe aggregation treatment on the training parameters of the local model of the new building and the training parameters of the local model of all the buildings in the similar building migration network together to obtain a shared energy efficiency prediction model which can be shared by the new building and all the buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
12. The apparatus for building energy efficiency prediction according to claim 11, wherein,
the control unit, for each building in the similar building migration network, makes the building perform local training by using the federal learning framework based on the running data of the building, and obtains training parameters of a local model of the building, including:
aiming at each building in the similar building migration network, on the local side of the building, enabling the building to perform local training based on the running data of the building, and using the basic energy efficiency prediction model in the federal learning frame to obtain training gradient of the local model of the building, and marking the training gradient as gradient information; encrypting the gradient information to obtain an encryption result, and taking the encryption result as a training parameter of a local model of the building;
correspondingly, the control unit performs a security aggregation process on training parameters of local models of all buildings in the similar building migration network to obtain a shared energy efficiency prediction model which can be shared by all the buildings in the similar building migration network, and the shared energy efficiency prediction model is recorded as a federal model, and comprises the following steps:
Aiming at the training parameters of the local models of all the buildings in the similar building migration network, carrying out safe aggregation treatment on the training parameters of the local models of all the buildings at a server side to obtain aggregation gradient information; and updating training parameters of the basic energy efficiency prediction model in the federal learning framework based on the aggregation gradient information to obtain a shared energy efficiency prediction model which can be shared by all buildings in the similar building migration network, and recording the shared energy efficiency prediction model as a federal model.
13. The apparatus for building energy efficiency prediction according to claim 12, wherein the control unit selects, for each building in the similar building migration network, at a local side of the building, an energy consumption characteristic parameter of the building as the operation data of the building itself; the energy consumption characteristic parameters of the building comprise: at least one of time characteristics, weather information, historical energy consumption, and building structure information.
14. The apparatus for predicting building energy efficiency according to any one of claims 8 to 13, wherein the control unit optimizes the federal model using the operation data of the target building to obtain an optimized federal model as a local federal model of the target building, comprising:
Aiming at the target building, enabling the target building to perform local training by utilizing the federal model based on the running data of the target building to obtain training parameters of the local model of the target building; and optimizing the training parameters of the federal model by utilizing the training parameters of the local model built by the target, so as to obtain an optimized federal model which is used as the local federal model of the target building.
15. A terminal, comprising: an apparatus for building energy efficiency prediction as claimed in any one of claims 8 to 14.
16. A storage medium comprising a stored program, wherein the program, when run, controls a device in which the storage medium is located to perform the method of building energy efficiency prediction of any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211586076.8A CN116187514A (en) | 2022-12-07 | 2022-12-07 | Method, device, terminal and storage medium for predicting building energy efficiency |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211586076.8A CN116187514A (en) | 2022-12-07 | 2022-12-07 | Method, device, terminal and storage medium for predicting building energy efficiency |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116187514A true CN116187514A (en) | 2023-05-30 |
Family
ID=86437354
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211586076.8A Pending CN116187514A (en) | 2022-12-07 | 2022-12-07 | Method, device, terminal and storage medium for predicting building energy efficiency |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116187514A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117808627A (en) * | 2023-12-29 | 2024-04-02 | 光谷技术有限公司 | Energy consumption supervision method and related device |
-
2022
- 2022-12-07 CN CN202211586076.8A patent/CN116187514A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117808627A (en) * | 2023-12-29 | 2024-04-02 | 光谷技术有限公司 | Energy consumption supervision method and related device |
CN117808627B (en) * | 2023-12-29 | 2024-08-09 | 光谷技术有限公司 | Energy consumption supervision method and related device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | A BP neural network prediction model based on dynamic cuckoo search optimization algorithm for industrial equipment fault prediction | |
Chung et al. | Distributed deep reinforcement learning for intelligent load scheduling in residential smart grids | |
Li et al. | Federated learning-based short-term building energy consumption prediction method for solving the data silos problem | |
Wang et al. | Ensemble learning for load forecasting | |
Chen et al. | Modeling and optimization of complex building energy systems with deep neural networks | |
CN114777192B (en) | Secondary network heat supply autonomous optimization regulation and control method based on data association and deep learning | |
Tran et al. | Using Fuzzy Clustering Chaotic-based Differential Evolution to solve multiple resources leveling in the multiple projects scheduling problem | |
CN116187514A (en) | Method, device, terminal and storage medium for predicting building energy efficiency | |
CN115854501B (en) | Airport terminal room temperature large hysteresis prediction control method based on passenger flow prediction | |
CN114265913A (en) | Space-time prediction algorithm based on federal learning on industrial Internet of things edge equipment | |
CN116205359A (en) | Method, device, terminal and storage medium for predicting daily power load | |
CN116306911A (en) | Distributed machine learning-based thermodynamic station load prediction and optimization control method | |
Zhang et al. | Grid-interactive multi-zone building control using reinforcement learning with global-local policy search | |
Liao et al. | MEMS: An automated multi-energy management system for smart residences using the DD-LSTM approach | |
Zhang et al. | Improving the accuracy of load forecasting for campus buildings based on federated learning | |
CN112183918B (en) | Intelligent generation method of power transmission line online inspection operation plan | |
CN116796633A (en) | Power grid monitoring model correction method based on digital twin technology | |
CN116717836A (en) | Intelligent heat supply operation system and method based on Internet of things | |
Zhou et al. | Green building project cost budgeting and cost control integrating interactive VR genetic algorithm | |
Lu et al. | Residential Energy Consumption Forecasting Based on Federated Reinforcement Learning with Data Privacy Protection. | |
CN115908056A (en) | Heat supply system customer service intelligent processing method and system based on metauniverse | |
Li et al. | Office building energy consumption prediction based on LSTM and SVR | |
Song | [Retracted] Building Construction Design Based on Particle Swarm Optimization Algorithm | |
CN114781606A (en) | Method for predicting medium and long term electricity consumption in region based on federal learning | |
Khan et al. | Occupancy Prediction in Multi-Occupant IoT Environments leveraging Federated Learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |