CN112232479B - Building energy time-consuming space factor characterization method based on deep cascade neural network and related products - Google Patents

Building energy time-consuming space factor characterization method based on deep cascade neural network and related products Download PDF

Info

Publication number
CN112232479B
CN112232479B CN202010956592.XA CN202010956592A CN112232479B CN 112232479 B CN112232479 B CN 112232479B CN 202010956592 A CN202010956592 A CN 202010956592A CN 112232479 B CN112232479 B CN 112232479B
Authority
CN
China
Prior art keywords
spatial
features
time sequence
energy consumption
time
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.)
Active
Application number
CN202010956592.XA
Other languages
Chinese (zh)
Other versions
CN112232479A (en
Inventor
胡书山
王广超
余日季
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hubei University
Original Assignee
Hubei University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hubei University filed Critical Hubei University
Priority to CN202010956592.XA priority Critical patent/CN112232479B/en
Publication of CN112232479A publication Critical patent/CN112232479A/en
Application granted granted Critical
Publication of CN112232479B publication Critical patent/CN112232479B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/06—Power analysis or power optimisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Computer Hardware Design (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Business, Economics & Management (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Architecture (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Structural Engineering (AREA)
  • Development Economics (AREA)
  • Civil Engineering (AREA)
  • Game Theory and Decision Science (AREA)
  • Computational Mathematics (AREA)
  • Medical Informatics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)

Abstract

The embodiment of the application discloses a building energy time-consuming space factor characterization method based on a deep cascade neural network and a related product, wherein the method comprises the following steps: inputting spatial factors related to building energy consumption into a symmetrical residual error network, and extracting spatial characteristics related to the building energy consumption through the symmetrical residual error network; carrying out local feature weight quantization on the spatial features through a spatial attention mechanism to obtain quantized spatial features; inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a two-way long-short-term memory network to obtain the time sequence characteristics related to the building energy consumption; and carrying out local feature weight quantization on the time sequence features through a time sequence attention mechanism to obtain space-time factor joint features, wherein the space-time factor joint features are used for predicting energy consumption data of the building.

Description

Building energy time-consuming space factor characterization method based on deep cascade neural network and related products
Technical Field
The application relates to the technical field of computers, in particular to a building energy time-consuming space factor characterization method based on a deep cascade neural network and a related product.
Background
Building energy consumption (i.e. operation energy consumption of a building) is the sum of the energy consumption of various devices (such as heating, ventilation, air conditioning, lighting and office equipment) in the operation process of the building. Building energy consumption is the first major of today's energy consumption, for example, about 35% of the total energy consumption in europe and the united states, about 30% of the total energy consumption in china, but most building hvac systems are high in energy consumption and carbon emissions. The total energy consumption of China 2015 building is up to 9.64 hundred million tons of standard coal, and the international energy agency predicts that the total energy consumption of China building in 2030 will reach 15.2 hundred million tons of standard coal.
The development of computer technology brings a new opportunity for building energy consumption research, and the academic research of intersecting the two becomes a research hot spot in recent years, and is focused by a plurality of researchers at home and abroad, and research subjects comprise building modeling, energy consumption optimization and the like. The building energy consumption prediction takes building body shape coefficients, climate states and the like as input, and a computer is utilized to calculate building thermal states and energy consumption time by time dynamically and accurately. As a multi-factor related complex nonlinear problem, the solution of the problem needs to combine mathematical modeling, engineering background knowledge and computer programming, and relies on the powerful calculation of the computer to output results.
The accurate prediction result of building energy consumption plays a vital role in optimizing building energy efficiency. Aiming at a single building, the energy consumption prediction can help a designer optimize the design of a building model and the use of building materials, and the energy efficiency of the design model is improved; the method can also assist the manager in analyzing equipment optimization and building reconstruction strategies, so that a more reasonable energy efficiency optimization strategy is selected. The method is oriented to regional multi-building, the energy consumption prediction can provide data support for electric energy scheduling decisions in the intelligent power grid, and is assisted with energy consumption storage equipment, so that the phenomenon of power consumption peaks is weakened, and the deployment and management cost of regional power supply equipment is reduced.
Disclosure of Invention
The embodiment of the application provides a building energy time-consuming space factor characterization method based on a deep cascade neural network and a related product, which can be used for extracting features.
A building energy time-consuming space factor characterization method based on a deep cascade neural network comprises the following steps:
Inputting spatial factors related to building energy consumption into a symmetrical residual error network, and extracting spatial characteristics related to the building energy consumption through the symmetrical residual error network;
Carrying out local feature weight quantization on the spatial features through a spatial attention mechanism to obtain quantized spatial features;
inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a two-way long-short-term memory network to obtain the time sequence characteristics related to the building energy consumption;
And carrying out local feature weight quantization on the time sequence features through a time sequence attention mechanism to obtain time-space factor joint features, wherein the time-space factor joint features are used for predicting energy consumption data of the building.
Further, the symmetric residual network comprises a convolution module and a deconvolution module, wherein the convolution module comprises K residual blocks, and the deconvolution module comprises K deconvolution blocks, wherein K is an integer greater than or equal to 1;
The step of inputting the spatial factors related to the building energy consumption into a symmetrical residual error network, and extracting the spatial characteristics related to the building energy consumption through the symmetrical residual error network comprises the following steps:
Performing convolution operation on the space factor through K residual blocks in the convolution module to obtain an intermediate feature;
And deconvoluting the intermediate features through K deconvolution blocks in the deconvolution module to obtain the spatial features related to the building energy consumption.
Further, the spatial attention mechanism comprises a channel attention module and a spatial attention module; the step of carrying out local feature weight quantization on the spatial features through a spatial attention mechanism to obtain quantized spatial features comprises the following steps:
inputting the spatial features to the channel attention module to obtain channel attention weights, and obtaining first features according to the spatial features and the channel attention weights;
And inputting the first feature into the spatial attention module to obtain spatial attention weight, and obtaining the quantized spatial feature according to the first feature and the spatial attention weight.
Further, the channel attention module comprises a first average pooling layer, a first maximum pooling layer and a fully connected layer; the step of inputting the spatial features to the channel attention module to obtain channel attention weights comprises the following steps:
respectively inputting the spatial features into a first average pooling layer and a first maximum pooling layer to obtain spatial features after average pooling and spatial features after maximum pooling;
Adding the spatial features after the average pooling and the spatial features after the maximum pooling to obtain the attention weight through the full connection layer;
the spatial attention module comprises a second average pooling layer, a second maximum pooling layer and a convolution layer; the step of inputting the first feature to the spatial attention module to obtain a spatial attention weight includes:
Inputting the first features into the second average pooling layer and the second maximum pooling layer respectively to obtain first features after average pooling and first features after maximum pooling;
and carrying out convolution operation on the average pooled first feature and the maximum pooled first feature through the convolution layer to obtain the space attention weight.
Further, the two-way long-short-term memory network includes a forward propagation network and a backward propagation network;
inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a two-way long-short-term memory network to obtain the time sequence characteristics related to the building energy consumption, wherein the time sequence factors comprise:
inputting the time sequence factors related to the quantized spatial features and the building energy consumption into the forward propagation network to obtain first time sequence features related to the building energy consumption;
Inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into the backward propagation network to obtain second time sequence characteristics related to the building energy consumption;
And constructing and obtaining the time sequence characteristics related to the building energy consumption according to the first time sequence characteristics and the second time sequence characteristics.
Further, the time sequence factors comprise time factors corresponding to N times, the time sequence features comprise time features corresponding to N times, and N is an integer greater than or equal to 1;
inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a two-way long-short-term memory network to obtain the time sequence characteristics related to the building energy consumption, wherein the time sequence factors comprise:
inputting the quantized spatial features and N time factors into the forward propagation network to obtain N first time features;
inputting the quantized spatial features and N time factors into the backward propagation network to obtain N second time features;
And respectively combining the first time feature and the second time feature corresponding to the same time to obtain N time features.
Further, the performing local feature weight quantization on the time sequence feature through a time sequence attention mechanism to obtain a space-time factor joint feature includes:
inputting N time characteristics into the time sequence attention mechanism to obtain time sequence weights corresponding to each time characteristic;
and carrying out weighted summation on the N time characteristics according to the time sequence weight to obtain a space-time factor joint characteristic.
A deep cascading neural network-based building energy-consuming space factor characterization device, comprising:
The space feature extraction module is used for inputting space factors related to building energy consumption into a symmetrical residual error network, and extracting space features related to the building energy consumption through the symmetrical residual error network;
the spatial feature weight quantization module is used for carrying out local feature weight quantization on the spatial features through a spatial attention mechanism to obtain quantized spatial features;
The time sequence feature extraction module is used for inputting the time sequence factors related to the quantized spatial features and the building energy consumption into a two-way long-short-term memory network to obtain the time sequence features related to the building energy consumption;
And the time sequence feature weight quantization module is used for carrying out local feature weight quantization on the time sequence features through a time sequence attention mechanism to obtain space-time factor joint features, wherein the space-time factor joint features are used for predicting energy consumption data of the building.
An electronic device comprising a memory having stored thereon computer executable instructions and a processor that when executing the computer executable instructions on the memory performs the above method.
A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the above method.
According to the deep cascade neural network-based building energy time-consuming space factor characterization method and the related product, the characteristics of the space factors can be extracted through the symmetrical residual neural network, and the characteristics of the time sequence factors can be extracted through the bidirectional long-short term memory network. And the space local feature weight is quantized through a space attention mechanism, and the time sequence attention mechanism and the time sequence local feature weight are quantized. The extracted features have more recognition power and robustness, and the accuracy of feature extraction is improved.
Drawings
In order to more clearly illustrate the technical solution of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly described.
Fig. 1 is a flow chart of a method for building energy-consuming time-space factor characterization based on a deep cascaded neural network in one embodiment.
FIG. 2 is a schematic diagram of energy consumption influencing factors in one embodiment.
FIG. 3 is a schematic diagram of spatial feature attention mechanisms in one embodiment.
FIG. 4 is a schematic diagram of a spatio-temporal factor joint feature extraction network in one embodiment.
FIG. 5 is a schematic diagram of a deep cascaded neural network based building energy-time-space factor characterization device in one embodiment.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to fall within the scope of the protection of the application.
It should be understood that the terms "comprises" and "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It is also to be understood that the terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
It should be further understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.
As used in this specification and the appended claims, the term "if" may be construed as "when..once" or "in response to a determination" or "in response to detection" depending on the context. Similarly, the phrase "if a condition or event described is determined" or "if a condition or event described is detected" may be interpreted in context to mean "upon determination" or "in response to determination" or "upon detection of a condition or event described" or "in response to detection of a condition or event described".
Fig. 1 is a flow chart of a method for characterizing building energy-consuming time-space factors based on a deep cascade neural network according to an embodiment, and the method includes steps 102 to 108 as shown in fig. 1. Wherein:
And 102, inputting the spatial factors related to the building energy consumption into a symmetrical residual error network, and extracting the spatial characteristics related to the building energy consumption through the symmetrical residual error network.
And 104, carrying out local feature weight quantization on the spatial features through a spatial attention mechanism to obtain quantized spatial features.
And 106, inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a bi-directional long-term and short-term memory network to obtain the time sequence characteristics related to the building energy consumption.
And step 108, carrying out local feature weight quantization on the time sequence features through a time sequence attention mechanism to obtain space-time factor joint features, wherein the space-time factor joint features are used for predicting energy consumption data of the building.
In one embodiment, a building is an interior space (e.g., office, hallway) formed by various building major elements and forms to meet the needs of a person's production or living. And analyzing the influence factors of the energy consumption of the building by taking the building as a unit.
FIG. 2 is a schematic diagram of energy consumption influencing factors in one embodiment. As shown in fig. 2, the energy consumption influence factor can be generalized into both spatial and temporal aspects. The space factor refers to the building shape characteristics, has important influence on the performances such as building heat preservation and illumination, for example, wall materials have important influence on the building heat preservation performance, and the window orientation has decisive effect on the lighting amount. The BIM (Building Information Modeling, building information model) model is a data source of space factors, and buildings in the BIM model can be split into a plurality of building spaces according to related standards. The single building space factor is divided into two parts, namely the space parameters of the building and the space parameters of the entity components, wherein the space parameters of the building comprise the area, the volume, the floor and the like. The entity components comprise walls, windows, roofs, floors, equipment and the like, the space factors of the first four components comprise parameters such as construction materials, areas, orientations, heat transfer coefficients and the like, and the equipment parameters comprise parameters such as illumination power, heating ventilation air conditioning power, temperature setting and the like.
The timing factor refers to the outdoor and indoor timing states of the building, and has an important effect on the energy consumption of indoor lighting and heating ventilation air conditioning systems, for example: the hot outdoor environment may lead to a drastic rise in indoor refrigeration load. The project takes weather and schedule data as data sources of time sequence factors, wherein the weather data comprises parameters such as outdoor air temperature, solar radiation quantity, solar angle, air humidity and the like. The schedule data comprises schedules of indoor personnel and schedules of light, heating ventilation and air conditioning, other equipment and the like, and parameters of the schedules of the indoor personnel comprise the number of indoor personnel and activity. The schedule parameters of the heating, ventilation and air conditioning system are switch state, cold and warm temperature setting and the like.
For the extraction of the building energy consumption related parameters, BIMserver can be used for completing the analysis of the BIM model, and the IFC (Industry Foundation Class) standard with the widest application is selected as the BIM model input format so as to improve the universality and expandability of the extraction algorithm. The extraction algorithm analyzes the building structure data and the equipment data, and regarding the structure data, builds a tree structure of the building space (IFCSPACE) with the entire building (IFCBUILDING) as a root node. Firstly, extracting the attribute (such as area, orientation and the like) of each building space, splitting each building space into a plurality of enclosure entities (such as a wall IFCWALLSTANDARDCASE, a window IFCWINDOW and the like), and extracting the parameters of three-dimensional geometric coordinate point sequences, materials, heat transfer coefficients and the like of each entity. For the device data, the extraction algorithm processes the air conditioning system (e.g., IFCBOILER, IFCCHILLER, etc.), the lighting device, the other devices in sequence, and the lighting device and the other devices only need to acquire their power and location information. Analysis of the air conditioning system in addition to obtaining basic parameters such as power, temperature settings, etc., requires analysis of the air circulation path (IFCDISTRIBUTIONPORT) to derive the building space served by the hvac system. The sign of the space factor and the time sequence factor related to the building energy consumption are shown in table 1.
TABLE 1 spatial and temporal factor symbols relating to building energy consumption
The parameters of the time sequence factors do not need to be extracted, but the data quality is ensured through data cleaning. Data cleaning includes missing data filling and outlier detection, and the missing data can be filled by using the mean value of the values of the points at the front and rear moments, namelyThe outlier refers to a data object which is significantly deviated from the general data distribution in the time sequence, and because the volume of time sequence data such as weather, schedule and the like related to building energy consumption is not large, the detection of the time sequence data outlier is completed by adopting an autoregressive integrated moving average line (ARIMA) model. The ARIMA (p, d, q) model can be expressed as formula (1), and the predicted value x t of the original sequence is expressed as formula (3) based on the second-order difference x t (formula (2)).
In one embodiment, the building energy consumption related space has heterogeneous interconnectivity with the timing factor, i.e., is independent of each other and organically related. And constructing a deep cascade neural network to complete the joint characterization of the space and the time sequence factors, and designing a double-attention mechanism for the cascade network to realize the local weighting of the space and the time sequence characteristics. The network model includes two main parts: the symmetrical residual neural network is responsible for space factor characterization and the two-way long-short term memory network negative time sequence factor characterization. In order to improve the recognition power and the robustness of the model, a spatial attention mechanism can be designed for the symmetrical residual neural network, and the spatial local feature weight can be quantized; a time sequence attention mechanism is designed for the two-way long-short-term memory network, and time sequence local characteristic weights are quantized.
According to the building energy time-consuming space factor characterization method based on the deep cascade neural network, which is provided by the embodiment, the characteristics of the space factors can be extracted through the symmetrical residual neural network, and the characteristics of the time sequence factors can be extracted through the bidirectional long-short term memory network. And the space local feature weight is quantized through a space attention mechanism, and the time sequence attention mechanism and the time sequence local feature weight are quantized. The extracted features have more recognition power and robustness, and the accuracy of feature extraction is improved.
In one embodiment, the symmetrical residual network comprises a convolution module and a deconvolution module, wherein the convolution module comprises K residual blocks, and the deconvolution module comprises K deconvolution blocks, wherein K is an integer greater than or equal to 1; inputting the spatial factors related to the building energy consumption into a symmetrical residual error network, extracting the spatial characteristics related to the building energy consumption through the symmetrical residual error network, and comprising the following steps: carrying out convolution operation on the space factors through K residual blocks in the convolution module to obtain intermediate features; and deconvoluting the intermediate features through K deconvolution blocks in the deconvolution module to obtain the spatial features related to the building energy consumption.
For example, the symmetric residual neural network is composed of two modules, a convolution module and a deconvolution module. The convolution module consists of five residual blocks, each residual block comprises five layers, the first three layers are residual convolution layers, the fourth layer is batch normalization (Batch Normalization), and the last layer is Dropout. The residual convolution principle is shown in formula (4), when the channel numbers of x and F are different, the dimension of x is changed (formula (5)), RReLU (RECTIFIED LINEAR Unit, linear rectification function) (formula (6)) is used as a convolution activation function, and a i accords with uniform distribution U (l, U), l < U, l & U epsilon [0, 1).
F＝W2*RReLU(W1x),y＝F(x,{Wi})+x (4)
y＝F(x，{Wi})+Wsx (5)
The residual block uses batch normalization (Batch Normalization) to pull the eigenvalue distribution back to the normal distribution, falling in the interval where the activation function is more sensitive to input. Batch normalization first calculates the small batch mean and variance (equation (7)), normalizes the calculation, and makes it in a better nonlinear region (equation (8)) through feature transformation.
The residual block adopts Dropout to alleviate the occurrence of overfitting, so that regularization effect is achieved to a certain extent, the Dropout principle is shown as a formula (9), wherein the Bernoulli function generates a probability r vector, namely a vector of 0 and 1 is randomly generated.
The deconvolution module consists of five deconvolution blocks, each of which includes two layers of two-dimensional transpose deconvolution and batch normalization. The two-dimensional transpose deconvolution principle is shown in equation (10), and k i represents the ith deconvolution kernel. Combining the intermediate feature maps of each deconvolutionAnd obtaining a deconvolution final output characteristic diagram.
In one embodiment, the spatial attention mechanism includes a channel attention module and a spatial attention module; carrying out local feature weight quantization on the spatial features through a spatial attention mechanism to obtain quantized spatial features, wherein the method comprises the following steps: inputting the spatial features to a channel attention module to obtain channel attention weights, and obtaining first features according to the spatial features and the channel attention weights; and inputting the first characteristic into a spatial attention module to obtain spatial attention weight, and obtaining quantized spatial characteristics according to the first characteristic and the spatial attention weight. The first feature is the intermediate feature of space weight quantization.
In one embodiment, the channel attention module includes a first average pooling layer, a first maximum pooling layer, and a fully connected layer; inputting the spatial features to a channel attention module to obtain channel attention weights, comprising: respectively inputting the spatial features into a first average pooling layer and a first maximum pooling layer to obtain the spatial features after average pooling and the spatial features after maximum pooling; adding the average pooled spatial features and the maximum pooled spatial features through a full-connection layer to obtain attention weights; the spatial attention module comprises a second average pooling layer, a second maximum pooling layer and a convolution layer; inputting the first feature to a spatial attention module to obtain a spatial attention weight, comprising: respectively inputting the first features into a second average pooling layer and a second maximum pooling layer to obtain first features after average pooling and first features after maximum pooling; and carrying out convolution operation on the average pooled first characteristic and the maximum pooled first characteristic through a convolution layer to obtain the spatial attention weight.
Specifically, the local feature weight of the spatial feature is quantized, specifically, the local feature weight is quantized by channel information and spatial information of the spatial feature output by the deconvolution module. As shown in fig. 3, the attention mechanism includes a channel attention module and a spatial attention module, where the channel attention module calculates global average pooling and global maximum pooling information of the feature map, and then adds the global average pooling and global maximum pooling information through the full connection layer to obtain a channel attention parameter (formula (11)), i.e., a channel attention weight. The spatial attention module carries out global maximization and average pooling on the coordinates of each channel characteristic diagram to obtain two characteristic diagrams, and then convolves the characteristic diagrams to obtain a spatial attention parameter (formula (12)), namely spatial attention weight.
In one embodiment, the two-way long-short-term memory network includes a forward-propagation network and a backward-propagation network; inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a two-way long-short-term memory network to obtain the time sequence characteristics related to the building energy consumption, wherein the time sequence factors comprise: inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a forward propagation network to obtain first time sequence characteristics related to the building energy consumption; inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a backward propagation network to obtain second time sequence characteristics related to the building energy consumption; and constructing and obtaining time sequence characteristics related to building energy consumption according to the first time sequence characteristics and the second time sequence characteristics.
For the time sequence factor, the project is characterized by adopting a two-way long-short-term memory network. The output at the current moment is not only related to the previous state, but also possibly related to the future state, the bidirectional long-short-term memory network characterizes the input state through the forward propagation direction and the backward propagation direction, and the final characterization result is obtained by integrating the two-direction results. The quantized spatial features and the timing factors are input into a two-way long-short-term memory network, a first timing feature related to building energy consumption is obtained through a forward propagation network, a second timing feature related to building energy consumption is obtained through backward propagation, and the timing features are obtained through construction according to the first timing feature and the second timing feature.
In the embodiment provided by the application, the time sequence factors comprise time sequence factors corresponding to N times, and the time sequence features comprise time features corresponding to N times, wherein N is an integer greater than or equal to 1; inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a two-way long-short-term memory network to obtain the time sequence characteristics related to the building energy consumption, wherein the time sequence factors comprise: inputting the quantized spatial features and N time factors into a forward propagation network to obtain N first time features; inputting the quantized spatial features and N time factors into a backward propagation network to obtain N second time features; and respectively combining the first time feature and the second time feature corresponding to the same time to obtain N time features.
Specifically, the local feature weight quantization is performed on the time sequence feature through a time sequence attention mechanism to obtain a space-time factor joint feature, which comprises the following steps: inputting the N time characteristics into a time sequence attention mechanism to obtain time sequence weights corresponding to each time characteristic; and weighting and summing the N moment characteristics according to the time sequence weight to obtain the space-time factor joint characteristic.
FIG. 4 is a schematic diagram of a spatio-temporal factor joint feature extraction network in one embodiment, as shown in FIG. 4. The space-time factor joint feature extraction network comprises a symmetrical residual network, a spatial attention mechanism, a two-way long-short term memory network and a time sequence attention mechanism. In the two-way long and short memory network, taking forward propagation as an example, as shown in formulas (a) - (e), each LSTM unit comprises a forgetting gateInput gate/>Output door/>Memory cell/>Hidden state based on last cell/>And memory state/>Hidden state output of current cell/>(18). Combining the feature information of forward propagation and backward propagation to generate a feature representation/>, of a time point t
To further quantify the importance of the local timing feature, the timing feature attention mechanism assigns a relative weight α t to the feature h t output at the time point t, and the calculation principle of the weight α t is shown in the formula (19) and the formula (20). After the calculation of the relative weights at all time points is completed, the output characteristics of the two-way long-short-term memory network can be expressed as follows
FIG. 5 is a schematic diagram of a deep cascaded neural network-based building energy-time-space factor characterization device in one embodiment, as shown in FIG. 5, comprising:
The spatial feature extraction module 502 is configured to input spatial factors related to building energy consumption into a symmetric residual error network, and extract spatial features related to the building energy consumption through the symmetric residual error network;
The spatial local feature weight quantization module 504 is configured to perform local feature weight quantization on the spatial features through a spatial attention mechanism, so as to obtain quantized spatial features;
The time sequence feature extraction module 506 is configured to input the time sequence factors related to the quantized spatial feature and the building energy consumption into a two-way long-short-term memory network to obtain a time sequence feature related to the building energy consumption;
And the time sequence local feature weight quantization module 508 is used for carrying out local feature weight quantization on the time sequence features through a time sequence attention mechanism to obtain space-time factor joint features, wherein the space-time factor joint features are used for predicting energy consumption data of the building.
The building energy time-consuming space factor characterization device based on the deep cascade neural network provided by the embodiment can extract the characteristics of space factors through the symmetrical residual neural network, and extract the characteristics of time sequence factors through the bidirectional long-short term memory network. And the space local feature weight is quantized through a space attention mechanism, and the time sequence attention mechanism and the time sequence local feature weight are quantized. The extracted features have more recognition power and robustness, and the accuracy of feature extraction is improved.
In one embodiment, the symmetrical residual network includes a convolution module and a deconvolution module, the convolution module includes K residual blocks, and the deconvolution module includes K deconvolution blocks, where K is an integer greater than or equal to 1; the spatial feature extraction module 502 is further configured to perform convolution operation on the spatial factor through K residual blocks in the convolution module, so as to obtain an intermediate feature; and deconvoluting the intermediate features through K deconvolution blocks in the deconvolution module to obtain the spatial features related to the building energy consumption.
In one embodiment, the spatial feature weight quantization module 504 is further configured to input the spatial feature to the channel attention module to obtain a channel attention weight, and obtain a first feature according to the spatial feature and the channel attention weight; and inputting the first characteristic into the spatial attention module to obtain a spatial attention weight, and obtaining the quantized spatial characteristic according to the first characteristic and the spatial attention weight.
In one embodiment, the channel attention module includes a first average pooling layer, a first maximum pooling layer, and a fully connected layer; the spatial attention module comprises a second average pooling layer, a second maximum pooling layer and a convolution layer; the spatial feature weight quantization module 504 is further configured to input the spatial features into a first average pooling layer and a first maximum pooling layer, respectively, to obtain an average pooled spatial feature and a maximum pooled spatial feature; adding the spatial features after the average pooling and the spatial features after the maximum pooling to obtain the attention weight through the full connection layer; inputting the first features into the second average pooling layer and the second maximum pooling layer respectively to obtain first features after average pooling and first features after maximum pooling; and carrying out convolution operation on the average pooled first feature and the maximum pooled first feature through the convolution layer to obtain the space attention weight.
In one embodiment, the two-way long-short-term memory network includes a forward propagation network and a backward propagation network; the timing characteristic extraction module 506 is further configured to input the timing factor related to the quantized spatial characteristic and the building energy consumption into the forward propagation network, so as to obtain a first timing characteristic related to the building energy consumption; inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into the backward propagation network to obtain second time sequence characteristics related to the building energy consumption; and constructing and obtaining the time sequence characteristics related to the building energy consumption according to the first time sequence characteristics and the second time sequence characteristics.
In one embodiment, the time sequence factors include time sequence factors corresponding to N times, and the time sequence features include time features corresponding to N times, wherein N is an integer greater than or equal to 1; the timing characteristic extraction module 506 is further configured to input the quantized spatial characteristics and N time factors into the forward propagation network, to obtain N first time characteristics; inputting the quantized spatial features and N time factors into the backward propagation network to obtain N second time features; and respectively combining the first time characteristic and the second time characteristic corresponding to the same time to obtain N time characteristics.
In one embodiment, the time sequence feature weight quantization module 508 is further configured to input N time sequence features to the time sequence attention mechanism, and obtain a time sequence weight corresponding to each of the time sequence features; and carrying out weighted summation on the N time characteristics according to the time sequence weight to obtain a space-time factor joint characteristic.
In one embodiment, an electronic device is provided that includes a memory having computer-executable instructions stored thereon and a processor that when executed performs the methods of the above embodiments.
A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method of the above embodiments.
In practical applications, the electronic device may further include other necessary elements, including but not limited to any number of input/output systems, processors, controllers, memories, etc., and all electronic devices capable of implementing the cross-cloud platform big data management method according to the embodiments of the present application are within the protection scope of the present application.
The memory includes, but is not limited to, random access memory (random access memory, RAM), read-only memory (ROM), erasable programmable read-only memory (erasable programmable read only memory, EPROM), or portable read-only memory (compact disc read to only memory, CD to ROM) for associated instructions and data.
The input system is used for inputting data and/or signals, and the output system is used for outputting data and/or signals. The output system and the input system may be separate devices or may be a single device.
The processor may include one or more processors, including for example one or more central processing units (central processing unit, CPU), which in the case of a CPU, may be a single-core CPU or a multi-core CPU. The processor may also include one or more special purpose processors, which may include GPUs, FPGAs, etc., for performing acceleration processing.
The memory is used to store program codes and data for the network device.
The processor is used to call the program code and data in the memory to perform the steps of the method embodiments described above. Reference may be made specifically to the description of the method embodiments, and no further description is given here.
In the several embodiments provided by the present application, it should be understood that the disclosed systems and methods may be implemented in other ways. For example, the division of the unit is merely a logic function division, and there may be another division manner when actually implemented, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted or not performed. The coupling or direct coupling or communication connection shown or discussed with each other may be through some interface, system or unit indirect coupling or communication connection, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the application, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable system. The computer instructions may be stored in or transmitted across a computer-readable storage medium. The computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, fiber optic, digital subscriber line (digital subscriber line, DSL)), or wireless (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a read-only memory (ROM), or a random-access memory (random access memory, RAM), or a magnetic medium such as a floppy disk, a hard disk, a magnetic tape, a magnetic disk, or an optical medium such as a digital versatile disk (DIGITAL VERSATILE DISC, DVD), or a semiconductor medium such as a Solid State Disk (SSD), or the like.
While the application has been described with reference to certain preferred embodiments, it will be understood by those skilled in the art that various changes and substitutions of equivalents may be made without departing from the spirit and scope of the application. Therefore, the protection scope of the application is subject to the protection scope of the claims.

Claims (10)

1. The method for characterizing the building energy time-consuming space factor based on the deep cascade neural network is characterized by comprising the following steps of:
Inputting spatial factors related to building energy consumption into a symmetrical residual error network, and extracting spatial characteristics related to the building energy consumption through the symmetrical residual error network, wherein the spatial factors refer to building body characteristics;
carrying out local feature weight quantization on the spatial features through a spatial attention mechanism to obtain quantized spatial features;
Inputting the quantized spatial characteristics and time sequence factors related to the building energy consumption into a two-way long-short-term memory network to obtain the time sequence characteristics related to the building energy consumption, wherein the time sequence factors refer to outdoor and indoor time sequence states of the building, and weather and schedule data are used as data sources of the time sequence factors;
And carrying out local feature weight quantization on the time sequence features through a time sequence attention mechanism to obtain space-time factor joint features, wherein the space-time factor joint features are used for predicting energy consumption data of the building.
2. The method according to claim 1, wherein the symmetric residual network comprises a convolution module and a deconvolution module, wherein the convolution module comprises K residual blocks, and the deconvolution module comprises K deconvolution blocks, wherein K is an integer greater than or equal to 1;
The step of inputting the spatial factors related to the building energy consumption into a symmetrical residual error network, and extracting the spatial characteristics related to the building energy consumption through the symmetrical residual error network comprises the following steps:
Performing convolution operation on the space factor through K residual blocks in the convolution module to obtain an intermediate feature;
and deconvoluting the intermediate features through K deconvolution blocks in the deconvolution module to obtain the spatial features related to the building energy consumption.
3. The method of claim 1, wherein the spatial attention mechanism comprises a channel attention module and a spatial attention module; the step of carrying out local feature weight quantization on the spatial features through a spatial attention mechanism to obtain quantized spatial features comprises the following steps:
Inputting the spatial features to the channel attention module to obtain channel attention weights, and obtaining first features according to the spatial features and the channel attention weights;
And inputting the first feature into the spatial attention module to obtain spatial attention weight, and obtaining the quantized spatial feature according to the first feature and the spatial attention weight.
4. A method according to claim 3, wherein the channel attention module comprises a first average pooling layer, a first maximum pooling layer, and a fully connected layer; the step of inputting the spatial feature to the channel attention module to obtain a channel attention weight includes:
Respectively inputting the spatial features into a first average pooling layer and a first maximum pooling layer to obtain spatial features after average pooling and spatial features after maximum pooling;
adding the spatial features after the average pooling and the spatial features after the maximum pooling to obtain the attention weight through the full connection layer;
The spatial attention module comprises a second average pooling layer, a second maximum pooling layer and a convolution layer; the step of inputting the first feature to the spatial attention module to obtain a spatial attention weight includes:
Inputting the first features into the second average pooling layer and the second maximum pooling layer respectively to obtain first features after average pooling and first features after maximum pooling;
and carrying out convolution operation on the average pooled first feature and the maximum pooled first feature through the convolution layer to obtain the space attention weight.
5. The method of any one of claims 1 to 4, wherein the two-way long-short-term memory network comprises a forward-propagation network and a backward-propagation network;
Inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a two-way long-short-term memory network to obtain the time sequence characteristics related to the building energy consumption, wherein the time sequence factors comprise:
Inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into the forward propagation network to obtain first time sequence characteristics related to the building energy consumption;
Inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into the backward propagation network to obtain second time sequence characteristics related to the building energy consumption;
and constructing and obtaining the time sequence characteristics related to the building energy consumption according to the first time sequence characteristics and the second time sequence characteristics.
6. The method of claim 5, wherein the timing factors include time factors corresponding to N times, and wherein the timing features include time features corresponding to N times, wherein N is an integer greater than or equal to 1;
Inputting the time sequence factors related to the quantized spatial characteristics and the building energy consumption into a two-way long-short-term memory network to obtain the time sequence characteristics related to the building energy consumption, wherein the time sequence factors comprise:
Inputting the quantized spatial features and N time factors into the forward propagation network to obtain N first time features;
Inputting the quantized spatial features and N time factors into the backward propagation network to obtain N second time features;
And respectively combining the first time characteristic and the second time characteristic corresponding to the same time to obtain N time characteristics.
7. The method of claim 6, wherein the performing local feature weighting on the time series feature by a time series attention mechanism to obtain a space-time factor joint feature comprises:
Inputting N time characteristics into the time sequence attention mechanism to obtain time sequence weights corresponding to each time characteristic;
and carrying out weighted summation on the N time characteristics according to the time sequence weight to obtain a space-time factor joint characteristic.
8. A deep cascading neural network-based building energy-consuming space factor characterization device, comprising:
The space feature extraction module is used for inputting space factors related to building energy consumption into a symmetrical residual error network, and extracting space features related to the building energy consumption through the symmetrical residual error network, wherein the space factors refer to building body features;
the spatial local feature weight quantization module is used for carrying out local feature weight quantization on the spatial features through a spatial attention mechanism to obtain quantized spatial features;
The time sequence feature extraction module is used for inputting the quantized spatial features and time sequence factors related to the building energy consumption into a two-way long-short-term memory network to obtain the time sequence features related to the building energy consumption, wherein the time sequence factors refer to outdoor and indoor time sequence states of the building, and weather and schedule data are used as data sources of the time sequence factors;
And the time sequence local feature weight quantization module is used for carrying out local feature weight quantization on the time sequence features through a time sequence attention mechanism to obtain space-time factor joint features, wherein the space-time factor joint features are used for predicting energy consumption data of the building.
9. An electronic device comprising a memory having stored thereon computer executable instructions and a processor which when executed performs the method of any of claims 1 to 7.
10. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method of any of claims 1 to 7.
CN202010956592.XA 2020-09-11 2020-09-11 Building energy time-consuming space factor characterization method based on deep cascade neural network and related products Active CN112232479B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010956592.XA CN112232479B (en) 2020-09-11 2020-09-11 Building energy time-consuming space factor characterization method based on deep cascade neural network and related products

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010956592.XA CN112232479B (en) 2020-09-11 2020-09-11 Building energy time-consuming space factor characterization method based on deep cascade neural network and related products

Publications (2)

Publication Number Publication Date
CN112232479A CN112232479A (en) 2021-01-15
CN112232479B true CN112232479B (en) 2024-06-14

Family

ID=74116752

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010956592.XA Active CN112232479B (en) 2020-09-11 2020-09-11 Building energy time-consuming space factor characterization method based on deep cascade neural network and related products

Country Status (1)

Country Link
CN (1) CN112232479B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117132135B (en) * 2023-10-23 2024-02-06 陕西天润科技股份有限公司 Urban informatization management system and method based on digital twinning

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10739741B2 (en) * 2009-06-22 2020-08-11 Johnson Controls Technology Company Systems and methods for detecting changes in energy usage in a building
US20160291622A1 (en) * 2015-03-31 2016-10-06 Enernoc, Inc. System for weather induced facility energy consumption characterization
CN109063903B (en) * 2018-07-19 2021-04-09 山东建筑大学 Building energy consumption prediction method and system based on deep reinforcement learning
US20200082269A1 (en) * 2018-09-12 2020-03-12 Nvidia Corporation Memory efficient neural networks
CN109389091B (en) * 2018-10-22 2022-05-03 重庆邮电大学 Character recognition system and method based on combination of neural network and attention mechanism
US11361225B2 (en) * 2018-12-18 2022-06-14 Microsoft Technology Licensing, Llc Neural network architecture for attention based efficient model adaptation
CN109754015B (en) * 2019-01-02 2021-01-26 京东方科技集团股份有限公司 Neural networks for drawing multi-label recognition and related methods, media and devices
GB2584614B (en) * 2019-05-17 2021-10-13 Arbnco Ltd Optimising building energy use
CN111126488B (en) * 2019-12-24 2023-08-18 威创集团股份有限公司 Dual-attention-based image recognition method
CN111260136A (en) * 2020-01-17 2020-06-09 上海电力大学 Building short-term load prediction method based on ARIMA-LSTM combined model

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
"Spatial and Channel Attention Based Convolutional Neural Networks for Modeling Noisy Speech";Sirui Xu，et.al;《ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)》;全文 *
"数据驱动的建筑使用能耗预测及其应用研究";周惟;《中国优秀硕士学位论文全文数据库》(第第1期期);全文 *

Also Published As

Publication number Publication date
CN112232479A (en) 2021-01-15

Similar Documents

Publication Publication Date Title
CN112183826B (en) Building energy consumption prediction method based on deep cascade generation countermeasure network and related products
Zhang et al. A systematic feature selection procedure for short-term data-driven building energy forecasting model development
Sha et al. Overview of computational intelligence for building energy system design
Jihad et al. Forecasting the heating and cooling load of residential buildings by using a learning algorithm “gradient descent”, Morocco
Wang et al. Multi-objective optimization (MOO) for high-rise residential buildings’ layout centered on daylight, visual, and outdoor thermal metrics in China
Park et al. Predictive model for PV power generation using RNN (LSTM)
CN103049612B (en) Building indoor environment optimization method based on model order reduction technology
Goudarzi et al. Predictive modelling of building energy consumption based on a hybrid nature-inspired optimization algorithm
Bucking et al. An information driven hybrid evolutionary algorithm for optimal design of a net zero energy house
Abediniangerabi et al. A data-driven framework for energy-conscious design of building facade systems
Wang et al. A novel efficient optimization algorithm for parameter estimation of building thermal dynamic models
CN112152840B (en) Sensor deployment method and system based on BIM and analog simulation
Kirimtat et al. Multi-objective optimization for shading devices in buildings by using evolutionary algorithms
CN114764521A (en) Garden building shape optimization method and system based on genetic algorithm
CN112232479B (en) Building energy time-consuming space factor characterization method based on deep cascade neural network and related products
CN116205425A (en) Low-carbon park cold-hot electric load prediction method based on typical database
Akbarzadeh et al. Heating-cooling monitoring and power consumption forecasting using lstm for energy-efficient smart management of buildings: A computational intelligence solution for smart homes
Ku et al. Building electric energy prediction modeling for BEMS using easily obtainable weather factors with Kriging model and data mining
Fang et al. Comprehensive clustering method to determine coincident design day for air-conditioning system design
US20230161929A1 (en) Estimation method, simulation method, estimation device, and estimation program
Yan et al. An improved temperature prediction technique for hvac units using intelligent algorithms
Zhang et al. Enhancing multi-scenario data-driven energy consumption prediction in campus buildings by selecting appropriate inputs and improving algorithms with attention mechanisms
CN114925929A (en) Building load prediction method and device based on degree days
Park et al. Predictive model for solar insolation using the deep learning technique
Chu et al. Data-Driven Energy Prediction in Residential Buildings using LSTM and 1-D CNN.

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
GR01 Patent grant
GR01 Patent grant