CN116227741A - Water chilling unit energy saving method and device based on self-adaptive algorithm and related medium - Google Patents
Water chilling unit energy saving method and device based on self-adaptive algorithm and related medium Download PDFInfo
- Publication number
- CN116227741A CN116227741A CN202310496732.3A CN202310496732A CN116227741A CN 116227741 A CN116227741 A CN 116227741A CN 202310496732 A CN202310496732 A CN 202310496732A CN 116227741 A CN116227741 A CN 116227741A
- Authority
- CN
- China
- Prior art keywords
- model
- load prediction
- cold load
- indoor
- data
- 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
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 104
- 238000000034 method Methods 0.000 title claims abstract description 90
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 74
- 238000012545 processing Methods 0.000 claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000001816 cooling Methods 0.000 claims description 51
- 230000003044 adaptive effect Effects 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 17
- 230000006872 improvement Effects 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 15
- 230000007613 environmental effect Effects 0.000 claims description 14
- 238000010276 construction Methods 0.000 claims description 12
- 230000002159 abnormal effect Effects 0.000 claims description 11
- 230000015654 memory Effects 0.000 claims description 11
- 238000011176 pooling Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 10
- 238000004134 energy conservation Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 claims description 8
- 238000003066 decision tree Methods 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 5
- 229910002092 carbon dioxide Inorganic materials 0.000 claims description 4
- 239000001569 carbon dioxide Substances 0.000 claims description 4
- 230000007717 exclusion Effects 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 241000965478 Darksidea epsilon Species 0.000 claims description 3
- 238000005265 energy consumption Methods 0.000 abstract description 14
- 238000004364 calculation method Methods 0.000 description 13
- 230000008901 benefit Effects 0.000 description 10
- 230000000694 effects Effects 0.000 description 8
- 238000013527 convolutional neural network Methods 0.000 description 7
- 230000004913 activation Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 238000009826 distribution Methods 0.000 description 6
- 239000004973 liquid crystal related substance Substances 0.000 description 6
- 239000011159 matrix material Substances 0.000 description 4
- 230000002787 reinforcement Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010187 selection method Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000004378 air conditioning Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 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
- G06N20/00—Machine learning
-
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Evolutionary Computation (AREA)
- Tourism & Hospitality (AREA)
- Mathematical Physics (AREA)
- General Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Medical Informatics (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Operations Research (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Air Conditioning Control Device (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a water chilling unit energy saving method, a device and a related medium based on a self-adaptive algorithm, wherein the method comprises the following steps: collecting environment data and cold station equipment data, and performing data processing on the environment data and the cold station equipment data to obtain a model data set; training and learning the model data set by adopting an XGBoost model to construct a first indoor cold load prediction model; training and learning the model data set by adopting a LightGBM model to construct a second indoor cold load prediction model; training and learning the model data set by adopting a CNN-LSTM model to construct a third indoor cold load prediction model; performing self-adaptive selection through a FRRMAB method to obtain a target indoor cold load prediction model, and taking a cold load prediction result of the target indoor cold load prediction model as a final cold load prediction result; and outputting a corresponding energy-saving strategy to the water chilling unit based on the final cold load prediction result. The invention can improve the energy consumption management efficiency of the water chilling unit and achieve the purpose of energy saving management.
Description
Technical Field
The invention relates to the technical field of energy consumption management, in particular to a water chilling unit energy saving method and device based on a self-adaptive algorithm and a related medium.
Background
Currently, building energy consumption accounts for a higher proportion of the total social energy consumption, wherein the public building energy consumption intensity (energy consumption per unit area) accounts for the largest proportion of the building energy consumption, and the air conditioning system energy consumption in the public building energy consumption accounts for a larger proportion of the whole building energy consumption. Meanwhile, in order to create a comfortable indoor environment, a cold water main machine in a building cooled by a water system is usually set at a lower temperature, or the temperature of the cold water main machine is simply controlled according to manual experience, and the two modes can provide surplus cold energy for the indoor environment, so that unnecessary energy consumption exists.
With the vigorous development of AIoT (artificial intelligence Internet of things) technology, environmental conditions and cold station parameter changes of project sites can be sensed through sensors at present, and temperature recommended values are further given by combining a cold load prediction algorithm, so that the effects of accurate energy conservation and accurate detail cost can be achieved while good office experience of users in a building is ensured. However, the cold load prediction algorithm in the prior art has certain defects, such as low applicability, insufficient prediction precision and the like, so that how to improve the cold load prediction precision, further improve the management efficiency of the cold water host machine, and realize the effects of energy conservation and cost saving are problems to be solved by those skilled in the art.
Disclosure of Invention
The embodiment of the invention provides a water chilling unit energy saving method, a device, computer equipment and a storage medium based on a self-adaptive algorithm, which aim to improve the energy consumption management efficiency of the water chilling unit and achieve the purpose of energy saving management.
In a first aspect, an embodiment of the present invention provides a water chiller energy saving method based on an adaptive algorithm, including:
collecting environment data and cold station equipment data, and performing data processing on the environment data and the cold station equipment data to obtain a model data set;
training and learning the model data set by adopting an XGBoost model to construct a first indoor cold load prediction model;
training and learning the model dataset by adopting a LightGBM model to construct a second indoor cold load prediction model;
training and learning the model data set by adopting a CNN-LSTM model to construct a third indoor cold load prediction model;
performing self-adaptive selection on the first indoor cold load prediction model, the second indoor cold load prediction model and the third indoor cold load prediction model through a FRRMAB method to obtain a target indoor cold load prediction model, and taking a cold load prediction result of the target indoor cold load prediction model as a final cold load prediction result;
And outputting a corresponding energy-saving strategy to the water chilling unit based on the final cold load prediction result.
In a second aspect, an embodiment of the present invention provides a water chiller energy saving device based on an adaptive algorithm, including:
the data acquisition unit is used for acquiring environment data and cold station equipment data, and carrying out data processing on the environment data and the cold station equipment data to obtain a model data set;
the first construction unit is used for training and learning the model data set by adopting an XGBoost model so as to construct a first indoor cold load prediction model;
the second construction unit is used for training and learning the model data set by adopting a LightGBM model so as to construct a second indoor cold load prediction model;
the third construction unit is used for training and learning the model data set by adopting a CNN-LSTM model so as to construct a third indoor cold load prediction model;
the model selection unit is used for adaptively selecting the first indoor cold load prediction model, the second indoor cold load prediction model and the third indoor cold load prediction model through a FRRMAB method to obtain a target indoor cold load prediction model, and taking the result of cold load prediction of the target indoor cold load prediction model as a final cold load prediction result;
And the strategy output unit is used for outputting a corresponding energy-saving strategy to the water chilling unit based on the final cold load prediction result.
In a third aspect, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor implements the water chiller energy saving method based on the adaptive algorithm according to the first aspect when executing the computer program.
In a fourth aspect, an embodiment of the present invention provides a computer readable storage medium, where a computer program is stored on the computer readable storage medium, where the computer program when executed by a processor implements the water chiller energy saving method based on the adaptive algorithm according to the first aspect.
According to the embodiment of the invention, the indoor cold load prediction model is respectively constructed through the XGBoost model, the LightGBM model and the CNN-LSTM model, the indoor cold load prediction model with the optimal current time period is selected through the self-adaptive algorithm so as to predict and output a final cold load prediction result, and then an energy-saving strategy is set for the water chilling unit based on the final cold load prediction result, so that the energy consumption management efficiency of the water chilling unit can be improved, and the energy-saving effect is achieved while the indoor environment comfort level is ensured.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a water chiller energy saving method based on an adaptive algorithm according to an embodiment of the present invention;
FIG. 2 is a network structure diagram of a water chiller energy saving method based on an adaptive algorithm according to an embodiment of the present invention;
FIG. 3 is a schematic sub-flowchart of a water chiller energy saving method based on an adaptive algorithm according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of another sub-flow of a water chiller energy saving method based on an adaptive algorithm according to an embodiment of the present invention;
FIG. 5 is a schematic block diagram of a chiller energy saving device based on an adaptive algorithm according to an embodiment of the present invention;
FIG. 6 is a sub-schematic block diagram of a chiller energy saving device based on an adaptive algorithm according to an embodiment of the present invention;
Fig. 7 is another sub-schematic block diagram of a water chiller energy saving device based on an adaptive algorithm according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention 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 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.
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 invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. 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.
Referring to fig. 1, an embodiment of the present invention provides a water chiller energy saving method based on an adaptive algorithm, which specifically includes: steps S101-S106.
S101, acquiring environment data and cold station equipment data, and performing data processing on the environment data and the cold station equipment data to obtain a model data set;
as shown in fig. 3, step S101 specifically includes: s301 to S304.
S301, collecting environmental data and cold station equipment data through a sensor; the environment data comprise outdoor temperature, outdoor humidity, indoor temperature, indoor humidity and indoor carbon dioxide concentration, and the cold station equipment data comprise cooling tower water inlet temperature, cooling tower water outlet temperature, chilled water supply temperature, chilled water return temperature, chilled water flow and total active power of a cold water host;
s302, carrying out missing value processing and abnormal value processing on the environment data and the cold station equipment data, and carrying out interpolation filling based on missing value processing results and abnormal value processing;
S303, calculating indoor cooling load according to the following formula according to the cooling station equipment data:
wherein Q represents the indoor cooling load, c represents the specific heat capacity of water
ρ represents the density of water
G represents flow, ++>
Indicating the return water temperature of chilled water->
Indicating chilled water supply temperature;
s304, taking the indoor cooling load as a tag, deleting the data with the cooling load of 0 when the water chiller is not cooled, and constructing a model data set related to the environmental data and the cold station equipment data.
In this embodiment, the sensors are installed to collect project site environment data and cold station equipment data, including, but not limited to, outdoor temperature, outdoor humidity, indoor temperature, indoor humidity, indoor carbon dioxide concentration, cooling tower water inlet temperature, cooling tower water outlet temperature, chilled water supply temperature, chilled water return temperature, chilled water flow rate and total active power of the cold water host, and data can be collected once every 5 minutes. And processing the missing value and the abnormal value of the collected environmental data and the cold station equipment data. When the abnormal value is identified, a reasonable value range of each data field is set, and data exceeding the range is identified as abnormal data. The total active power of the water chilling host can also be normally distributed by taking a certain time interval as a unit, points outside 3 sigma are identified as abnormal data caused by measuring drift in the running process, and the identified abnormal value can be subjected to line interpolation together with the missing value after being emptied.
Further, the indoor cooling load is calculated according to the chilled water flow, the chilled water supply temperature and the chilled water return temperature, as follows:
wherein Q represents indoor cooling load, c is specific heat capacity of water
ρ is the density of water
G is flow, unit is->
，/>
Indicating the return water temperature of chilled water->
The water supply temperature of the chilled water is expressed in degrees celsius.
And (3) taking the indoor cold load as a tag, deleting the data with the cold load of 0 when not refrigerating, and obtaining a model dataset by using the environmental data collected by the sensor and the cold station equipment data construction characteristics so as to be used for constructing each subsequent model.
S102, training and learning the model data set by adopting an XGBoost model to construct a first indoor cold load prediction model;
specifically, step S102 includes:
inputting the model dataset into each decision tree sub-model of the XGBoost model, and outputting a corresponding first cooling load prediction result by the XGBoost model;
optimizing the XGBoost model by taking the gradient descent direction of the loss function as an optimization target according to the following steps:
in the method, in the process of the invention,
representing the first cold load prediction result of the XGBoost model,/>
Representing the first cooling load true value, L in Obj represents a differentiable convex loss function for measuring the first cooling load predictor +. >
And a first cooling load true value
Error of each other>
Representing a regular term for controlling the complexity of the model to avoid overfitting, T representing the number of leaf nodes,/->
Is a coefficient of->
Is a regularized term of L2.
XGBoost and LightGBM are both algorithms based on GBDT (Gradient Boosting Decision Tree), a decision tree is taken as a submodel, the gradient descent direction of a loss function is taken as an optimization target, and weak learners are combined to obtain a strong learner. The objective function Obj for optimization by the XGBoost model is difficult to optimize in the euclidean space by the conventional method, so the XGBoost model introduces a second-order taylor formula to approximate and simplify the objective function.
For each sub-model, the smaller the objective function value, the better the model effect, as follows:
wherein t represents the iteration of the t-th round,
is the model obtained by iteration of the t-th round, and the t-1 round of cold load prediction result
Plus t-round model fitting residual case and the real value of the cold load +.>
Comparing the calculated losses and the regularization term +.>
Prevent overfitting, the whole model is towards decrease +.>
Is a direction iteration of (a). For each sub-model, it is generally not possible to enumerate all possible candidate trees when selecting the optimal cut point, so XGBoost uses a greedy algorithm to evaluate the split feature by calculating the amount of change in the objective function values before and after node splitting, starting from the root node of the tree.
The XGBoost model carries out second-order Taylor expansion on the loss function, not only can the accuracy be increased, but also the self-definition of the loss function can be supported, and the regular term is added in the objective function, so that the over-fitting is prevented, and the superiority in effect is presented in the time sequence prediction problem in each field.
S103, training and learning the model data set by adopting a LightGBM model to construct a second indoor cold load prediction model;
specifically, step S103 includes:
sampling the model data set by adopting a single-side gradient sampling method, retaining a% of large gradient samples before the model data set, and randomly extracting b% of small gradient samples;
performing mutual exclusion feature binding processing on the large gradient sample to reduce the feature quantity of the large gradient sample;
and discretizing the large gradient sample with the reduced feature quantity into k discrete features by using a histogram algorithm, and traversing the k discrete features to find the optimal splitting point and the splitting threshold value so as to obtain a second cold load prediction result corresponding to the LightGBM model.
The LightGBM has the characteristics of light weight and high training speed, adopts a single-edge Gradient Sampling (GOSS) method to sample samples, retains a% of large Gradient samples before Sampling and randomly extracts b% of small Gradient samples, and simultaneously gives the small Gradient samples when calculating information gain in order not to change the data distribution of the samples 
Weight coefficient of (c) in the above-mentioned formula (c).
In addition, the LightGBM also adopts a mutual exclusion feature binding (Exclusive Feature Bundling, EFB) method, and fusion binding is carried out on some features to reduce the number of the features, a histogram algorithm is used for discretizing continuous features into k discrete features, meanwhile, a histogram with the width of k is constructed for statistical information, and the optimal splitting point and splitting threshold value can be found by traversing k barrels. The LightGBM algorithm has low memory consumption, reduces time complexity by adopting various modes, and has good effect on a plurality of time sequence prediction problems.
S104, training and learning the model data set by adopting a CNN-LSTM model to construct a third indoor cold load prediction model;
specifically, step S104 specifically includes:
inputting the data in the model data set into a convolution layer of a CNN network, and extracting data characteristics in a self-adaptive manner by using a wide convolution kernel;
carrying out pooling treatment on the data features by using a maximum pooling layer so as to reduce the dimension of the data features;
inputting the data characteristics with reduced dimensionality into an LSTM layer, and outputting corresponding sequence characteristics by the LSTM layer;
and classifying and outputting the sequence characteristics through the full connection layer to obtain a third cooling load prediction result corresponding to the CNN-LSTM model.
The CNN-LSTM network consists of CNN (Convolution Neural Network, convolutional neural network) and LSTM (Long Short Term Memory, long and short term memory network). The core of the CNN is a convolution layer and a pooling layer, the convolution layer creates a feature map through convolution and activation operation, the pooling layer reduces the feature map through pooling operation to reduce the calculation time, and the specific formula is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
representing the output of the convolution operation,/>
Input features representing convolutional layers, +.>
Is a convolution kernel +.>
Is a bias item->
Representation featureSyndrome/pattern of (I/O)>
Representing an activation function;
wherein, the liquid crystal display device comprises a liquid crystal display device,
representing the pooled output, +.>
Representing a pooling operation.
LSTM belongs to one of the circulating neural networks, can effectively transfer and express information in a long-time sequence, and solves the problems of gradient elimination and gradient explosion existing in the traditional circulating neural network. The basic units of the LSTM network introduce input gates, forget gates, and output gates to control the memorization, updating, and outputting of information. The input gate determines whether the current information is important to be memorized, the forgetting gate controls the updating of the content of the current memory unit, the output gate controls whether the memory content should be output under the current input condition, and the forward propagation process can be divided into four steps. First, using the current input characteristics 
And the output of the last moment +.>
To calculate the sigmoid activation value +.>
Control the last moment status information +>
Is not limited, and is not limited; second, deciding to memorize in the current state +.>
The new information in (a) is calculated by calculating the candidate value +.>
And the activation value of the input gate +.>
，/>
And->
Respectively, tanh and sigmoid; third, calculating the current state using the result of the previous step>
，/>
Representing matrix dot product operation; fourth step, calculating the sigmoid activation value of the output gate +.>
And (2) the current state->
Combining to obtain an output->
The specific formula is as follows:
in the method, in the process of the invention,
、/>
、/>
、/>
、/>
、/>
、/>
and->
As a matrix of weights, the weight matrix,
、/>
、/>
and->
For bias item->
、/>
And->
Activation values of forget gate, input gate and output gate, respectively, +.>
As a candidate value +.>
For the current state +.>
Is output. Input sample feature->
The cold load prediction result is obtained after the transformation, and the result is compared with the cold loadAnd comparing the charge reality values, and reversely updating each weight matrix and the bias term in the direction of reducing the difference between the charge reality values and the bias term.
CNN is good at processing data with grid topological structure, can effectively extract implicit characteristics in the data, LSTM network is good at processing time sequence data, CNN-LSTM network can combine the advantages of the two, CNN is used for extracting local characteristics, LSTM is used for selectively memorizing previous information to influence later output, and time sequence prediction problem can be effectively processed.
S105, adaptively selecting the first indoor cold load prediction model, the second indoor cold load prediction model and the third indoor cold load prediction model through a FRRMAB method to obtain a target indoor cold load prediction model, and taking a cold load prediction result of the target indoor cold load prediction model as a final cold load prediction result;
as shown in fig. 4, step S105 specifically includes: steps S401 to S404.
S401, calculating an adaptation rate improvement FIR at the current moment for each indoor cold load prediction model according to the following formula:
wherein m is t Indoor cold load prediction model representing t moment selection, m t ∈{XGBoost，LightGBM，CNN-LSTM}，Pred(m t T) represents the model m selected at time t t An adaptability evaluation value of the obtained cold load prediction result;
s402, setting a sliding window with the size of W so as to store an indoor cold load prediction model selected at the latest W moments and a corresponding adaptability improvement rate;
s403, giving credit values FRR to the indoor cold load prediction models according to the following formula by combining attenuation factors:
in the method, in the process of the invention,
representation model m t D represents the attenuation factor, D.epsilon.0, 1]，/>
Representation model m t Model m t Is equal to the sum of the corresponding FIR values within the sliding window, + >
Representation model m t A corresponding ranking;
s404, calculating model scores for each indoor cold load prediction model based on the credit value FRR according to the following formula:
in the formula (I), score (m) t T) represents the indoor cold load prediction model m selected at the moment t t Score at this time, FRR (m t T) represents model m t The reputation value at time t, C is the exploration factor in the FRRMAB method,
representation model m t Number of times selected in past t moments, +.>
The sum of the logarithms of the number of times the K indoor cold load prediction models are selected in t moments is represented, and K=3.
The present embodiment adopts the FRRMAB method (Fitness Rate Rank Multi-Arm Bandits Method) using an adaptive selection algorithm to select a model used at each moment in the prediction process, i.e., a target indoor cooling load prediction model. The FRRMAB method is an adaptive selection method based on a classical reinforcement learning algorithm UCB, and the method balances the exploration and utilization of the algorithm by means of an adaptive selection operator (operator). The reason for selecting FRRMAB here is two: firstly, as an algorithm in reinforcement learning, FRRMAB is very suitable for solving the problem of time sequence data prediction; secondly, for the problem of self-adaptive selection of multiple models, FRRMAB can obtain that different alternative operators and models perform better at which stage of the problem through calculation in the running process, which is also the aim of self-adaptive selection improvement, a single algorithm and a model may have optimal solution and performance on the problem in a unilateral or local way, but in order to achieve the overall expected optimal effect, the improvement of self-adaptive selection can reasonably utilize the advantages of different algorithms, and compared with the fusion of a single model and a simple model, the FRRMAB has flexibility, generalization and expandability.
The UCB method (Upper Confidence Bound) is a classical reinforcement learning method that is used to solve the problem of maximizing revenue for multi-arm slot machines. The rocker arm selection problem of multi-arm slot machines is a classical reinforcement learning problem, the essence of which is the dilemma of balance exploration and utilization (Exploration vs Exploitation, evE), where players gain by selecting to shake down one of the rocker arms. After selecting the rocker arm, the player receives a profit of 0 or 1, while the probability of the different rocker arms receiving the profit is taken from different mathematical distributions, and the player does not know the specific situation of the distribution in advance. The player's goal is to select the rocker arm by some strategy to obtain the maximum desired benefit. In this embodiment, the 3 indoor cold load prediction models are regarded as "rocker arms", the cold water host controlled by the algorithm is regarded as "multi-arm slot machine", and the expected maximum benefit is the expected error epsilon of minimizing the cold load prediction, namely, the accuracy Acc of maximizing the cold load prediction, and the calculation formula is as follows:
utilization (explicit): the ideal strategy is to select the algorithm with the smallest expected error for each cold load prediction to obtain the highest accumulated expected benefit, but the algorithm with the highest expected benefit cannot be found under normal conditions because the actual probability distribution is not known and the benefit generated by the algorithm each time accords with the random probability distribution;
Exploration (expression): the selection of different algorithms is explored randomly, and more accurate gain probability estimates for each algorithm are obtained by updating continuously according to the results of each time. But the greater the number of times the algorithm is explored means that some algorithms that have not been highly profitable before need to be selected, with some loss of revenue.
UCB concrete flow: assuming the number of algorithms is K, all algorithms are selected at least once, the objective of the traversal is to ensure that the player has at least one profit experience for each algorithm, and at time t (t > K), for a certain algorithm j (j e {1, …, K }), the scoring formula is defined as follows:
wherein T is j,t Representing the number of times algorithm j is selected in t moments,
defined as the average gain obtained in the past t moments by selecting the jth algorithm, the re, i.e. the accuracy Acc of the predicted value of the cold load, is defined as follows:
as can be seen from the analysis of the calculation formula in the UCB method, it is mainly composed of two parts and takes on different roles. Wherein, the liquid crystal display device comprises a liquid crystal display device,
the main function of the method is to give a higher score to an algorithm with more accurate average cold load prediction, and embody the property of developing and utilizing an excellent algorithm; as can be seen from the right side of the equation, when the time t passes, the smaller the number of times the algorithm j is selected, the larger the term is, namely the larger probability is selected, the meaning is that the algorithm with smaller trial selection times is encouraged, and the exploratory property is reflected. Furthermore, the denominator under the root number on the right side of the equation is always greater than 0, since it is guaranteed that each algorithm is selected at least once during the start phase.
The UCB method has the advantages that the algorithm with higher prediction accuracy is selected to obtain larger total benefits, the exploratory is also considered, and a certain opportunity is given to the algorithm with smaller selected times to explore whether better selection exists.
The FRRMAB method based on UCB method improvement is totally called an adaptive operator selection method based on fitness improvement rate ranking, and mainly comprises two flows: reputation value assignment (Credit Assignment) and operator selection (Operator Selection). The purpose of reputation value allocation is to assign different reputation values to different operators, and the reputation values are used as judgment basis in the selection of the next operator. In this embodiment, models corresponding to different cold load prediction algorithms (i.e., XGBoost model, lightGBM model, and CNN-LSTM model) may be regarded as different operators for selection by the FRRMAB.
First stage reputation value assignment: the method mainly solves two problems, namely, how to measure the performance of different models in the process of algorithm operation, and how to calculate different reputation values allocated to each model based on the performance measurement.
Fitness improvement rate: FIR (Fitness Improvement Rate) which is defined as the rate of increase of the adaptation value of the predicted value of the cold load at the present time compared with the estimated value at the previous time, i.e. 
Employing an operator/model m at time t t （m t E { XGBoost, lightGBM, CNN-LSTM }) the resulting FIR calculation formula is as follows:
wherein Pred (m t T) represents the adaptability evaluation value of the cold load predicted value obtained by the model mt selected at the time t, and the performances of different indoor cold load predicted models at different times can be quantitatively evaluated through FIR.
Sliding window: after the FIR value is calculated, the FRRMAB stores the operators/models selected at the last W times and the FIR corresponding thereto by setting a sliding window with a fixed size of WThe formula is shown in table 1 below: wherein m is t ∈{XGBoost, LightGBM, CNN-LSTM},(t=1,2,…,W)。
TABLE 1
| t | 1 | 2 | … | W |
| m t | m 1 | m 2 | … | m w |
| FIR | FIR(m 1 ,1) | FIR(m 2 ,2) | … | FIR(m w ,W) |
Reputation value calculation and distribution: in order to convert the performance metrics FIR obtained in the sliding window into reputation values for different models, the sum of the FIR values of the different models in the sliding window needs to be counted first as rewards (Reward) of the different models, namely
The models are then ordered according to Reward, model m t The corresponding rank is
. In order to get higher reputation value for the top ranked models, an attenuation factor +.>
The rewards of each model are converted into an attenuation value (Decay) with the following calculation formula:
thereafter, the ratio of the attenuation value to the sum of all attenuation values is assigned to the model as reputation value FRR (19):
Therefore, the algorithm distributes a credit value to each indoor cold load prediction model for the subsequent indoor cold load prediction model selection step, and the size of the attenuation factor D can be seen to influence the duty ratio of the performance of the model per se in the credit value, so that the exploration and utilization of the whole algorithm are further controlled.
Second stage model selection: according to the reputation value of the different indoor cold load prediction models obtained in the previous step, the FRRMAB method selects the model to be used at the next moment, the step is mainly based on the balance thought of the UCB method, the model Score (Score) is divided into two parts, the performance and the exploration degree are respectively measured, the main difference is that the calculation of the UCB on the model performance is to directly calculate the average benefit of the model, and the FRRMAB method adopts the FRR value generated in the previous step as a substitute.
At the beginning of FRRMAB, the algorithm will give each model an equal probability until each model is selected at least once, after which each model is weighted summed with the degree of exploration, the score calculated, and the model with the highest score selected as the next choice. The score calculation formula is as follows:
Here, score (mt, t) represents the Score of the indoor cold load prediction model mt selected at time t at that time. Of the two parts to the right of the equal sign, the FRR (m t T) is m t The reputation value at the time t is used for promoting selection of a cold load prediction algorithm with better performance; in the second part, the denominator part inside the root represents the model m t The number of times selected in the past t times, the molecular part is K kinds of calculationThe sum of the logarithms of the number of times a method (k=3) is selected in t times, and C outside the root can be regarded as a coefficient controlling the degree of exploration, whereby it is obtained that the smaller the number of times a certain model is selected in the past t times, the larger the value of this term, i.e., the smaller the attempt to select a cold load prediction algorithm is encouraged.
Thus, the FRRMAB performs calculation at each time, thereby adaptively selecting a function of a model suitable for the current time, and pursuing minimization of an overall expected error of the cooling load prediction, and maximization of accuracy.
S106, outputting a corresponding energy-saving strategy to the water chilling unit based on the final cold load prediction result.
Specifically, the step S106 includes:
obtaining a final cold load prediction result in a preset time threshold, and obtaining an average value to obtain an average cold load prediction result in the preset time threshold;
Setting corresponding water supply temperature for the water chilling unit by using the average cold load prediction result;
and re-acquiring a final cold load prediction result at intervals of a preset time threshold value, so as to dynamically adjust the energy-saving strategy.
For example, a central air conditioner of a building is typically turned on earlier, at which time most of the terminal fan coils are not yet turned on, and the chilled water circulation is less cold to exchange with the room, so that it can be set to a higher temperature, for example, 10 degrees, when turned on. After the starting, a certain time m (unit is minutes) is taken as an interval, for example, 60 minutes, an FRRMAB method is adopted to predict indoor cold load prediction results of m/5 time points in the future, and then the indoor cold load prediction results are averaged to obtain corresponding average cold load prediction results, and the average cold load prediction results are taken as indoor average required cold loads in the future. And then adjusting the water supply temperature of the main machine of the cold water according to the average cold load prediction result. Specifically, the specific heat capacity c of water and the density ρ of water are constant, the fluctuation of the chilled water flow G is negligible under the condition that the starting quantity of the chilled pump is determined and the variable frequency control is not performed, and the chilled water backwater temperature at the current moment is known
And the indoor average required cooling load for a period of time in the future, the water supply temperature of the chilled water of the cooling load required in the future can be calculated by a cooling load calculation formula (namely the formula of the step S303) under the condition of taking the current backwater temperature as a reference 
。
Further, in consideration of the temperature control accuracy and the temperature range of the main chiller, the calculated chilled water supply temperature can be calculated in 0.5 units when the main chiller temperature is set
Rounding down and limiting the set temperature value after rounding down to [7, 12]In the section, if the setting temperature is lower than 7 degrees after rounding down, the setting temperature is required to be 7 degrees, and if the setting temperature is higher than 12 degrees, the setting temperature is required to be 12 degrees. Of course, in other scenarios, the unit of the downward rounding and the temperature control range can be limited according to the specific situation of the cold water host, and the unit of 0.5 and the control range are not necessarily required to be [7, 12 ]]And (3) the room(s). Furthermore, the adjustment is performed at intervals, such as 1 hour, during the start-up period of the cold water main machine, specifically, the time required by the water supply temperature of the cold water main machine reaching a certain error range of the steady state value and the time condition that the historical cold load changes are referred to for determining, and the cold water main machine is automatically issued to execute after the adjustment time is determined and the corresponding temperature is set during each adjustment.
In combination with the illustration of fig. 2, the embodiment of the invention constructs an indoor cold load prediction model through an XGBoost model, a LightGBM model and a CNN-LSTM model respectively, selects an indoor cold load prediction model with the optimal current time period through a self-adaptive algorithm FRRMAB to predict and output a final cold load prediction result, and then sets an energy-saving strategy for the water chiller based on the final cold load prediction result, so that the energy consumption management efficiency of the water chiller can be improved, and the energy-saving effect is achieved while the indoor environment comfort level is ensured.
In addition, a sliding window method can be adopted for variables in constructing characteristics for the CNN-LSTM model, and the method is characterized in thatBesides a sliding window method, the method can also adopt a moving window to calculate statistics and difference mode on the basis of the sliding window, and add time characteristics such as hours, working days and the like when constructing the characteristics for XGBoost and LightGBM models. Here, the data used in constructing the features using the sliding window method is the predicted point locations
Before the point of time->
Strip observation data->
Indicates the number of bits to be predicted in future time, < >>
And the size of the sliding window. When constructing the model, model tuning can be performed according to the average absolute percentage error MAPE (Mean Absolute Percentage Error) of the evaluation index on the model data set so as to determine the sliding window size and super-parameter setting of the CNN-LSTM model, and the characteristics and the optimal super-parameters of the XGBoost model and the LightGBM model are determined so as to obtain the optimal model under each algorithm frame. The MAPE calculation is as follows:
where n is the data length of the model dataset,
is the real value of the cooling load of the data of the kth time t,
the predicted value of the cooling load is the data of the kth time t.
Fig. 5 is a schematic block diagram of a water chiller energy saving device 500 based on an adaptive algorithm according to an embodiment of the present invention, where the device 500 includes:
The data acquisition unit 501 is configured to acquire environmental data and cold station equipment data, and perform data processing on the environmental data and the cold station equipment data to obtain a model data set;
a first construction unit 502, configured to perform training learning on the model dataset by using an XGBoost model, so as to construct a first indoor cold load prediction model;
a second construction unit 503, configured to perform training learning on the model dataset by using a LightGBM model, so as to construct a second indoor cold load prediction model;
a third construction unit 504, configured to perform training learning on the model dataset by using a CNN-LSTM model, so as to construct a third indoor cooling load prediction model;
a model selecting unit 505, configured to adaptively select the first indoor cold load prediction model, the second indoor cold load prediction model, and the third indoor cold load prediction model by using a FRRMAB method, obtain a target indoor cold load prediction model, and use a result of cold load prediction of the target indoor cold load prediction model as a final cold load prediction result;
and the strategy output unit 506 is configured to output a corresponding energy saving strategy to the water chiller based on the final cooling load prediction result.
In one embodiment, as shown in fig. 6, the data acquisition unit 501 includes:
a sensor acquisition unit 601, configured to acquire environmental data and cold station equipment data through a sensor; the environment data comprise outdoor temperature, outdoor humidity, indoor temperature, indoor humidity and indoor carbon dioxide concentration, and the cold station equipment data comprise cooling tower water inlet temperature, cooling tower water outlet temperature, chilled water supply temperature, chilled water return temperature, chilled water flow and total active power of a cold water host;
the data processing unit 602 is configured to perform missing value processing and abnormal value processing on the environmental data and the cold station device data, and perform interpolation filling based on a missing value processing result and abnormal value processing;
a cooling load calculating unit 603 for calculating an indoor cooling load according to the cooling station apparatus data according to the following formula:
wherein Q represents the indoor cooling load, c represents the specific heat capacity of water
ρ represents the density of water
G represents flow, ++>
Indicating the return water temperature of chilled water->
Indicating chilled water supply temperature;
and the data set constructing unit 604 is configured to delete the data with the indoor cooling load as a tag and the cooling load of 0 when the chiller is not refrigerating, so as to construct a model data set related to the environmental data and the cold station equipment data.
In an embodiment, the first construction unit 502 includes:
the data input unit is used for inputting the model data set into each decision tree sub-model of the XGBoost model, and outputting a corresponding first cooling load prediction result by the XGBoost model;
the model optimizing unit is used for optimizing the XGBoost model by taking the gradient descending direction of the loss function as an optimizing target according to the following formula:
in the method, in the process of the invention,
representing the first cold load prediction result of the XGBoost model,/>
Representing the first cold load true value, L in Obj represents a differentiable convex loss function for measuring the first cold load pre-loadMeasurement of->
And a first cooling load true value
Error of each other>
Representing a regular term for controlling the complexity of the model to avoid overfitting, T representing the number of leaf nodes,/->
Is a coefficient of->
Is a regularized term of L2.
In an embodiment, the second construction unit 503 includes:
the sampling processing unit is used for sampling the model data set by adopting a single-side gradient sampling method, retaining a% of large gradient samples before a%, and randomly extracting b% of small gradient samples;
the binding processing unit is used for carrying out mutual exclusion feature binding processing on the large gradient sample so as to reduce the feature quantity of the large gradient sample;
And the characteristic discrete unit is used for discretizing the large gradient sample with the reduced characteristic quantity into k discrete characteristics by using a histogram algorithm, traversing the k discrete characteristics to find the optimal splitting point and the splitting threshold value, and obtaining a second cooling load prediction result corresponding to the LightGBM model.
In an embodiment, the third building unit 504 includes:
the feature extraction unit is used for inputting the data in the model data set into a convolution layer of a CNN network and extracting data features in a self-adaptive mode by utilizing a wide convolution kernel;
the pooling processing unit is used for pooling the data features by utilizing the maximum pooling layer so as to reduce the dimension of the data features;
the sequence output unit is used for inputting the data characteristics with reduced dimensionality into the LSTM layer and outputting corresponding sequence characteristics by the LSTM layer;
and the classification output unit is used for classifying and outputting the sequence characteristics through the full connection layer to obtain a third cooling load prediction result corresponding to the CNN-LSTM model.
In an embodiment, as shown in fig. 7, the model selection unit 505 includes:
an improvement rate calculation unit 701, configured to calculate an fitness improvement rate FIR at the current time for each indoor cooling load prediction model according to the following formula:
Wherein m is t Indoor cold load prediction model representing t moment selection, m t ∈{XGBoost，LightGBM，CNN-LSTM}，Pred(m t T) represents the model m selected at time t t An adaptability evaluation value of the obtained cold load prediction result;
a window storage unit 702, configured to set a sliding window with a size W, so as to store the indoor cold load prediction models selected at the last W moments and the corresponding fitness improvement rates;
a reputation value giving unit 703 for giving a reputation value FRR to each indoor cold load prediction model in combination with the attenuation factor according to the following formula:
in the method, in the process of the invention,
representation model m t D represents the attenuation factor, D.epsilon.0, 1]，/>
Representation model m t Model m t Is equal to the sum of the corresponding FIR values within the sliding window, +>
Representation modelM is as follows t A corresponding ranking;
the model score calculating unit 704 is configured to calculate a model score for each indoor cooling load prediction model based on the reputation value FRR according to the following formula:
in the formula (I), score (m) t T) represents the indoor cold load prediction model m selected at the moment t t Score at this time, FRR (m t T) represents model m t The reputation value at time t, C is the exploration factor in the FRRMAB method,
representation model m t Number of times selected in past t moments, +.>
The sum of the logarithms of the number of times the K indoor cold load prediction models are selected in t moments is represented, and K=3.
In an embodiment, the policy output unit 506 includes:
obtaining a final cold load prediction result in a preset time threshold, and obtaining an average value to obtain an average cold load prediction result in the preset time threshold;
setting corresponding water supply temperature for the water chilling unit by using the average cold load prediction result;
and re-acquiring a final cold load prediction result at intervals of a preset time threshold value, so as to dynamically adjust the energy-saving strategy.
Since the embodiments of the apparatus portion and the embodiments of the method portion correspond to each other, the embodiments of the apparatus portion are referred to the description of the embodiments of the method portion, and are not repeated herein.
The embodiment of the present invention also provides a computer readable storage medium having a computer program stored thereon, which when executed can implement the steps provided in the above embodiment. The storage medium may include: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The embodiment of the invention also provides a computer device, which can comprise a memory and a processor, wherein the memory stores a computer program, and the processor can realize the steps provided by the embodiment when calling the computer program in the memory. Of course, the computer device may also include various network interfaces, power supplies, and the like.
In the description, each embodiment is described in a progressive manner, and each embodiment is mainly described by the differences from other embodiments, so that the same similar parts among the embodiments are mutually referred. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section. It should be noted that it would be obvious to those skilled in the art that various improvements and modifications can be made to the present application without departing from the principles of the present application, and such improvements and modifications fall within the scope of the claims of the present application.
It should also be noted that in this specification, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
Claims (10)
1. The water chilling unit energy saving method based on the self-adaptive algorithm is characterized by comprising the following steps of:
collecting environment data and cold station equipment data, and performing data processing on the environment data and the cold station equipment data to obtain a model data set;
training and learning the model data set by adopting an XGBoost model to construct a first indoor cold load prediction model;
training and learning the model dataset by adopting a LightGBM model to construct a second indoor cold load prediction model;
training and learning the model data set by adopting a CNN-LSTM model to construct a third indoor cold load prediction model;
performing self-adaptive selection on the first indoor cold load prediction model, the second indoor cold load prediction model and the third indoor cold load prediction model through a FRRMAB method to obtain a target indoor cold load prediction model, and taking a cold load prediction result of the target indoor cold load prediction model as a final cold load prediction result;
and outputting a corresponding energy-saving strategy to the water chilling unit based on the final cold load prediction result.
2. The adaptive algorithm-based chiller plant energy conservation method of claim 1, wherein the collecting environmental data and cold station equipment data and performing data processing on the environmental data and cold station equipment data to obtain a model dataset comprises:
Collecting environmental data and cold station equipment data through a sensor; the environment data comprise outdoor temperature, outdoor humidity, indoor temperature, indoor humidity and indoor carbon dioxide concentration, and the cold station equipment data comprise cooling tower water inlet temperature, cooling tower water outlet temperature, chilled water supply temperature, chilled water return temperature, chilled water flow and total active power of a cold water host;
carrying out missing value processing and abnormal value processing on the environment data and the cold station equipment data, and carrying out interpolation filling based on missing value processing results and abnormal value processing;
calculating indoor cooling load according to the cooling station equipment data:
wherein Q represents the indoor cooling load, c represents the specific heat capacity of water
ρ represents the density of water
G represents flow, ++>
Indicating the return water temperature of chilled water->
Indicating chilled water supply temperature;
and taking the indoor cooling load as a tag, deleting the data with the cooling load of 0 when the water chiller is not cooled, and constructing a model data set related to the environmental data and the cold station equipment data.
3. The adaptive algorithm-based chiller plant energy conservation method of claim 1, wherein the training learning the model dataset using an XGBoost model to construct a first indoor cold load prediction model comprises:
Inputting the model dataset into each decision tree sub-model of the XGBoost model, and outputting a corresponding first cooling load prediction result by the XGBoost model;
optimizing the XGBoost model by taking the gradient descent direction of the loss function as an optimization target according to the following steps:
in the method, in the process of the invention,
representing the first cold load prediction result of the XGBoost model,/>
Representing the first cooling load true value, L in Obj represents a differentiable convex loss function for measuring the first cooling load predictor +.>
And a first cold load realism value +.>
Error of each other>
Representing a regular term for controlling the complexity of the model to avoid overfitting, T representing the number of leaf nodes,/->
As the coefficient of the light-emitting diode,
is a regularized term of L2.
4. The adaptive algorithm-based chiller plant energy conservation method of claim 1, wherein the training learning the model dataset using a LightGBM model to construct a second indoor cold load prediction model comprises:
sampling the model data set by adopting a single-side gradient sampling method, retaining a% of large gradient samples before the model data set, and randomly extracting b% of small gradient samples;
performing mutual exclusion feature binding processing on the large gradient sample to reduce the feature quantity of the large gradient sample;
And discretizing the large gradient sample with the reduced feature quantity into k discrete features by using a histogram algorithm, and traversing the k discrete features to find the optimal splitting point and the splitting threshold value so as to obtain a second cold load prediction result corresponding to the LightGBM model.
5. The adaptive algorithm-based chiller plant energy conservation method of claim 1, wherein the training learning the model dataset using a CNN-LSTM model to construct a third indoor cold load prediction model comprises:
inputting the data in the model data set into a convolution layer of a CNN network, and extracting data characteristics in a self-adaptive manner by using a wide convolution kernel;
carrying out pooling treatment on the data features by using a maximum pooling layer so as to reduce the dimension of the data features;
inputting the data characteristics with reduced dimensionality into an LSTM layer, and outputting corresponding sequence characteristics by the LSTM layer;
and classifying and outputting the sequence characteristics through the full connection layer to obtain a third cooling load prediction result corresponding to the CNN-LSTM model.
6. The adaptive algorithm-based chiller plant energy conservation method of claim 1, wherein the adaptively selecting the first indoor cold load prediction model, the second indoor cold load prediction model, and the third indoor cold load prediction model by the FRRMAB method to obtain a target indoor cold load prediction model comprises:
The fitness improvement rate FIR at the current moment is calculated for each indoor cold load prediction model according to the following steps:
wherein m is t Indoor cold load prediction model representing t moment selection, m t ∈{XGBoost，LightGBM，CNN-LSTM}，Pred(m t T) represents the model m selected at time t t An adaptability evaluation value of the obtained cold load prediction result;
setting a sliding window with the size of W so as to store the indoor cold load prediction model selected at the latest W moments and the corresponding adaptability improvement rate;
the reputation value FRR is given to each indoor cold load prediction model in combination with the attenuation factor according to the following formula:
in the method, in the process of the invention,
representation model m t D represents the attenuation factor, D.epsilon.0, 1]，/>
Representation model m t Model m t Is equal to the sum of the corresponding FIR values within the sliding window, +>
Representation model m t A corresponding ranking;
calculating model scores for each indoor cold load prediction model based on the reputation value FRR according to:
in the formula (I), score (m) t T) represents the indoor cold load prediction model m selected at the moment t t Score at this time, FRR (m t T) represents model m t The reputation value at time t, C is the exploration factor in the FRRMAB method,
representation model m t Number of times selected in past t moments, +.>
The sum of the logarithms of the number of times the K indoor cold load prediction models are selected in t moments is represented, and K=3.
7. The method for saving energy of a chiller based on an adaptive algorithm according to claim 1, wherein the outputting of a corresponding energy saving strategy to the chiller based on the final cooling load prediction result comprises:
obtaining a final cold load prediction result in a preset time threshold, and obtaining an average value to obtain an average cold load prediction result in the preset time threshold;
setting corresponding water supply temperature for the water chilling unit by using the average cold load prediction result;
and re-acquiring a final cold load prediction result at intervals of a preset time threshold value, so as to dynamically adjust the energy-saving strategy.
8. The utility model provides a chiller economizer based on self-adaptation algorithm which characterized in that includes:
the data acquisition unit is used for acquiring environment data and cold station equipment data, and carrying out data processing on the environment data and the cold station equipment data to obtain a model data set;
the first construction unit is used for training and learning the model data set by adopting an XGBoost model so as to construct a first indoor cold load prediction model;
the second construction unit is used for training and learning the model data set by adopting a LightGBM model so as to construct a second indoor cold load prediction model;
The third construction unit is used for training and learning the model data set by adopting a CNN-LSTM model so as to construct a third indoor cold load prediction model;
the model selection unit is used for adaptively selecting the first indoor cold load prediction model, the second indoor cold load prediction model and the third indoor cold load prediction model through a FRRMAB method to obtain a target indoor cold load prediction model, and taking the result of cold load prediction of the target indoor cold load prediction model as a final cold load prediction result;
and the strategy output unit is used for outputting a corresponding energy-saving strategy to the water chilling unit based on the final cold load prediction result.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the adaptive algorithm-based chiller plant energy conservation method of any one of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program, which when executed by a processor, implements the adaptive algorithm-based water chiller energy conservation method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310496732.3A CN116227741A (en) | 2023-05-05 | 2023-05-05 | Water chilling unit energy saving method and device based on self-adaptive algorithm and related medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310496732.3A CN116227741A (en) | 2023-05-05 | 2023-05-05 | Water chilling unit energy saving method and device based on self-adaptive algorithm and related medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116227741A true CN116227741A (en) | 2023-06-06 |
Family
ID=86573488
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310496732.3A Pending CN116227741A (en) | 2023-05-05 | 2023-05-05 | Water chilling unit energy saving method and device based on self-adaptive algorithm and related medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116227741A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104008430A (en) * | 2014-05-29 | 2014-08-27 | 华北电力大学 | Method for establishing virtual reality excavation dynamic smart load prediction models |
| CN107036232A (en) * | 2017-03-07 | 2017-08-11 | 深圳达实智能股份有限公司 | A kind of hospital's water cool-storage technology system cold operational energy efficiency optimal control method and device |
| CN110472778A (en) * | 2019-07-29 | 2019-11-19 | 上海电力大学 | A kind of short-term load forecasting method based on Blending integrated study |
| CN111476422A (en) * | 2020-04-10 | 2020-07-31 | 北京石油化工学院 | L ightGBM building cold load prediction method based on machine learning framework |
| CN111932015A (en) * | 2020-08-12 | 2020-11-13 | 武汉中电节能有限公司 | Method and system for predicting cold and heat load of regional cooling and heating |
| CN113762387A (en) * | 2021-09-08 | 2021-12-07 | 东北大学 | Data center station multi-load prediction method based on hybrid model prediction |
| CN114169434A (en) * | 2021-12-07 | 2022-03-11 | 国电南瑞科技股份有限公司 | Load prediction method |
| CN114857743A (en) * | 2022-01-25 | 2022-08-05 | 西安建筑科技大学 | Terminal valve optimization control method and system based on market partition load prediction |
| CN115130741A (en) * | 2022-06-20 | 2022-09-30 | 北京工业大学 | Multi-model fusion based multi-factor power demand medium and short term prediction method |
-
2023
- 2023-05-05 CN CN202310496732.3A patent/CN116227741A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104008430A (en) * | 2014-05-29 | 2014-08-27 | 华北电力大学 | Method for establishing virtual reality excavation dynamic smart load prediction models |
| CN107036232A (en) * | 2017-03-07 | 2017-08-11 | 深圳达实智能股份有限公司 | A kind of hospital's water cool-storage technology system cold operational energy efficiency optimal control method and device |
| CN110472778A (en) * | 2019-07-29 | 2019-11-19 | 上海电力大学 | A kind of short-term load forecasting method based on Blending integrated study |
| CN111476422A (en) * | 2020-04-10 | 2020-07-31 | 北京石油化工学院 | L ightGBM building cold load prediction method based on machine learning framework |
| CN111932015A (en) * | 2020-08-12 | 2020-11-13 | 武汉中电节能有限公司 | Method and system for predicting cold and heat load of regional cooling and heating |
| CN113762387A (en) * | 2021-09-08 | 2021-12-07 | 东北大学 | Data center station multi-load prediction method based on hybrid model prediction |
| CN114169434A (en) * | 2021-12-07 | 2022-03-11 | 国电南瑞科技股份有限公司 | Load prediction method |
| CN114857743A (en) * | 2022-01-25 | 2022-08-05 | 西安建筑科技大学 | Terminal valve optimization control method and system based on market partition load prediction |
| CN115130741A (en) * | 2022-06-20 | 2022-09-30 | 北京工业大学 | Multi-model fusion based multi-factor power demand medium and short term prediction method |
Non-Patent Citations (3)
| Title |
|---|
| KE LI等: "Adaptive Operator Selection With Bandits for a Multiobjective Evolutionary Algorithm Based on Decomposition", 《IEEE》, pages 114 - 130 * |
| 徐湘君: "基于非平衡数据集的***欺诈检测研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》, no. 3, pages 140 - 583 * |
| 边策: "某航站楼候机厅冷负荷预测与空调末端节能策略研究", 《中国优秀硕士学位论文全文数据库 工程科技辑》, no. 2, pages 031 - 1263 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111795484B (en) | Intelligent air conditioner control method and system | |
| CN107704875A (en) | Based on the building load Forecasting Methodology and device for improving IHCMAC neutral nets | |
| CN111735178A (en) | Air conditioner energy-saving system based on elite meaning and SVR regression algorithm and optimization method | |
| CN112415924A (en) | Energy-saving optimization method and system for air conditioning system | |
| CN113112077B (en) | HVAC control system based on multi-step prediction deep reinforcement learning algorithm | |
| CN111580382B (en) | Unit-level heat supply adjusting method and system based on artificial intelligence | |
| CN110781595B (en) | Method, device, terminal and medium for predicting energy use efficiency (PUE) | |
| CN112527037B (en) | Greenhouse environment regulation and control method and system with environment factor prediction function | |
| CN113762387B (en) | Multi-element load prediction method for data center station based on hybrid model prediction | |
| CN106682764A (en) | Method for predicting other day air-conditioning load of public building based on parallel prediction strategy | |
| CN112926795A (en) | SBO (statistical analysis) -based CNN (continuous casting) optimization-based high-rise residential building group heat load prediction method and system | |
| CN109919356A (en) | One kind being based on BP neural network section water demand prediction method | |
| CN114362175B (en) | Wind power prediction method and system based on depth certainty strategy gradient algorithm | |
| CN116562514B (en) | Method and system for immediately analyzing production conditions of enterprises based on neural network | |
| CN110519816A (en) | A kind of radio roaming control method, device, storage medium and terminal device | |
| CN114580762A (en) | Hydrological forecast error correction method based on XGboost | |
| CN114353872A (en) | Prediction method and device for machine room temperature | |
| Fu et al. | A Sarsa-based adaptive controller for building energy conservation | |
| CN116227741A (en) | Water chilling unit energy saving method and device based on self-adaptive algorithm and related medium | |
| CN114065646B (en) | Energy consumption prediction method based on hybrid optimization algorithm, cloud computing platform and system | |
| CN115796364A (en) | Intelligent interactive decision-making method for discrete manufacturing system | |
| CN114909707A (en) | Heat supply secondary network regulation and control method based on intelligent balancing device and reinforcement learning | |
| CN113177675A (en) | Air conditioner cold load prediction method based on optimization neural network of longicorn group algorithm | |
| CN112529183A (en) | Knowledge distillation-based model self-adaptive updating method | |
| CN114444737A (en) | Intelligent pavement maintenance planning method based on transfer 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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230606 |