CN113887801A - Building cold load prediction method, system, equipment and readable storage medium - Google Patents

Building cold load prediction method, system, equipment and readable storage medium Download PDF

Info

Publication number
CN113887801A
CN113887801A CN202111156381.9A CN202111156381A CN113887801A CN 113887801 A CN113887801 A CN 113887801A CN 202111156381 A CN202111156381 A CN 202111156381A CN 113887801 A CN113887801 A CN 113887801A
Authority
CN
China
Prior art keywords
neural network
grnn
cold load
prediction
building
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
Application number
CN202111156381.9A
Other languages
Chinese (zh)
Inventor
于军琪
李蕴
赵安军
周敏
高之坤
杨思远
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.)
Xian University of Architecture and Technology
Original Assignee
Xian University of Architecture and Technology
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 Xian University of Architecture and Technology filed Critical Xian University of Architecture and Technology
Priority to CN202111156381.9A priority Critical patent/CN113887801A/en
Publication of CN113887801A publication Critical patent/CN113887801A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • 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
    • 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)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Human Resources & Organizations (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Primary Health Care (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a method, a system, equipment and a medium for predicting building cold load, wherein the method comprises the following steps: acquiring data of optimal influence constituent elements of the cold load of the historical building; respectively training a GRNN neural network and an LSTM neural network, and optimizing key factors of the GRNN neural network by adopting a BSA algorithm; determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model, and constructing to obtain a BSA-GRNN & LSTM combined neural network prediction model; using the data of the optimal influence component elements of the building cold load to be predicted as the input of a BSA-GRNN & LSTM combined neural network prediction model, and outputting to obtain the building cold load prediction result; the method effectively solves the problems of low prediction precision and poor generalization performance of a single model, and effectively improves the precision of the building cold load prediction result.

Description

Building cold load prediction method, system, equipment and readable storage medium
Technical Field
The invention belongs to the technical field of building energy consumption prediction, and particularly relates to a building cold load prediction method, a system, equipment and a readable storage medium.
Background
In recent years, with the acceleration of urbanization, people have higher requirements for quality of life, and central air conditioning systems are widely used in modern buildings in order to provide comfortable indoor thermal environments. The operation energy consumption of the air conditioning system accounts for about 40 percent of the total energy consumption of the building, and the air conditioning system is the largest single energy consumption system in the building. The air conditioning system energy conservation plays an important role in building operation energy efficiency, and accurate building cold load prediction is the key for improving the air conditioning operation energy efficiency, so the building cold load prediction has important significance in energy conservation and emission reduction. The accurate building cold load prediction is the key for improving the building energy efficiency and implementing the building energy conservation, so the research on the mechanism and the rule of the building cold load prediction and the establishment of an accurate and effective building cold load prediction model not only have very important practical significance, but also are a plurality of energy management tasks, such as: fault detection and diagnosis strategies, demand side management and optimal control lay the necessary foundation.
At present, a prediction model based on an artificial neural network can be effectively applied to the prediction of the cold load of a building, and the actual application requirement can be met. However, the input variables of the existing prediction model are not comprehensive in type, and most of the input variables are outdoor meteorological parameters; the building cold load is also influenced by the characteristics of the building body, indoor variables and the like, which also need to be analyzed and considered; and the building energy consumption data has the characteristics of nonlinearity and strong volatility, so that the prediction precision obtained by using a single prediction model is low, the generalization capability is poor, the method cannot be well suitable for the characteristics of nonlinearity and strong volatility of the building energy consumption data, and the prediction precision obtained by using the single prediction model is low, the generalization capability is poor, and the method cannot be well suitable for building cold load prediction.
Disclosure of Invention
Aiming at the technical problems in the prior art, the invention provides a building cold load prediction method, a system, equipment and a readable storage medium, which are used for solving the technical problems of low prediction precision and poor generalization capability when a single prediction model is used in the existing building cold load prediction process due to the fact that building energy consumption data has the characteristics of nonlinearity and strong volatility.
In order to achieve the purpose, the invention adopts the technical scheme that:
the invention provides a building cold load prediction method, which comprises the following steps:
acquiring data of optimal influence constituent elements of the historical building cold load, and constructing a training subset and a testing subset;
training the GRNN neural network by using the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA (bovine serum albumin) algorithm to obtain an optimized GRNN neural network model;
training the LSTM neural network by using the training subset to obtain a trained LSTM neural network model;
respectively predicting the test subsets by using the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment;
determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result at each time, and constructing to obtain a BSA-GRNN & LSTM combined neural network prediction model;
and taking the data of the optimal influence component elements of the building cold load to be predicted as the input of the BSA-GRNN & LSTM combined neural network prediction model, and outputting to obtain the building cold load prediction result.
Further, the process of acquiring the data of the optimal influence component elements of the historical building cold load and constructing a training subset and a testing subset is specifically as follows:
acquiring influence component data of the historical building cold load;
performing feature extraction on the influence component data of the historical building cold load by adopting a random forest algorithm to obtain main influence component data of the historical building cold load;
processing the main influence component data of the historical building cold load by adopting a recursive characteristic elimination method to obtain the optimal influence component data of the historical building cold load;
and constructing a training subset and a testing subset by adopting a random extraction mode on the optimal influence component data of the historical building cold load.
Further, the optimal influence component data of the historical building cold load comprises east window-wall ratio, south window-wall ratio, building system coefficient, room personnel, outdoor relative humidity, outdoor dry bulb temperature, illumination use condition, solar irradiance at the previous moment and load at the previous moment.
Further, training the GRNN neural network by utilizing the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA (bovine serum albumin) algorithm to obtain an optimized GRNN neural network model, wherein the key factors of the GRNN neural network are smoothing factors of the GRNN neural network; wherein, the smoothing factor of the GRNN neural network is set to be a random number in [0,1 ].
Further, a BSA algorithm is adopted to optimize key factors of the GRNN neural network to obtain an optimized GRNN neural network model, and the average absolute error of the GRNN neural network model to a training set is used as a fitness function of BSA;
wherein, the expression of the fitness function of the BSA is as follows:
Figure BDA0003288500010000031
wherein, the fitness is a fitness function of the BSA; y isiOutput values of the ith sample in the training subset are obtained for the GRNN neural network model;
Figure BDA0003288500010000032
the real value of the ith sample in the training subset is obtained; n is a radical oftestFor trainingNumber of samples of the set.
Further, the optimized GRNN neural network model comprises an input layer, a mode layer, a summation layer and an output layer; wherein the number of neurons in the input layer is the same as the input vector dimension of the training subset; the neurons of the input layer directly transfer the input vectors of the training subset to the pattern layer;
the expression of the transfer function of the neurons of the pattern layer is:
Figure BDA0003288500010000033
wherein, PiA transfer function of neurons that are a mode layer; x is an input vector of the training subset; xiAn ith sample corresponding to an ith neuron in the mode layer; σ is a key factor of the GRNN neural network.
The summation layer includes a first neuron and a second neuron; a first neuron for arithmetically summing the outputs of all neurons in a mode layer; the second neuron is used for carrying out weighted summation on the outputs of all neurons in the mode layer;
the expression of the output layer is:
Figure BDA0003288500010000041
wherein Y is the output value of the output layer; sNAn output of a first neuron that is a summation layer; sDIs the output of the second neuron of the summation layer.
The invention also provides a building cold load prediction system, which comprises:
the data acquisition module is used for acquiring the data of the optimal influence component elements of the cold load of the historical building, and constructing a training subset and a testing subset;
the GRNN neural network module is used for training the GRNN neural network by utilizing the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA algorithm to obtain an optimized GRNN neural network model;
the LSTM neural network model is used for training the LSTM neural network by utilizing the training subset to obtain a trained LSTM neural network model;
the average relative error module is used for predicting the test subsets by respectively utilizing the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment;
the combined model module is used for determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result at each time, and constructing a BSA-GRNN & LSTM combined neural network prediction model;
and the prediction module is used for taking the data of the optimal influence component elements of the building cold load to be predicted as the input of the BSA-GRNN & LSTM combined neural network prediction model and outputting to obtain the building cold load prediction result.
Furthermore, the data acquisition module comprises a constituent element data module, a feature extraction module, a recursive feature elimination module and a subset module;
the component data module is used for acquiring influence component data of the historical building cold load;
the characteristic extraction module is used for extracting the characteristics of the influence component data of the historical building cold load by adopting a random forest algorithm to obtain the main influence component data of the historical building cold load;
the recursive characteristic elimination module is used for processing the main influence component element data of the historical building cold load by adopting a recursive characteristic elimination method to obtain the optimal influence component element data of the historical building cold load;
and the subset module is used for constructing a training subset and a testing subset by adopting a random extraction mode on the data of the constituent elements with the optimal influence of the cold load of the historical building.
The present invention also provides a building cold load prediction device, including:
a memory for storing a computer program;
a processor for implementing the steps of the building cold load prediction method when executing the computer program.
The invention also provides a computer-readable storage medium, which stores a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method for predicting the building cold load.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a building cold load prediction method, which comprises the steps of determining the prediction application time of an optimized GRNN neural network model and a trained LSTM neural network model according to the average relative error of a prediction result at each time, constructing a BSA-GRNN & LSTM combined neural network prediction model, and predicting the building cold load by adopting the BSA-GRNN & LSTM combined neural network prediction model; the LSTM neural network model can be suitable for processing dynamic and long-interval data, has strong nonlinear mapping capability, organically combines the GRNN neural network model optimized by the BSA algorithm with the LSTM neural network model, fully utilizes the advantages of the two neural network models, effectively overcomes the problems of low prediction precision and poor generalization performance of a single model, and effectively improves the precision of a building cold load prediction result.
Furthermore, a random forest algorithm is adopted to perform feature extraction on the influence component data of the historical building cold load, and influence coefficients of the influence components of the building cold load on the cold load are determined, the random forest algorithm can effectively analyze data with nonlinearity and colinearity, so that features with large influence on the cold load can be better extracted, and the accuracy of a prediction model is improved; the optimal influence component data of the historical building cold load is determined by adopting a recursive characteristic elimination method, the dimensionality of the input variables of the BSA-GRNN & LSTM combined neural network prediction model is reduced by the recursive characteristic elimination method, the convergence speed is improved, and the operation cost is saved.
Furthermore, the smoothing factor of the GRNN neural network model is optimized by adopting a BAS algorithm, the average absolute error of the GRNN neural network model to the building cold load training set is set as a fitness function of the BAS, the problem that the prediction accuracy is not high due to subjectivity and randomness of the selection of the smoothing factor of the GRNN neural network model is solved, the BAS algorithm and the GRNN neural network model are organically combined, the building cold load prediction error of the GRNN neural network model is reduced, and the prediction performance is effectively improved.
Drawings
FIG. 1 is a graph of influence coefficients of the elements affecting the cooling load of a building in an embodiment;
FIG. 2 is a diagram of an optimal feature set-up for the BAS-GRNN & LSTM neural network model in an embodiment;
FIG. 3 is a flow chart of modeling of the BAS-GRNN & LSTM neural network model in an embodiment;
FIG. 4 is a diagram of a generalized recurrent neural network in an embodiment;
FIG. 5 is a diagram of an embodiment of a long term memory network;
FIG. 6 is a diagram showing the modeling results of the BAS-GRNN & LSTM neural network model in the example;
FIG. 7 is a comparison graph of the predicted results of four different models in the example;
FIG. 8 is a graph of regression fit of true and predicted values for an example;
FIG. 9 is an iterative comparison of three different algorithms in the example;
the generalized capability demonstration chart of the building cold load prediction method in the embodiment of fig. 10.
Detailed Description
In order to make the technical problems, technical solutions and advantageous effects of the present invention more apparent, the following embodiments further describe the present invention in detail. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention provides a building cold load prediction method, which comprises the following steps:
step 1, obtaining data of the optimal influence component elements of the historical building cold load, and constructing a training subset and a testing subset.
The process of acquiring the data of the optimal influence component elements of the cold load of the historical building specifically comprises the following steps:
step 11, obtaining influence component data of the historical building cold load; wherein, the influence component data of the historical building cold load comprises: building area, building story height, ratio of each facing window to wall, building form factor, outdoor weather, wind speed, room personnel, outdoor relative humidity, outdoor dry bulb temperature, lighting usage, solar irradiance at a previous time, and load at a previous time.
Step 12, performing feature extraction on the influence component data of the historical building cold load by adopting a random forest algorithm to obtain main influence component data of the historical building cold load; the main influence component element data of the historical building cold load comprise window-wall ratio of each direction, building object shape coefficient, outdoor weather, wind speed, room personnel, outdoor relative humidity, outdoor dry bulb temperature, lighting use condition, solar irradiance at the previous moment and cold load at the previous moment.
Step 13, processing the main influence component data of the historical building cold load by adopting a recursive characteristic elimination method to obtain the optimal influence component data of the historical building cold load; the optimal influence component data of the historical building cold load is used for constructing a training subset and a testing subset; the optimal influence component element data of the historical building cold load comprise east window-wall ratio, south window-wall ratio, building system coefficient, room personnel, outdoor relative humidity, outdoor dry bulb temperature, illumination use condition, solar irradiance at the previous moment and load at the previous moment.
Step 2, training the GRNN neural network by using the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA (bovine serum albumin) algorithm to obtain an optimized GRNN neural network model; wherein, the key factor of the GRNN neural network is a smoothing factor of the GRNN neural network; wherein, the smoothing factor of the GRNN neural network is set as a random number in [0,1 ];
in the invention, the average absolute error of a GRNN neural network model to a training set is used as a fitness function of a BSA algorithm; the expression of the fitness function of the BSA algorithm is as follows:
Figure BDA0003288500010000081
wherein, the fitness is a fitness function of the BSA; y isiOutput values of the ith sample in the training subset are obtained for the GRNN neural network model;
Figure BDA0003288500010000082
the real value of the ith sample in the training subset is obtained; n is a radical oftestIs the number of samples of the training subset.
The optimized GRNN neural network model comprises an input layer, a mode layer, a summation layer and an output layer; wherein the number of neurons in the input layer is the same as the input vector dimension of the training subset; the neurons of the input layer directly transfer the input vectors of the training subset to the pattern layer;
the expression of the transfer function of the neurons of the pattern layer is:
Figure BDA0003288500010000083
wherein, PiA transfer function of neurons that are a mode layer; x is an input vector of the training subset; xiAn ith sample corresponding to an ith neuron in the mode layer; σ is a key factor of the GRNN neural network.
The summation layer includes a first neuron and a second neuron; a first neuron for arithmetically summing the outputs of all neurons in a mode layer; the second neuron is used for carrying out weighted summation on the outputs of all neurons in the mode layer;
the expression of the output layer is:
Figure BDA0003288500010000084
wherein Y is the output value of the output layer; sNAn output of a first neuron that is a summation layer; sDIs the output of the second neuron of the summation layer.
And 3, training the LSTM neural network by using the training subset to obtain a trained LSTM neural network model.
And 4, predicting the test subsets by respectively using the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment.
And step 5, determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result at each time, and constructing to obtain the BSA-GRNN & LSTM combined neural network prediction model.
And 6, taking the data of the optimal influence component of the building cold load to be predicted as the input of a BSA-GRNN & LSTM combined neural network prediction model, and outputting to obtain the building cold load prediction result.
The invention also provides a building cold load prediction system which comprises a data acquisition module, a GRNN neural network module, an LSTM neural network model, an average relative error module, a combined model module and a prediction module.
And the data acquisition module is used for acquiring the data of the optimal influence component elements of the cold load of the historical building, and constructing a training subset and a testing subset.
And the GRNN neural network module is used for training the GRNN neural network by utilizing the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA algorithm to obtain an optimized GRNN neural network model.
And the LSTM neural network model is used for training the LSTM neural network by utilizing the training subset to obtain the trained LSTM neural network model.
And the average relative error module is used for predicting the test subsets by respectively utilizing the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment.
And the combined model module is used for determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result at each time, and constructing to obtain the BSA-GRNN & LSTM combined neural network prediction model.
And the prediction module is used for taking the data of the optimal influence component elements of the building cold load to be predicted as the input of the BSA-GRNN & LSTM combined neural network prediction model and outputting to obtain the building cold load prediction result.
In the invention, the data acquisition module comprises a component data module, a feature extraction module, a recursive feature elimination module and a subset module.
And the component data module is used for acquiring the influence component data of the historical building cold load.
And the characteristic extraction module is used for extracting the characteristics of the influence component data of the historical building cold load by adopting a random forest algorithm to obtain the main influence component data of the historical building cold load.
And the recursive characteristic elimination module is used for processing the main influence component element data of the historical building cold load by adopting a recursive characteristic elimination method to obtain the optimal influence component element data of the historical building cold load.
And the subset module is used for constructing a training subset and a testing subset by adopting a random extraction mode on the data of the constituent elements with the optimal influence of the cold load of the historical building.
The invention also provides a building cold load prediction device, which comprises a memory and a processor; a memory for storing a computer program; a processor for implementing the steps of the building cold load prediction method when executing the computer program.
When the processor executes the computer program, the steps of the building cold load prediction method are realized, for example:
acquiring data of optimal influence constituent elements of the historical building cold load, and constructing a training subset and a testing subset; training the GRNN neural network by using the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA (bovine serum albumin) algorithm to obtain an optimized GRNN neural network model; training the LSTM neural network by using the training subset to obtain a trained LSTM neural network model; respectively predicting the test subsets by using the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment; determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result at each time, and constructing to obtain a BSA-GRNN & LSTM combined neural network prediction model; and taking the data of the optimal influence component elements of the building cold load to be predicted as the input of the BSA-GRNN & LSTM combined neural network prediction model, and outputting to obtain the building cold load prediction result.
Alternatively, the processor implements the functions of the modules in the system when executing the computer program, for example: and the data acquisition module is used for acquiring the data of the optimal influence component elements of the cold load of the historical building, and constructing a training subset and a testing subset. And the GRNN neural network module is used for training the GRNN neural network by utilizing the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA algorithm to obtain an optimized GRNN neural network model. And the LSTM neural network model is used for training the LSTM neural network by utilizing the training subset to obtain the trained LSTM neural network model. And the average relative error module is used for predicting the test subsets by respectively utilizing the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment. And the combined model module is used for determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result at each time, and constructing to obtain the BSA-GRNN & LSTM combined neural network prediction model. And the prediction module is used for taking the data of the optimal influence component elements of the building cold load to be predicted as the input of the BSA-GRNN & LSTM combined neural network prediction model and outputting to obtain the building cold load prediction result.
Illustratively, the computer program may be partitioned into one or more modules/units that are stored in the memory and executed by the processor to implement the invention. The one or more modules/units may be a series of computer program instruction segments capable of performing preset functions, the instruction segments describing the execution process of the computer program in the building cold load prediction device. For example, the computer program may be divided into: the device comprises a data acquisition module, a GRNN neural network module, an LSTM neural network model, an average relative error module, a combined model module and a prediction module; the specific functions of each module are as follows:
and the data acquisition module is used for acquiring the data of the optimal influence component elements of the cold load of the historical building, and constructing a training subset and a testing subset. And the GRNN neural network module is used for training the GRNN neural network by utilizing the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA algorithm to obtain an optimized GRNN neural network model. And the LSTM neural network model is used for training the LSTM neural network by utilizing the training subset to obtain the trained LSTM neural network model. And the average relative error module is used for predicting the test subsets by respectively utilizing the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment. And the combined model module is used for determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result at each time, and constructing to obtain the BSA-GRNN & LSTM combined neural network prediction model. And the prediction module is used for taking the data of the optimal influence component elements of the building cold load to be predicted as the input of the BSA-GRNN & LSTM combined neural network prediction model and outputting to obtain the building cold load prediction result.
The building cold load prediction device can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing devices. The building cooling load prediction device may include, but is not limited to, a processor, a memory. It will be understood by those skilled in the art that the foregoing is merely an example of a building cold load prediction device, and does not constitute a limitation of the building cold load prediction device, and may include more components, or some components in combination, or different components, for example, the building cold load prediction device may further include an input-output device, a network access device, a bus, etc.
The processor may be a Central Processing Unit (CPU), other general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general purpose processor may be a microprocessor or the processor may be any conventional processor or the like, the processor being the control center of the building cooling load prediction device, with various interfaces and lines connecting the various parts of the overall building cooling load prediction device.
The memory may be used to store the computer programs and/or modules, and the processor may implement the various functions of the building cooling load prediction device by running or executing the computer programs and/or modules stored in the memory and invoking data stored in the memory.
The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, a phonebook, etc.) created according to the use of the cellular phone, and the like. In addition, the memory may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid state storage device.
The invention also provides a computer-readable storage medium, in which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the method for predicting the cold load of a building.
The building cooling load prediction system integrated module/unit, if implemented in the form of a software functional unit and sold or used as a separate product, may be stored in a computer readable storage medium.
Based on such understanding, the present invention realizes all or part of the processes of the building cooling load prediction method, and can also be implemented by a computer program to instruct related hardware, where the computer program can be stored in a computer readable storage medium, and when the computer program is executed by a processor, the steps of the building cooling load prediction method can be realized. Wherein the computer program comprises computer program code, which may be in source code form, object code form, executable file or preset intermediate form, etc.
The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer memory, Read-only memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution medium, etc.
It should be noted that the computer readable storage medium may contain content that is subject to appropriate increase or decrease as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, computer readable storage media that does not include electrical carrier signals and telecommunications signals in accordance with legislation and patent practice.
Examples
The embodiment provides a method for predicting the cold load of a building by taking a certain large commercial building as a research object, which specifically comprises the following steps:
step 1, obtaining influence component data of historical building cold load; the influence component element data of the historical building cold load comprise building area, building layer height, each window wall facing ratio, building object shape coefficient, outdoor weather, wind speed, room personnel, outdoor relative humidity, outdoor dry bulb temperature, lighting use condition, solar irradiance at the previous moment and cold load at the previous moment.
And 2, performing feature extraction on the influence component data of the historical building cold load by adopting a random forest algorithm, namely an RF algorithm, to obtain main influence component data of the historical building cold load, namely main influence index data of the building cold load.
The Random Forest (RF) is used as a supervised machine learning algorithm, has low requirement on a data set, runs stably, does not have the problems of overfitting and collinearity, and can effectively analyze data with nonlinearity and collinearity, thereby well extracting characteristic elements with large influence on building cold load.
The method adopts a random forest algorithm to perform a characteristic extraction process on the influence component data of the cold load of the historical building, and specifically comprises the following steps:
step 21, carrying out data standardization processing on the influence component data of the historical building cold load so as to eliminate the influence of different component data dimensions; the formula of the data standardization processing is as follows:
Figure BDA0003288500010000141
wherein z is*The influence component data after the standardization processing; z is the mean of the original impact constituent data; z is a radical ofmeanConstructing element data for the original influence; z is a radical ofstdIs the standard deviation of the original impact component data.
Step 22, randomly extracting n sample sets from the standardized historical building cold load influence component data set by adopting a bootstrap method; wherein, m samples in each sample set, that is, m normalized influence component data in each sample set; samples not drawn are off-bag data.
Step 23, performing model training on the n sample sets to generate n decision tree models;
step 24, selecting one sample with the highest splitting capability from m samples to perform node splitting on the characteristic tree type of the single decision tree model by using the minimum variance criterion; the node splitting process is carried out according to the following formula:
Figure BDA0003288500010000151
wherein I is an optimal split sample; xsIs the value of the sample; xs' is the average of the samples; s is the embedded sample dimension.
Step 24, integrating each decision tree into a random forest after maximum growth without pruning, and further calculating the influence of the data outside the bag on the decision tree model by using a Mean Square Error (MSE); wherein, the expression of mean square error MSE is:
Figure BDA0003288500010000152
wherein MSE is the mean square error; y isiThe real value of the ith group of samples; y isi' is a predicted value of the ith group of samples; k is oob sample size.
And step 26, voting by all decision tree models, representing the importance of input variables by the MSE reduction of the mean square error, and obtaining a mean square residual sequence by using the data outside the bag.
Wherein, the expression of the mean square residual sequence is:
[MSE1,MSE2,…,MSEn]
for the input variables, the influence coefficients are as follows:
Figure BDA0003288500010000153
wherein, the Score is an influence coefficient of an input variable; MSEjThe mean square residual error of the ith group of samples; MSEnjMake a decision for the nthThe mean square residual of the tree; sEThe standard error of n decision trees.
In the embodiment, in the process of selecting the features, a 5-fold cross validation model training mode is applied; decomposing a historical building cold load influence component data set into 5 parts of sub-sample sets; wherein, training the decision tree model by taking 4 parts of the training set and 1 part of the test set alternately; and finally, averaging the output (influence coefficient) of 5 times of training to obtain a parameter influence coefficient, so that the stability of the model can be improved, and the reliability of a sample selection result is increased.
As shown in fig. 1, in this embodiment, a random forest algorithm is adopted to perform feature extraction on the influence component data of the historical building cold load, so as to obtain the main influence component data of the historical building cold load; the main influence component element data of the historical building cold load comprise window-wall ratio of each direction, building object shape coefficient, outdoor weather, wind speed, room personnel, outdoor relative humidity, outdoor dry bulb temperature, lighting use condition, solar irradiance at the previous moment and cold load at the previous moment.
Step 3, processing the main influence component data of the historical building cold load by adopting a recursive characteristic elimination method to obtain the optimal influence component data of the historical building cold load; the optimal influence component data of the historical building cold load is used for constructing a training subset and a testing subset;
the recursive feature elimination method (RFE) mainly comprises the steps of repeatedly establishing a model, sorting according to feature scores, selecting features with the highest scores, repeating the process on the remaining features, traversing all the features and then stopping to obtain the optimal feature number; as shown in fig. 2, in this embodiment, a result of selecting the optimal feature quantity for the data of the main influence constituent elements of the historical building cold load by using the recursive feature elimination method is adopted, and the selected optimal feature quantity is 9; the optimal influence component element data of the historical building cold load comprise east window-wall ratio, south window-wall ratio, building system coefficient, room personnel, outdoor relative humidity, outdoor dry bulb temperature, illumination use condition, solar irradiance at the previous moment and load at the previous moment.
Step 4, establishing a hybrid prediction model of a generalized regression neural network optimized by a longicorn stigma search algorithm and a long-term memory neural network, namely BAS-GRNN & LSTM combined neural network prediction; specifically, the testing subsets are predicted by utilizing a general regression neural network BAS-GRNN and a long-term memory neural network LSTM model optimized by a Tianniu must search algorithm, the prediction results of all components are compared to obtain a final prediction value, modeling of the BAS-GRNN and LSTM is completed, prediction performances of different models are fully exerted, and therefore the purpose of improving prediction accuracy is achieved.
As shown in FIG. 3, the specific process of establishing the BAS-GRNN & LSTM neural network prediction model is as follows:
step 41, establishing a generalized regression neural network (BAS-GRNN) prediction model optimized by a longicorn stigma search algorithm; firstly, splitting a building cold load training set subjected to Random Forest (RF) and Recursive Feature Elimination (RFE) feature extraction into a training subset and a testing subset, and training and optimizing a GRNN neural network by using the training set; in the training optimization process, a smoothing factor of a Generalized Regression Neural Network (GRNN) algorithm is optimized by adopting a Tianniu whisker search algorithm (BAS) to obtain an optimized GRNN neural network model.
The optimization process is carried out by adopting a Tianniu whisker search algorithm (BAS), and the method specifically comprises the following steps:
step 411, setting a prediction precision parameter of the GRNN neural network model, and determining a prediction precision parameter range of the GRNN neural network model; the prediction precision parameters of the GRNN neural network model comprise smoothing factors; in this embodiment, the value range of the smoothing factor is set to be a random number within [0,1 ].
Step 412, setting an optimization range of the BSA algorithm influence parameters according to the prediction precision parameter range of the GRNN neural network model; the influence parameters of the BSA algorithm are mainly fitness functions, the average absolute error of the GRNN neural network model to the training subset is set as the fitness function of the BAS algorithm, the smaller the fitness function value of the BAS algorithm is, the more accurate the predicted value is, and the more excellent the smoothing factor of the obtained GRNN neural network model is;
the expression of the fitness function of the BSA algorithm is as follows:
Figure BDA0003288500010000171
wherein, the fitness is a fitness function of the BSA; y isiOutput values of the ith sample in the training subset are obtained for the GRNN neural network model;
Figure BDA0003288500010000172
the real value of the ith sample in the training subset is obtained; n is a radical oftestIs the number of samples of the training subset.
413, obtaining a predicted value of the GRNN neural network model by utilizing the training subset; and calculating the fitness values of all individuals in the current population according to the predicted value of the GRNN neural network model, selecting the current optimal fitness individual, and setting the position of the individual as the current optimal.
Step 414, obtaining a direction vector according to the direction randomness of the skyhook head by using the BAS algorithm, defining the positions of the left antenna and the right antenna, and calculating the fitness value f (X) of the left antenna and the right antennal) And f (X)r) And judging the direction and the position of the next iteration of the longicorn according to the size relationship of the direction and the position, then calculating the fitness value of the longicorn after moving, judging whether the iteration ending condition is met, ending when the iteration ending condition is met, and otherwise, repeating the process.
And 415, after each iteration updating, transmitting the optimized prediction precision parameters to the GRNN neural network model.
Step 416, judging whether the iteration cycle number reaches a preset value, if so, stopping optimizing the prediction precision parameter to obtain the optimal parameter of the GRNN neural network model; if not, the iterative updating is continued.
In the embodiment, the Generalized Regression Neural Network (GRNN) is one of the radial basis function neural networks, has simple design, fast convergence, strong fault tolerance and robustness, and is suitable for the prediction problem of unstable data such as building cold load; a typical GRNN network structure, as shown in fig. 4, includes an input layer, a mode layer, a summation layer, and an output layer, in the GRNN neural network model, the number of neurons in the input layer is equal to the input vector dimension of the training sample, and each neuron directly passes the input vector to the mode layer; the n neurons of the pattern layer correspond to the n training samples.
Wherein, the expression of the transfer function of the neurons of the mode layer is as follows:
Figure BDA0003288500010000181
wherein, PiA transfer function of neurons that are a mode layer; x is an input vector of the training subset; xiAn ith sample corresponding to an ith neuron in the mode layer; σ is a key factor of the GRNN neural network.
The summation layer includes a first neuron and a second neuron; a first neuron for arithmetically summing the outputs of all neurons in a mode layer; the second neuron is used for carrying out weighted summation on the outputs of all neurons in the mode layer; in the present embodiment, two types of neurons are used in the summation layer; one is to sum the outputs of all neurons in the mode layer arithmetically, the connection weight between the neurons is 1, and the transfer function is:
Figure BDA0003288500010000182
wherein S isDArithmetically summing the outputs of the first neuron of the summation layer, i.e., the outputs of all neurons in the mode layer; p is a radical ofiIs the output of the ith neuron in the pattern layer.
The other is weighted sum of neuron outputs, and the connection weight is output value y of the neuron in the mode layeriThe transfer function is:
Figure BDA0003288500010000191
wherein S isDWeighted summation is carried out on the output of a second neuron of the summation layer, namely the output of all neurons in the mode layer; y isiIs the output weight of the ith neuron in the pattern layer.
The output Y in the output layer is:
Figure BDA0003288500010000192
wherein Y is the output value of the output layer.
And step 42, establishing an LSTM prediction model. Firstly, dividing the optimal influence component element data of the historical building cold load after the characteristic extraction by RF and a recursive characteristic elimination method (RFE) algorithm into a training subset and a testing subset; and training the LSTM neural network by using the training subset to obtain a trained LSTM neural network model.
The Long Short-Term Memory Neural Network (LSTM) is a special Recurrent Neural Network (RNN), and can learn Long-Term dependency and overcome the problem of RNN gradient disappearance. The method has the advantages of high accuracy, strong distributed storage and learning capabilities and the like, has strong robustness and fault-tolerant capability on noise nerves, can fully approximate to a complex nonlinear relation, and simultaneously has the function of associative memory, so that the time sequence problem can be effectively processed. Unlike RNN, LSTM adds a structure called Memory Cell (Memory Cell) to the neural nodes of the hidden layer to memorize past information, and adds three "gate" structures to control the use of history information, namely a forgetting gate, an input gate and an output gate, as shown in fig. 5.
In the LSTM neural network model, for each time step t, the input x of the previous time step is usedtAnd an output ht-1Calculating the current cell state ctAnd an output ht(ii) a First-layer forgetting gate f of LSTM neural network model controls state c of last unit cellt-1Degree of forgetfulness of ft(ii) a The input gate i updates the cell state according to the newly input informationThe updated cell state is transmitted to the next cell it(ii) a The output gate o filters the cell state according to the input data to generate an output result o at the moment tt。
And 5, predicting the test subsets by respectively utilizing the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment.
And 6, determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result of each time, and comparing to determine the prediction use time of the optimized GRNN neural network model and the trained LSTM neural network model, namely constructing the BSA-GRNN & LSTM combined neural network prediction model.
And 7, taking the data of the optimal influence component of the building cold load to be predicted as the input of the BSA-GRNN & LSTM combined neural network prediction model, and outputting to obtain the building cold load prediction result.
And (3) test results:
in this embodiment, the learning and testing of the prediction model are performed on the building cold load energy consumption related data acquired by a certain large commercial building energy consumption monitoring platform in a certain city.
As shown in FIG. 1, FIG. 1 shows a graph of influence coefficients of the influence elements of the building cold load in the embodiment. It can be seen from fig. 2 that after the influence coefficients trained by the model are averaged, the obtained characteristic variables are: each orientation window-to-wall ratio, building shape factor, outdoor weather, wind speed, room personnel, outdoor relative humidity, outdoor dry bulb temperature, lighting usage, solar irradiance at a previous time, and load at a previous time; therefore, the characteristic variables are main influence factors of the screened public building energy consumption.
As shown in FIG. 6, FIG. 6 is a graph of the results of modeling the BAS-GRNN & LSTM neural network model in an embodiment. It can be seen from fig. 6 that the average relative error of the optimized GRNN neural network model and the trained LSTM neural network model for the cold load prediction at different times is smaller than that of the LSTM neural network model, and the prediction error of the optimized GRNN neural network model obtained at 8, 12, 13, 14, 15, and 19 is smaller than that of the LSTM neural network model, so that it is determined that the above times are predicted by the optimized GRNN neural network model, and the rest times are predicted by the trained LSTM neural network model, thereby completing the modeling of BAS-GRN & LSTM model.
Meanwhile, in order to verify the prediction effect of the model, a Root Mean Square Error (RMSE) and an average Absolute Percentage Error (MAPE) are selected as main evaluation indexes of the prediction precision of the model, and the formula is as follows:
Figure BDA0003288500010000211
Figure BDA0003288500010000212
as shown in FIG. 7, FIG. 7 shows the comparison of the prediction results of the cooling load of the large-scale commercial building. As can be seen from FIG. 7, compared with the three prediction models of the Generalized Regression Neural Network (GRNN), the long-term memory (LSTM) and the optimized GRNN neural network model, the prediction result of the prediction model of the BSA-GRNN & LSTM combined neural network in the embodiment is superior to that of other models, the prediction precision is higher, and the Root Mean Square Error (RMSE) and the Mean Absolute Percentage Error (MAPE) of the prediction model of the BSA-GRNN & LSTM combined neural network are 1.4009 and 0.27% respectively; therefore, the difference between the predicted value and the true value of the prediction model of the BSA-GRNN & LSTM combined neural network is minimum, partial values are almost completely overlapped, and the fitting effect is better. In addition, compared with the generalized regression neural network, the prediction accuracy of the optimized GRNN neural network model is improved.
As shown in fig. 8, a regression fitting graph of the real value and the predicted value in the present embodiment is shown in fig. 8. As can be seen from fig. 8, for the prediction of the cold load of the large commercial building, compared with the regression fitting curve of the predicted values and the true values of the three prediction models, i.e., the generalized regression neural network, the long-term memory neural network and the optimized GRNN neural network model, the predicted cold load value of the BSA-GRNN & LSTM combined neural network prediction model described in this embodiment is concentrated near the straight line y ═ x, which indicates that the prediction effect is better.
In order to further verify the parameter capability of the longicorn stigma search algorithm (BAS) optimized Generalized Regression Neural Network (GRNN) provided by the embodiment, the generalized regression neural network (BAS-GRNN) & LSTM prediction model optimized by the longicorn stigma search algorithm is used for predicting the cold load of the case large-scale commercial building in 7 months, and the Genetic Algorithm (GA), the Particle Swarm Optimization (PSO) and the longicorn stigma search algorithm (BAS) are respectively adopted for carrying out iterative optimization on the smoothing factor of the Generalized Regression Neural Network (GRNN) for experimental comparison.
As shown in fig. 9, fig. 9 shows a comparison graph of different algorithm iterations in this embodiment. As can be seen from fig. 9, the fitness value, i.e., the average absolute error, of the optimized GRNN parameter of the three algorithms is in a generally downward trend. However, compared to Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) algorithms, the longicorn whisker search algorithm (BAS) converges the mean error in a shorter time, resulting in the smallest error value. Meanwhile, compared with a Genetic Algorithm (GA) and a Particle Swarm Optimization (PSO), a convergence error iteration curve obtained by a Tianniu whisker search algorithm (BAS) is flatter, and the BAS algorithm has good stability and convergence and can be suitable for GRNN parameter optimization.
In order to further verify the generalization performance of the prediction model of the BAS-GRNN & LSTM combined neural network proposed in this embodiment, the BAS-GRNN & LSTM combined neural network model is used to predict the cooling load of the last four days in five months from 4 months to 8 months of a large commercial building, respectively, the cooling load data of the rest days in the current month is used as the training data of the model, and experimental comparison is performed by using whether a feature extraction algorithm is used as an independent variable.
As shown in fig. 10, fig. 10 shows a generalization capability demonstration diagram in the present embodiment; as can be seen from the attached figure 10, the BAS-GRNN & LSTM combined neural network model has strong generalization capability, has good prediction effect on cold loads of different months, and has an error range stabilized in a small interval range; meanwhile, compared with the method without feature extraction, the prediction effect is better after feature extraction is carried out by combining Random Forest (RF) and Recursive Feature Elimination (RFE), the fact that dimension reduction processing is successfully carried out on the original feature set is shown, the selected features can better establish a prediction model, and therefore model prediction accuracy is further improved.
The above-described embodiment is only one of the embodiments that can implement the technical solution of the present invention, and the scope of the present invention is not limited by the embodiment, but includes any variations, substitutions and other embodiments that can be easily conceived by those skilled in the art within the technical scope of the present invention disclosed.

Claims (10)

1. A building cold load prediction method is characterized by comprising the following steps:
acquiring data of optimal influence constituent elements of the historical building cold load, and constructing a training subset and a testing subset;
training the GRNN neural network by using the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA (bovine serum albumin) algorithm to obtain an optimized GRNN neural network model;
training the LSTM neural network by using the training subset to obtain a trained LSTM neural network model;
respectively predicting the test subsets by using the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment;
determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result at each time, and constructing to obtain a BSA-GRNN & LSTM combined neural network prediction model;
and taking the data of the optimal influence component elements of the building cold load to be predicted as the input of the BSA-GRNN & LSTM combined neural network prediction model, and outputting to obtain the building cold load prediction result.
2. The method for predicting the building cooling load according to claim 1, wherein the process of obtaining the data of the optimal influence constituent elements of the historical building cooling load and constructing the training subset and the testing subset is specifically as follows:
acquiring influence component data of the historical building cold load;
performing feature extraction on the influence component data of the historical building cold load by adopting a random forest algorithm to obtain main influence component data of the historical building cold load;
processing the main influence component data of the historical building cold load by adopting a recursive characteristic elimination method to obtain the optimal influence component data of the historical building cold load;
and constructing a training subset and a testing subset by adopting a random extraction mode on the optimal influence component data of the historical building cold load.
3. The method for predicting the building cold load according to claim 1, wherein the optimal influence component data of the historical building cold load comprise east window-wall ratio, south window-wall ratio, building system coefficient, room personnel, outdoor relative humidity, outdoor dry bulb temperature, illumination use condition, solar irradiance at the previous moment and load at the previous moment.
4. The method according to claim 1, wherein the GRNN neural network is trained using the training subset, and the key factors of the GRNN neural network are optimized using a BSA algorithm to obtain an optimized GRNN neural network model, the key factors of the GRNN neural network being smoothing factors of the GRNN neural network; wherein, the smoothing factor of the GRNN neural network is set to be a random number in [0,1 ].
5. The method according to claim 1, wherein a BSA algorithm is used to optimize key factors of the GRNN neural network to obtain an optimized GRNN neural network model, and an average absolute error of the GRNN neural network model to a training set is used as a fitness function of BSA;
wherein, the expression of the fitness function of the BSA is as follows:
Figure FDA0003288500000000021
wherein, the fitness is a fitness function of the BSA; y isiOutput values of the ith sample in the training subset are obtained for the GRNN neural network model;
Figure FDA0003288500000000022
the real value of the ith sample in the training subset is obtained; n is a radical oftestIs the number of samples of the training subset.
6. The method of claim 4, wherein the optimized GRNN neural network model comprises an input layer, a pattern layer, a summation layer, and an output layer; wherein the number of neurons in the input layer is the same as the input vector dimension of the training subset; the neurons of the input layer directly transfer the input vectors of the training subset to the pattern layer;
the expression of the transfer function of the neurons of the pattern layer is:
Figure FDA0003288500000000023
wherein, PiA transfer function of neurons that are a mode layer; x is an input vector of the training subset; xiAn ith sample corresponding to an ith neuron in the mode layer; sigma is a key factor of the GRNN neural network;
the summation layer includes a first neuron and a second neuron; a first neuron for arithmetically summing the outputs of all neurons in a mode layer; the second neuron is used for carrying out weighted summation on the outputs of all neurons in the mode layer;
the expression of the output layer is:
Figure FDA0003288500000000031
wherein Y is the output value of the output layer; sNAn output of a first neuron that is a summation layer; sDIs the output of the second neuron of the summation layer.
7. A building cooling load prediction system, comprising:
the data acquisition module is used for acquiring the data of the optimal influence component elements of the cold load of the historical building, and constructing a training subset and a testing subset;
the GRNN neural network module is used for training the GRNN neural network by utilizing the training subset, and optimizing key factors of the GRNN neural network by adopting a BSA algorithm to obtain an optimized GRNN neural network model;
the LSTM neural network model is used for training the LSTM neural network by utilizing the training subset to obtain a trained LSTM neural network model;
the average relative error module is used for predicting the test subsets by respectively utilizing the optimized GRNN neural network model and the trained LSTM neural network model, and calculating to obtain the average relative error of each prediction moment;
the combined model module is used for determining the prediction application time of the optimized GRNN neural network model and the trained LSTM neural network model according to the average relative error of the prediction result at each time, and constructing a BSA-GRNN & LSTM combined neural network prediction model;
and the prediction module is used for taking the data of the optimal influence component elements of the building cold load to be predicted as the input of the BSA-GRNN & LSTM combined neural network prediction model and outputting to obtain the building cold load prediction result.
8. The building cold load prediction system of claim 7, wherein the data acquisition module comprises a constituent element data module, a feature extraction module, a recursive feature elimination module and a subset module;
the component data module is used for acquiring influence component data of the historical building cold load;
the characteristic extraction module is used for extracting the characteristics of the influence component data of the historical building cold load by adopting a random forest algorithm to obtain the main influence component data of the historical building cold load;
the recursive characteristic elimination module is used for processing the main influence component element data of the historical building cold load by adopting a recursive characteristic elimination method to obtain the optimal influence component element data of the historical building cold load;
and the subset module is used for constructing a training subset and a testing subset by adopting a random extraction mode on the data of the constituent elements with the optimal influence of the cold load of the historical building.
9. A building cold load prediction device, comprising:
a memory for storing a computer program;
a processor for implementing the steps of a method of predicting the cooling load of a building as claimed in any one of claims 1 to 7 when said computer program is executed.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of a method for predicting the cooling load of a building as claimed in any one of claims 1 to 7.
CN202111156381.9A 2021-09-29 2021-09-29 Building cold load prediction method, system, equipment and readable storage medium Pending CN113887801A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111156381.9A CN113887801A (en) 2021-09-29 2021-09-29 Building cold load prediction method, system, equipment and readable storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111156381.9A CN113887801A (en) 2021-09-29 2021-09-29 Building cold load prediction method, system, equipment and readable storage medium

Publications (1)

Publication Number Publication Date
CN113887801A true CN113887801A (en) 2022-01-04

Family

ID=79004618

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111156381.9A Pending CN113887801A (en) 2021-09-29 2021-09-29 Building cold load prediction method, system, equipment and readable storage medium

Country Status (1)

Country Link
CN (1) CN113887801A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114662751A (en) * 2022-03-19 2022-06-24 特斯联科技集团有限公司 Park multifunctional short-term load prediction and optimization method based on LSTM
CN116127840A (en) * 2023-01-05 2023-05-16 无锡钊源电力电子有限公司 Data center load prediction method based on data driving
CN116227725A (en) * 2023-03-17 2023-06-06 广东热矩智能科技有限公司 Load prediction method and device for building air conditioning system and electronic equipment
CN116401935A (en) * 2023-02-21 2023-07-07 哈尔滨工业大学 Building dynamic thermal load neural network prediction method and system
CN116734918A (en) * 2023-06-05 2023-09-12 宁夏中昊银晨能源技术服务有限公司 Indoor environment monitoring system suitable for near zero energy consumption building
CN117928052A (en) * 2024-03-25 2024-04-26 烟台市市级机关服务中心 Energy-saving control method and system for central air conditioner

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114662751A (en) * 2022-03-19 2022-06-24 特斯联科技集团有限公司 Park multifunctional short-term load prediction and optimization method based on LSTM
CN116127840A (en) * 2023-01-05 2023-05-16 无锡钊源电力电子有限公司 Data center load prediction method based on data driving
CN116127840B (en) * 2023-01-05 2023-09-22 无锡钊源电力电子有限公司 Data center load prediction method based on data driving
CN116401935A (en) * 2023-02-21 2023-07-07 哈尔滨工业大学 Building dynamic thermal load neural network prediction method and system
CN116227725A (en) * 2023-03-17 2023-06-06 广东热矩智能科技有限公司 Load prediction method and device for building air conditioning system and electronic equipment
CN116734918A (en) * 2023-06-05 2023-09-12 宁夏中昊银晨能源技术服务有限公司 Indoor environment monitoring system suitable for near zero energy consumption building
CN116734918B (en) * 2023-06-05 2024-05-14 宁夏中昊银晨能源技术服务有限公司 Indoor environment monitoring system suitable for near zero energy consumption building
CN117928052A (en) * 2024-03-25 2024-04-26 烟台市市级机关服务中心 Energy-saving control method and system for central air conditioner
CN117928052B (en) * 2024-03-25 2024-06-07 烟台市市级机关服务中心 Energy-saving control method and system for central air conditioner

Similar Documents

Publication Publication Date Title
CN113887801A (en) Building cold load prediction method, system, equipment and readable storage medium
CN113282122B (en) Commercial building energy consumption prediction optimization method and system
CN112116144B (en) Regional power distribution network short-term load prediction method
CN111027772A (en) Multi-factor short-term load prediction method based on PCA-DBILSTM
CN114970362B (en) Power grid load scheduling prediction method and system under multi-energy structure
CN111915092B (en) Ultra-short-term wind power prediction method based on long-short-term memory neural network
CN112819136A (en) Time sequence prediction method and system based on CNN-LSTM neural network model and ARIMA model
CN109583635B (en) Short-term load prediction modeling method for operational reliability
CN112734135B (en) Power load prediction method, intelligent terminal and computer readable storage medium
Lu et al. A hybrid model based on convolutional neural network and long short-term memory for short-term load forecasting
CN114358213B (en) Error ablation processing method, system and medium for nonlinear time series data prediction
CN113177673B (en) Air conditioner cold load prediction optimization method, system and equipment
CN113553755B (en) Power system state estimation method, device and equipment
CN113361785A (en) Power distribution network short-term load prediction method and device, terminal and storage medium
CN114021483A (en) Ultra-short-term wind power prediction method based on time domain characteristics and XGboost
CN115759415A (en) Power consumption demand prediction method based on LSTM-SVR
CN115995810A (en) Wind power prediction method and system considering weather fluctuation self-adaptive matching
CN110738363B (en) Photovoltaic power generation power prediction method
CN113850438A (en) Public building energy consumption prediction method, system, equipment and medium
CN113095547B (en) Short-term wind power prediction method based on GRA-LSTM-ICE model
CN116822672A (en) Air conditioner cold load prediction optimization method and system
CN116470491A (en) Photovoltaic power probability prediction method and system based on copula function
CN115545164A (en) Photovoltaic power generation power prediction method, system, equipment and medium
CN114444763A (en) Wind power prediction method based on AFSA-GNN
CN113254857A (en) SSA-ELM-based short-term power load prediction method

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