Public building energy consumption prediction method, system, equipment and medium

Through the Lasso algorithm, CEEMDAN algorithm and TSA algorithm, the RBF and LSTM neural network are optimized, and the TSA-RBF&LSTM combination model is formed, which solves the problems of low energy consumption prediction accuracy and poor generalization ability of public buildings, and achieves more efficient energy consumption prediction and improvement Fast model training.

CN113850438BActive Publication Date: 2024-11-15XI'AN UNIVERSITY OF ARCHITECTURE AND TECHNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111153011.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-29
Publication Date
2024-11-15
Estimated Expiration
2041-09-29

AI Technical Summary

Technical Problem

The existing public building energy consumption prediction model has low accuracy and poor generalization capabilities, especially because building energy consumption is affected by many factors and the data is nonlinear and volatile, resulting in low prediction accuracy of a single prediction model and convergence of model training. Slow speed.

Method used

The Lasso algorithm is used for feature extraction, the CEEMDAN algorithm is used for data decomposition, and the frequency band is divided by approximate entropy theory. The RBF neural network and LSTM neural network are optimized by combining the TSA algorithm to form a TSA-RBF&LSTM combined neural network model for public building energy consumption prediction.

Benefits of technology

It improves the accuracy and generalization performance of energy consumption prediction of public buildings, reduces the dimension of model input variables, improves the convergence speed of the model, reduces the calculation cost, eliminates the problem of optimal parameter selection, and enhances the prediction performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113850438B_ABST
    Figure CN113850438B_ABST
Patent Text Reader

Abstract

The invention discloses a method, system, device and medium for predicting energy consumption of public buildings, comprising: using the main influencing indicator data and low-frequency components of historical public building energy consumption to train a RBF neural network, and using a TSA algorithm to optimize the key parameters of the RBF neural network; using the main influencing indicator data and high-frequency components of historical public building energy consumption to train an LSTM neural network; superimposing the optimized RBF neural network model and the trained LSTM prediction model to obtain a TSA-RBF&LSTM combined neural network prediction model; using the main influencing indicator data of the public building energy consumption to be predicted as the input of the TSA-RBF&LSTM combined neural network prediction model, and outputting the prediction result of the public building energy consumption; compared with the existing single model prediction, the invention effectively improves the prediction performance, prediction accuracy and generalization performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of building energy consumption prediction, and in particular relates to a method, system, equipment and medium for predicting energy consumption of a public building. Background Art

[0002] In recent years, with the acceleration of urbanization and the improvement of residents' quality of life, the proportion of building operation energy consumption in the total social energy consumption has reached more than 20%; among them, the energy consumption of public buildings is the most serious; therefore, the energy management of public buildings is of great significance to energy conservation and emission reduction. Accurate building energy consumption prediction is the key to improving building energy efficiency and implementing building energy conservation; therefore, studying the mechanism and law of public building energy consumption and establishing an accurate and effective public building energy consumption prediction model are of great practical significance.

[0003] At present, the prediction model based on artificial neural network can be effectively applied to the prediction of energy consumption of public buildings and can meet the actual application needs. However, since building energy consumption is affected by many factors, such as outdoor temperature and humidity, indoor personnel flow, and historical energy consumption, and building energy consumption data has the characteristics of nonlinearity and strong volatility, the prediction accuracy and generalization ability of a single prediction model are low, and it is not well suitable for the prediction of energy consumption of large public buildings. At the same time, the network parameter training of radial basis function neural network (RBF) has problems such as slow convergence speed and easy to fall into local optimality, which affects the prediction performance of the model. Summary of the invention

[0004] In view of the technical problems existing in the prior art, the present invention provides a method, system, device and medium for predicting energy consumption of public buildings to solve the technical problems of low accuracy and poor generalization ability of existing public building energy consumption prediction.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] The present invention provides a method for predicting energy consumption of public buildings, comprising the following steps:

[0007] Obtain data on the impact components of historical public building energy consumption;

[0008] The Lasso algorithm is used to extract the features of the influencing factors of the energy consumption of historical public buildings, and the main influencing indicator data of the energy consumption of historical public buildings are obtained;

[0009] The CEEMDAN algorithm is used to decompose the data of the influencing factors of energy consumption of historical public buildings, the intrinsic mode function components of linear steady state and the residual components;

[0010] Using the approximate entropy theory, the intrinsic mode function components and residual components of the linear steady state are divided according to the complexity to obtain high-frequency components and low-frequency components.

[0011] The RBF neural network is trained using the main influencing indicator data and low-frequency components of historical public building energy consumption, and the TSA algorithm is used to optimize the key parameters of the RBF neural network to obtain the optimized RBF neural network model.

[0012] The main influencing indicator data and high-frequency components of historical public building energy consumption are used to train the LSTM neural network to obtain the trained LSTM prediction model;

[0013] The optimized RBF neural network model and the trained LSTM prediction model are superimposed to obtain the TSA-RBF&LSTM combined neural network prediction model;

[0014] The main influencing indicator data of the public building energy consumption to be predicted is used as the input of the TSA-RBF&LSTM combined neural network prediction model, and the prediction result of the public building energy consumption is obtained as the output.

[0015] Furthermore, the data on the influencing factors of public building energy consumption include indoor lighting usage data of buildings, indoor personnel data, outdoor dry-bulb temperature, outdoor wet-bulb temperature, outdoor relative humidity, outdoor dew point temperature, wind speed, wind direction, historical current solar irradiance, solar irradiance at the previous historical moment and public building energy consumption at the previous historical moment.

[0016] Furthermore, the Lasso algorithm is used to extract features from the data of influencing components of historical public building energy consumption. In the process of obtaining the main influencing indicator data of historical public building energy consumption, the data of influencing components of historical public building energy consumption whose coefficient vector of the Lasso regression optimization target model is not zero are used as the main influencing indicator data of historical public building energy consumption.

[0017] Furthermore, the main influencing indicator data of historical public building energy consumption include building indoor lighting usage data, indoor personnel data, outdoor dry-bulb temperature, outdoor relative humidity, wind speed, historical current solar irradiance, solar irradiance at the previous historical moment and public building energy consumption at the previous historical moment.

[0018] Furthermore, the TSA algorithm is used to optimize the key parameters of the RBF neural network, and the key parameters of the RBF neural network include the center vector, the base width vector and the weight vector.

[0019] Furthermore, the expression of the optimized RBF neural network model is:

[0020]

[0021] in, is the output of the optimized RBF neural network model; j is the jth neuron; m is the number of neurons in the hidden layer; w j is the optimized output weight of the jth node in the hidden layer; X is the input vector; C j is the optimized center vector of the jth node in the hidden layer; b j is the optimized basis width parameter of the jth node in the hidden layer.

[0022] Furthermore, the expression of the trained LSTM prediction model is:

[0023] f t =sigmoid(b f +W f,h *h t-1 +W f,x *x t )

[0024] i t =sigmoid(b i +W i,h *h t-1 +W i,x *X t )

[0025] c t =tanh(b c +W c,x *X t +W c,h *h t-1 )

[0026] c t =c t-1 *f t +i t *c t

[0027] o t =sigmoid(b o +W o,h *h t-1 +W o,x *x t )

[0028] h t =o t *tanh(c t )

[0029] Where t is an arbitrary time step; f t is the output of the forget gate at time t; i t is the output of the input gate at time t; c tInformation used to update cell status; c t is the cell state at time t; o t is the output of the output gate at time t; h t are the outputs of the hidden layer at time t; b f , W f,h , W f,x are the coefficients corresponding to the forget gates; b i , W i,h , W i,x are the coefficients corresponding to the input gates; b c , W c,h , W c,x is the coefficient corresponding to the cell state; b o , W o,h , W o,x are the coefficients corresponding to the output gates; h t-1 is the output of the hidden layer at time t-1; x t is the input of the LSTM prediction model; sigmoid and tanh are activation functions respectively.

[0030] The present invention also provides a public building energy consumption prediction system, comprising:

[0031] A data acquisition module is used to obtain data on the influencing components of historical public building energy consumption;

[0032] The feature extraction module is used to extract the features of the data of the influencing elements of the energy consumption of historical public buildings by using the Lasso algorithm, and obtain the main influencing indicator data of the energy consumption of historical public buildings;

[0033] Decomposition module, used to decompose the data of the influencing elements of energy consumption of historical public buildings using CEEMDAN algorithm, the intrinsic mode function components of linear steady state and the residual components;

[0034] A partitioning module is used to partition the intrinsic mode function components and residual components of the linear steady state according to the complexity by using the approximate entropy theory to obtain high-frequency components and low-frequency components;

[0035] The RBF model module is used to train the RBF neural network using the main influencing indicator data and low-frequency components of historical public building energy consumption, and to optimize the key parameters of the RBF neural network using the TSA algorithm to obtain the optimized RBF neural network model;

[0036] The LSTM model module is used to train the LSTM neural network using the main influencing indicator data and high-frequency components of historical public building energy consumption to obtain a trained LSTM prediction model;

[0037] The model superposition module is used to superimpose the optimized RBF neural network model and the trained LSTM prediction model to obtain the TSA-RBF&LSTM combined neural network prediction model;

[0038] The prediction output module is used to use the main influencing indicator data of the public building energy consumption to be predicted as the input of the TSA-RBF&LSTM combined neural network prediction model, and output the prediction result of the public building energy consumption.

[0039] The present invention also provides a public building energy consumption prediction device, comprising:

[0040] Memory for storing computer programs;

[0041] A processor is used to implement the steps of the method for predicting energy consumption of public buildings when executing the computer program.

[0042] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and is characterized in that when the computer program is executed by a processor, the steps of the method for predicting energy consumption of a public building are implemented.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention provides a public building energy consumption prediction method and system, which adopts Lasso algorithm to extract features of influencing elements of public building energy consumption, evaluates the importance of influencing elements of public building energy consumption, obtains main influencing index data of public building energy consumption, and effectively improves the prediction accuracy of the model; at the same time, reduces the dimension of input variables of TSA-RBF&LSTM combined neural network prediction model, improves the convergence speed of the model, and saves computing cost; adopts CEEMDAN algorithm to decompose the influencing element data of historical public building energy consumption, decomposes the influencing element data with nonlinearity and non-stationarity into linear steady-state intrinsic mode function components and residual components, and effectively reduces the complexity of the influencing element data; at the same time, adopts approximate entropy theory to analyze and calculate the linear The complexity of the steady-state intrinsic mode function components and residual components is divided into high-frequency components and low-frequency components, which provides different training data for prediction using models with different characteristics; the TSA algorithm is used to optimize the RBF neural network, eliminating the problem of optimal parameter selection of the RBF neural network in public building energy consumption prediction; the TSA algorithm is combined with the RBF neural network to reduce the prediction error of the RBF neural network model, and the prediction performance is effectively improved; the RBF neural network model is trained using low-frequency components, and the LSTM prediction model is trained using high-frequency components, and the optimized RBF neural network model and the trained LSTM prediction model are combined and superimposed, which can make full use of the characteristics of the two models and effectively improve the prediction performance, prediction accuracy and generalization performance compared with the existing single model prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flowchart of the method for predicting energy consumption of public buildings described in the embodiment;

[0046] Figure 2 It is the Lasso fitting coefficient trajectory diagram of the influencing factors of energy consumption of public buildings in the embodiment;

[0047] Figure 3 It is a CEEMDAN decomposition flow chart in the embodiment;

[0048] Figure 4 It is a CEEMDAN energy consumption decomposition result diagram in the embodiment;

[0049] Figure 5 is the approximate entropy value of the energy consumption component decomposed by CEEMADAN in the embodiment;

[0050] Figure 6 RBF neural network structure diagram used in the embodiment;

[0051] Figure 7LSTM neural network structure diagram used in the embodiment;

[0052] Figure 8 A comparison chart of prediction results using four different prediction models in the embodiment;

[0053] Fig. 9 A comparison chart of prediction accuracy using four different prediction models in the embodiment;

[0054] Fig.10 This is a diagram demonstrating the generalization ability of the TSA-RBF&LSTM prediction model in the embodiment. DETAILED DESCRIPTION

[0055] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention more clearly understood, the present invention is further described in detail in the following specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0056] The present invention provides a method for predicting energy consumption of public buildings, comprising the following steps:

[0057] Step 1, obtain the data of the influencing elements of the energy consumption of historical public buildings; wherein the data of the influencing elements of the energy consumption of public buildings include the building indoor lighting usage data, indoor personnel data, outdoor dry bulb temperature, outdoor wet bulb temperature, outdoor relative humidity, outdoor dew point temperature, wind speed, wind direction, historical current solar irradiance, solar irradiance at the previous moment in history and public building energy consumption at the previous moment in history.

[0058] Step 2, using the Lasso algorithm to extract features from the influencing constituent element data of historical public building energy consumption, and obtain the main influencing indicator data of historical public building energy consumption; in the present invention, the influencing constituent element data of historical public building energy consumption whose coefficient vector of the Lasso regression optimization target model is not zero is used as the main influencing indicator data of historical public building energy consumption; wherein, the main influencing indicator data of historical public building energy consumption include building indoor lighting usage data, indoor personnel data, outdoor dry-bulb temperature, outdoor relative humidity, wind speed, historical current solar irradiance, historical solar irradiance at the previous moment and public building energy consumption at the previous moment.

[0059] Step 3: Use the CEEMDAN algorithm to decompose the data of the influencing factors of historical public building energy consumption, the linear steady-state intrinsic mode function components and residual components.

[0060] Step 4: Using the approximate entropy theory, the intrinsic mode function components and residual components of the linear steady state are divided according to the complexity to obtain high-frequency components and low-frequency components.

[0061] Step 5: Use the main influencing indicator data and low-frequency components of historical public building energy consumption to train the RBF neural network, and use the TSA algorithm to optimize the key parameters of the RBF neural network to obtain the optimized RBF neural network model; among them, the key parameters of the RBF neural network include the center vector, base width vector and weight vector.

[0062] Among them, the expression of the optimized RBF neural network model is:

[0063]

[0064] in, is the output of the optimized RBF neural network model; j is the jth neuron; m is the number of neurons in the hidden layer; w j is the optimized output weight of the jth node in the hidden layer; X is the input vector; C j is the optimized center vector of the jth node in the hidden layer; b j is the optimized basis width parameter of the jth node in the hidden layer.

[0065] Step 6: Use the main influencing indicator data and high-frequency components of historical public building energy consumption to train the LSTM neural network and obtain the trained LSTM prediction model.

[0066] Among them, the expression of the trained LSTM prediction model is:

[0067] f t =sigmoid(b f +W f,h *h t-1 +W f,x *x t )

[0068] i t =sigmoid(b i +W i,h *h t-1 +W i,x *X t )

[0069] c t =tanh(b c +W c,x *X t +W c,h *h t-1 )

[0070] c t =c t-1 *f t +i t *c t

[0071] o t =sigmoid(b o +W o,h *h t-1 +W o,x *x t )

[0072] h t =o t *tanh(c t )

[0073] Where t is an arbitrary time step; f t is the output of the forget gate at time t; i t is the output of the input gate at time t; c t Information used to update cell status; c t is the cell state at time t; o t is the output of the output gate at time t; h t are the outputs of the hidden layer at time t; b f , W f,h , W f,x are the coefficients corresponding to the forget gates; b i , W i,h , W i,x are the coefficients corresponding to the input gates; b c , W c,h , W c,x is the coefficient corresponding to the cell state; b o , W o,h , W o,x are the coefficients corresponding to the output gates; h t-1 is the output of the hidden layer at time t-1; x t is the input of the LSTM prediction model; sigmoid and tanh are activation functions respectively.

[0074] Step 7: Superimpose the optimized RBF neural network model and the trained LSTM prediction model to obtain the TSA-RBF&LSTM combined neural network prediction model.

[0075] Step 8: Use the main influencing indicator data of the public building energy consumption to be predicted as the input of the TSA-RBF&LSTM combined neural network prediction model, and output the prediction result of the public building energy consumption.

[0076] The public building energy consumption prediction method described in the present invention adopts the Lasso algorithm to extract features of the influencing factors of public building energy consumption, evaluates the importance of the influencing factors of public building energy consumption, obtains the main influencing indicator data of public building energy consumption, and effectively improves the prediction accuracy of the model; at the same time, it reduces the dimension of the input variables of the TSA-RBF&LSTM combined neural network prediction model, improves the convergence speed of the model, and saves computing costs.

[0077] The CEEMDAN algorithm is used to decompose the data of the influencing factors of historical public building energy consumption, and the nonlinear and non-stationary influencing factor data are decomposed into linear steady-state intrinsic mode function components and residual components, which effectively reduces the complexity of the influencing factor data; at the same time, the complexity of the linear steady-state intrinsic mode function components and residual components calculated by approximate entropy theory are divided into high-frequency components and low-frequency components, which provides different training data for prediction using models with different characteristics.

[0078] The TSA algorithm is used to optimize the RBF neural network, eliminating the problem of optimal parameter selection of the RBF neural network in public building energy consumption prediction; the TSA algorithm is combined with the RBF neural network to reduce the prediction error of the RBF neural network model, and the prediction performance is effectively improved; the RBF neural network model is trained with low-frequency components, and the LSTM prediction model is trained with high-frequency components, and the optimized RBF neural network model and the trained LSTM prediction model are combined and superimposed, which can make full use of the characteristics of the two models. Compared with the existing single model prediction, it effectively improves the prediction performance, prediction accuracy and generalization performance.

[0079] The present invention also provides a public building energy consumption prediction system, including a data acquisition module, a feature extraction module, a decomposition module, a partitioning module, an RBF model module, an LSTM model module, a model superposition module and a prediction output module; the data acquisition module is used to obtain the data of the influencing constituent elements of the historical public building energy consumption; the feature extraction module is used to use the Lasso algorithm to perform feature extraction on the data of the influencing constituent elements of the historical public building energy consumption, and obtain the main influencing indicator data of the historical public building energy consumption; the decomposition module is used to use the CEEMDAN algorithm to decompose the data of the influencing constituent elements of the historical public building energy consumption, and obtain the eigenmode function components and residual components of the linear steady state; the partitioning module is used to use the approximate entropy theory to divide the eigenmode function components and the residual components of the linear steady state according to the complexity to obtain the high-frequency component and the low-frequency component; the RBF model The model module is used to train the RBF neural network by adopting the main influencing indicator data and low-frequency components of the historical public building energy consumption, and optimize the key parameters of the RBF neural network by adopting the TSA algorithm to obtain the optimized RBF neural network model; the LSTM model module is used to train the LSTM neural network by adopting the main influencing indicator data and high-frequency components of the historical public building energy consumption, and obtain the trained LSTM prediction model; the model superposition module is used to superimpose the optimized RBF neural network model and the trained LSTM prediction model to obtain the TSA-RBF&LSTM combined neural network prediction model; the prediction output module is used to use the main influencing indicator data of the public building energy consumption to be predicted as the input of the TSA-RBF&LSTM combined neural network prediction model, and output the prediction result of the public building energy consumption.

[0080] The present invention also provides a public building energy consumption prediction device, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to implement the steps of the public building energy consumption prediction method when executing the computer program.

[0081] When the processor executes the computer program, the steps of the above-mentioned public building energy consumption prediction method are implemented, for example:

[0082] Obtain the data of the influencing elements of the energy consumption of historical public buildings; use the Lasso algorithm to extract the features of the data of the influencing elements of the energy consumption of historical public buildings, and obtain the main influencing indicator data of the energy consumption of historical public buildings; use the CEEMDAN algorithm to decompose the data of the influencing elements of the energy consumption of historical public buildings, the eigenmode function components and the residual components of the linear steady state; use the approximate entropy theory to divide the eigenmode function components and the residual components of the linear steady state according to the complexity, and obtain high-frequency components and low-frequency components; use the main influencing indicator data and low-frequency components of the energy consumption of historical public buildings to train the RBF neural network, and The TSA algorithm is used to optimize the key parameters of the RBF neural network to obtain the optimized RBF neural network model; the main influencing indicator data and high-frequency components of historical public building energy consumption are used to train the LSTM neural network to obtain the trained LSTM prediction model; the optimized RBF neural network model and the trained LSTM prediction model are superimposed to obtain the TSA-RBF&LSTM combined neural network prediction model; the main influencing indicator data of the public building energy consumption to be predicted is used as the input of the TSA-RBF&LSTM combined neural network prediction model, and the prediction result of the public building energy consumption is obtained as the output.

[0083] Alternatively, when the processor executes the computer program, the functions of each module in the above system are realized, for example: a data acquisition module is used to obtain the data of the influencing constituent elements of the energy consumption of historical public buildings; a feature extraction module is used to use the Lasso algorithm to extract the features of the data of the influencing constituent elements of the energy consumption of historical public buildings, and obtain the main influencing indicator data of the energy consumption of historical public buildings; a decomposition module is used to use the CEEMDAN algorithm to decompose the data of the influencing constituent elements of the energy consumption of historical public buildings, and obtain the eigenmode function components and residual components of the linear steady state; a partitioning module is used to use the approximate entropy theory to divide the eigenmode function components and residual components of the linear steady state according to the complexity to obtain high-frequency components and low-frequency components; an RBF model module is used to use the main influencing factors of the energy consumption of historical public buildings The RBF neural network is trained based on the indicator data and low-frequency components, and the key parameters of the RBF neural network are optimized using the TSA algorithm to obtain the optimized RBF neural network model; the LSTM model module is used to train the LSTM neural network using the main influencing indicator data and high-frequency components of historical public building energy consumption to obtain the trained LSTM prediction model; the model superposition module is used to superimpose the optimized RBF neural network model and the trained LSTM prediction model to obtain the TSA-RBF&LSTM combined neural network prediction model; the prediction output module is used to use the main influencing indicator data of the public building energy consumption to be predicted as the input of the TSA-RBF&LSTM combined neural network prediction model, and output the prediction result of the public building energy consumption.

[0084] Exemplarily, the computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments that can complete preset functions, and the instruction segments are used to describe the execution process of the computer program in the public building energy consumption prediction device. For example, the computer program can be divided into a data acquisition module, a feature extraction module, a decomposition module, a partitioning module, an RBF model module, an LSTM model module, a model superposition module and a prediction output module. The specific functions of each module are as follows: a data acquisition module, which is used to obtain the data of the influencing constituent elements of the energy consumption of historical public buildings; a feature extraction module, which is used to use the Lasso algorithm to extract the features of the influencing constituent elements of the energy consumption of historical public buildings, and obtain the main influencing indicator data of the energy consumption of historical public buildings; a decomposition module, which is used to use the CEEMDAN algorithm to decompose the data of the influencing constituent elements of the energy consumption of historical public buildings, and obtain the intrinsic mode function components and residual components of the linear steady state; a partitioning module, which is used to use the approximate entropy theory to divide the intrinsic mode function components and residual components of the linear steady state according to the complexity to obtain high-frequency components and low-frequency components; RB The F model module is used to train the RBF neural network using the main influencing indicator data and low-frequency components of historical public building energy consumption, and use the TSA algorithm to optimize the key parameters of the RBF neural network to obtain the optimized RBF neural network model; the LSTM model module is used to train the LSTM neural network using the main influencing indicator data and high-frequency components of historical public building energy consumption to obtain the trained LSTM prediction model; the model superposition module is used to superimpose the optimized RBF neural network model and the trained LSTM prediction model to obtain the TSA-RBF&LSTM combined neural network prediction model; the prediction output module is used to use the main influencing indicator data of the public building energy consumption to be predicted as the input of the TSA-RBF&LSTM combined neural network prediction model, and output the prediction result of the public building energy consumption.

[0085] The public building energy consumption prediction device can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The public building energy consumption prediction device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above is only an example of a public building energy consumption prediction device and does not constitute a limitation on the public building energy consumption prediction device. It may include more components, or a combination of certain components, or different components. For example, the public building energy consumption prediction device may also include input and output devices, network access devices, buses, etc.

[0086] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The processor is the control center of the public building energy consumption prediction device, and uses various interfaces and lines to connect various parts of the entire public building energy consumption prediction device.

[0087] The memory can be used to store the computer program and / or module, and the processor implements various functions of the public building energy consumption prediction device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0088] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0089] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for predicting energy consumption of public buildings are implemented.

[0090] If the modules / units integrated in the public building energy consumption prediction system are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0091] Based on such understanding, the present invention implements all or part of the processes in the above-mentioned public building energy consumption prediction method, and can also be completed by instructing related hardware through a computer program, and 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 above-mentioned public building energy consumption prediction method can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or preset intermediate form, etc.

[0092] The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0093] It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0094] Example

[0095] As attached Figure 1 As shown, taking a large public building in a city as a research object, this embodiment provides a method for predicting energy consumption of a public building, including the following steps:

[0096] Step 1, obtain the data of the influencing factors of historical public building energy consumption; wherein, the influencing factors of historical public building energy consumption include the building indoor lighting usage data, indoor personnel data, outdoor dry bulb temperature, outdoor wet bulb temperature, outdoor relative humidity, outdoor dew point temperature, wind speed, wind direction, historical current solar irradiance, solar irradiance at the previous moment in history and public building energy consumption at the previous moment in history.

[0097] Step 2: Use the Lasso algorithm to extract features of the influencing factors of the energy consumption of historical public buildings, and obtain the main influencing indicators of the energy consumption of historical public buildings; among them, the influencing factor data of the energy consumption of historical public buildings that makes the coefficient vector of the Lasso regression optimization target model not zero is used as the main influencing indicator data of the energy consumption of historical public buildings.

[0098] The Lasso algorithm is the least absolute shrinkage and selection operator (Lasso), which is a feature selection method that can select stable and interpretable feature subsets. The basic idea of ​​the Lasso algorithm is to add a constraint that the sum of the absolute values ​​of all independent variable coefficients is less than a threshold, so that when solving the multivariate regression model, the independent variable coefficients that are weakly correlated with the prediction target are zero, thereby playing a role in feature selection. Compared with traditional feature selection algorithms, the Lasso algorithm has better prediction accuracy and interpretability than the least squares method, and its calculation process is simpler than the exponential algorithm.

[0099] In this embodiment, the process of using the Lasso algorithm to extract features of the influencing elements of the energy consumption of historical public buildings and obtaining the main influencing indicators of the energy consumption of historical public buildings specifically includes the following steps:

[0100] Step 21: Perform data standardization on the acquired historical public building energy consumption influencing factor data to eliminate the different influences of different feature dimensions and obtain standardized feature data; wherein, the data standardization is performed according to the following data standardization formula:

[0101] The data normalization formula is:

[0102]

[0103] Among them, z * is the standardized feature data, z mean is the arithmetic mean of the original impact component data, z std is the standard deviation of the original data influencing the constituent element data, and z is the original data influencing the constituent element data.

[0104] Step 22: Use N groups of standardized feature data to construct an initial training set (X i1 ,y i1 ), i1=1,2,...,N; where X i1 is the independent variable of the initial training set, y i1 is the response variable, and i1 is the number of characteristic data sets after standardization of the i1th influencing factor.

[0105] In this embodiment, the independent variable X of the initial training set i1 The expression is:

[0106]

[0107] in, is the pth characteristic variable in the i1th influencing component, and p is the number of characteristic variables of a certain influencing component data.

[0108] Step 23: Establish a Lasso regression target model; let the coefficient vector of the Lasso regression target model be Coefficient vector The expression is in, is the regression coefficient of the pth characteristic variable.

[0109] In this embodiment, the expression of the Lasso regression target model is:

[0110]

[0111] in, is a constant term; is the regression coefficient; is the j1th characteristic variable in the i1th influencing factor; t1 is the constraint value, and the constraint value t1 is the regression coefficient L1 norm penalty, t1≥0; j1 is the j1th feature variable.

[0112] In this embodiment, when the value of the constraint value t is less than the preset value, the coefficients with low correlation are compressed to zero, thereby achieving the purpose of feature selection; for all constraint values ​​t1, the constant term The estimation formula is:

[0113]

[0114] Step 24, set is the complete least squares estimate, and make In this embodiment, the mean square error (MSE) is used as the evaluation criterion of the Lasso regression target model, and a cross-validation model training method is adopted to solve the optimal regularization parameter λ of the model.

[0115] In this embodiment, the expression of mean square error MSE is:

[0116]

[0117] Among them, MSE is the mean square error, n1 is the number of test samples, y i1 is the true value of the response variable, is the predicted value of the response variable.

[0118] Step 25: According to the optimal regularization parameter λ of the model, a feature subset whose coefficient vector of the Lasso regression optimization target model is not zero is obtained, that is, the influencing component data of the energy consumption of historical public buildings whose coefficient vector of the Lasso regression optimization target model is not zero is used as the main influencing indicator data of the energy consumption of historical public buildings.

[0119] In this embodiment, in the process of feature extraction, a 10-fold cross-validation model training method is adopted; specifically, the overall initial training sample set is decomposed into 10 sub-sample sets, 9 of which are alternately used as training sets, and the remaining 1 sub-sample set is used as a test set to train the model, and finally the mean square error MSE obtained from 10 trainings is averaged; it can effectively improve the stability of the model, increase the credibility of the feature selection results, solve the regularization parameter λ corresponding to the minimum mean square error MSE, and obtain the feature subset whose coefficient vector of the Lasso regression optimization target model is not zero, which is the main influencing indicator data of historical public building energy consumption.

[0120] As attached Figure 2 As shown, attached Figure 2 The result of extracting the features of the elements influencing the energy consumption of historical public buildings by using the Lasso algorithm is given in the figure, i.e., the Lasso fitting coefficient trajectory diagram of the energy consumption influencing factors of a large public building in this embodiment; Figure 2 It can be seen that when the model error is the smallest, the characteristic variables whose model coefficients are not zero include lighting usage, room occupancy, outdoor dry-bulb temperature, outdoor relative humidity, wind speed, solar irradiance, solar irradiance at the previous moment, and energy consumption at the previous moment; therefore, in this embodiment, the main influencing indicator data of historical public building energy consumption selected include building indoor lighting usage data, indoor occupancy data, outdoor dry-bulb temperature, outdoor relative humidity, wind speed, historical current solar irradiance, historical solar irradiance at the previous moment, and public building energy consumption at the previous moment.

[0121] Step 3: Use the CEEMDAN algorithm to decompose the data of the influencing factors of historical public building energy consumption, the linear steady-state intrinsic mode function components and residual components.

[0122] Empirical Mode Decomposition (EMD) is a widely used time-frequency signal decomposition method that can adaptively decompose nonlinear and non-stationary signals into the sum of linear and steady-state intrinsic mode functions (IMF) and a residual component. The EMD method does not need to set basis functions when decomposing signals, and processes signals completely according to the time scale characteristics of the data itself. Compared with short-time Fourier transform, wavelet decomposition and other methods, the decomposition process of EMD is simpler and more intuitive. The advantages of EMD are strong adaptability and good completeness, but modal aliasing seriously affects the decomposition accuracy of EMD. As an improved method of EMD, the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) algorithm based on adaptive noise effectively solves the modal aliasing problem of the EMD algorithm.

[0123] As attached Figure 3 As shown in the figure, the CEEMDAN algorithm is used to decompose the data of the influencing factors of the energy consumption of historical public buildings, and the process of the linear steady-state intrinsic mode function components and residual components specifically includes the following steps:

[0124] Step 31: Generate a public building energy consumption signal set x containing Gaussian white noise a (t); where x i The expression of (t) is:

[0125] x a (t) = x(t) + ω a (t)

[0126] Among them, x(t) is the original public building energy consumption signal; ω a (t) is white noise that satisfies Gaussian distribution, a=1,2,...,A; A is the number of times white noise is added, that is, the total number of samples of the public building energy consumption signal set.

[0127] Step 32: The public building energy consumption signal set x containing Gaussian white noise a (t) Perform empirical mode decomposition to obtain the first-order linear steady-state intrinsic mode function component IMF of each sample a 1; The first-order linear steady-state intrinsic mode function component IMF of each sample a The mean value of 1 is taken as the first-order linear steady-state intrinsic mode function component of the original public building energy consumption signal; that is:

[0128]

[0129] in, is the first-order intrinsic mode function component obtained by CEEMDAN decomposition.

[0130] Step 33, calculate the first-order residual component and the second-order linear steady-state intrinsic mode function component; wherein the calculation formulas for the first-order residual component and the second-order linear steady-state intrinsic mode function component are as follows:

[0131]

[0132]

[0133] Where r1(t) is the first-order residual component; ε1 is the white noise amplitude corresponding to the first-order residual component; E1(·) is the first-order linear steady-state intrinsic mode function component of the signal obtained by EMD decomposition; ω a (t) is the white noise added for the ath time.

[0134] Step 34, calculate the k-order residual component and the k+1-order linear steady-state intrinsic mode function component; wherein k=2, 3, ..., K, K is the highest order of the linear steady-state intrinsic mode function component; the calculation formula of the k-order residual component and the k+1-order linear steady-state intrinsic mode function component is as follows:

[0135]

[0136]

[0137] Among them, r k (t) is the k-order residual component; r k-1 (t) is the k-1 order residual component; is the k-order intrinsic mode function component obtained by CEEMDAN decomposition; is the k+1th order intrinsic mode function obtained by CEEMDAN decomposition; ε a is the white noise amplitude corresponding to the a-order residual component; E k (·) is the k-order linear steady-state intrinsic mode function component of the signal obtained by EMD decomposition; ω i (t) is the white noise added for the ith time.

[0138] Step 35, repeat step 34 until the number of extreme points of the residual component R(t) is less than 3, then it cannot be decomposed any further and the algorithm terminates; the final decomposition result is expressed as:

[0139]

[0140] In this embodiment, the CEEMDAN algorithm is used to decompose the data of the influencing factors of the energy consumption of historical public buildings. The white noise amplitude is set to 0.2, the number of white noise additions is 100 times, and the number of iterations is 500 times. The final decomposition result is shown in the attached figure. Figure 4 shown.

[0141] Step 4: Using the approximate entropy theory, the intrinsic mode function components and residual components of the linear steady state are divided according to the complexity to obtain high-frequency components and low-frequency components.

[0142] The approximate entropy theory was proposed by Pincus and is a method used to measure the complexity of time series. The larger the approximate entropy value, the more complex the sequence. In this embodiment, the approximate entropy solution process is as follows:

[0143] Step 41: Assume that the building energy consumption data is an M-dimensional time series u(1), u(2), ..., u(M); wherein u(M) is the Mth energy consumption data sample.

[0144] Step 42: Determine the embedding dimension b, and obtain a b-dimensional vector X(c); wherein the expression of the b-dimensional vector X(c) is:

[0145] X(c)=[u(c),u(c+1),u(c+b-1)]

[0146] Among them, u(c) is the cth energy consumption data sample; u(c+b-1) is the c+b-1th energy consumption data sample; c is the dimension of the cth building energy consumption data sample, c=1,2,...,M-b+1.

[0147] Step 43: Calculate the distance d[X(c), X(e)] between the c-th energy consumption data sample vector X(c) and the e-th energy consumption data sample vector X(e) in the phase space; wherein the expression of the distance d[X(c), X(e)] is:

[0148] d[X(c),X(e)]=max 0~b-1 |u(c+g)-u(e+g)|

[0149] Among them, u(c+g) is the c+g-th energy consumption data sample; u(e+g) is the e+g-th energy consumption data sample; and b is the embedding dimension.

[0150] Set the threshold ε, count the number of distances below the threshold ε, and calculate its ratio to the total number:

[0151]

[0152] Among them, C b c(ε) is the cth distance statistic parameter when the embedding dimension is b under the threshold ε; U{d(X(c),X(e))<ε} is the number of vectors X(c) and X(e) whose distance is less than the threshold ε.

[0153] Step 44: C b c (ε) Take the logarithm and then find the average value:

[0154]

[0155] Among them, φ m (ε) is the mean of the distance statistics parameter when the embedding dimension is b under the threshold ε.

[0156] Step 45, increase the dimension to b+1, repeat steps 42-44; get the distance statistic parameter mean φ when the embedding dimension is b+1 under the threshold ε b+1 (ε).

[0157] Step 46: Calculate the approximate entropy estimate:

[0158] ApEn(b,ε,M)=φ b (ε)-φ b+1 (ε)

[0159] Among them, ApEn(b,ε,M) is the approximate entropy value when the embedding dimension is b, the threshold is ε, and the data sample dimension is M.

[0160] In this embodiment, the approximate entropy theory is used to calculate the linear steady-state intrinsic mode function components and residual components after the decomposition of the influencing factor data of historical public building energy consumption; the parameters are set as: b=2, ε=0.2Std; where Std is the standard deviation of the original time series.

[0161] In this embodiment, the calculation results are shown in the attached Figure 5 As shown, IMF1-IMF4 are divided into high-frequency components, and IMF5-IMF7 and the residual component R are divided into low-frequency components.

[0162] From the attached Figure 4-5 It can be seen that the energy consumption data of a large public building in this embodiment can be decomposed into 7 IMF components and 1 residual component, and can be divided into high-frequency components and low-frequency components through the approximate entropy value, among which IMF1~IMF4 are high-frequency components, and IMF5~IMF7 and the residual component R are low-frequency components.

[0163] Step 5: Use the main influencing indicator data and high-frequency components of historical public building energy consumption to train the RBF neural network, and use the TSA algorithm to optimize the key parameters of the RBF neural network to obtain the optimized RBF neural network model; among which, the key parameters of the RBF neural network include the center vector, base width vector and weight vector; use the main influencing indicator data and high-frequency components of historical public building energy consumption to train the LSTM neural network to obtain the trained LSTM prediction model; superimpose the optimized RBF neural network model and the trained LSTM prediction model to obtain the TSA-RBF&LSTM combined neural network prediction model, which can give full play to the prediction performance of different models, thereby achieving the purpose of improving the prediction accuracy.

[0164] The specific process is as follows:

[0165] Step 51, establishing an optimized RBF neural network model; first, combining the main influencing indicator data of historical public building energy consumption obtained by feature extraction using the Lasso algorithm with the low-frequency components in step 4, and constructing a training set and a test set; using the training set and the test set to train and optimize the RBF neural network; wherein, in the training optimization process, the TSA algorithm is used to optimize the center vector, base width vector and weight vector of the RBF neural network to obtain the optimized RBF model; the optimization process using the TSA algorithm specifically includes the following steps:

[0166] Step 511, determine the key parameters of the RBF neural network, and determine the parameter range of the key parameters of the RBF neural network; in this embodiment, the key parameters of the RBF neural network include a center vector, a base width vector and a weight vector; wherein each component in the center vector is set to a random number within [-3, 3]; each component of the base width vector is set to a random number within [0, 3]; each component of the weight vector is set to a random number within [-1, 1].

[0167] Step 512: according to the parameter range of the key parameters of the RBF neural network, set the optimization range of the TSA algorithm influencing parameters; the TSA algorithm influencing parameters include population size, search trend parameters, optimization dimension, maximum number of iterations, upper and lower limits of tree species population positions, positions of randomly initialized tree species populations, and the fitness function of the TSA algorithm.

[0168] In this embodiment, the fitness function of the TSA algorithm is set to the root mean square error (Root Mean Square Error). The smaller the fitness function value is, the more accurate the prediction value is, and the obtained center vector, base width vector and weight vector of the RBF neural network are better. The root mean square error calculation formula is as follows:

[0169]

[0170] Among them, fitness is the fitness function; n1 is the number of test samples, is the true value of the response variable, is the predicted value of the response variable.

[0171] Step 513: Utilize the low-frequency component training set and tree species population information to obtain the prediction value of the RBF neural network; and calculate the fitness values ​​of all individuals in the current population based on the prediction value of the RBF neural network, select the individual with the current optimal fitness, and set the position of the individual as the current optimal.

[0172] Step 514, use the control parameter search trend ST of the TSA algorithm to iteratively update the individual seed positions; when the random number generated in [0,1] is less than ST, update according to the positions of the seed's mother tree, the current best tree and another random tree; otherwise, update according to the positions of the seed's mother tree and another random tree.

[0173] Step 515: After each iterative update, the key parameters of the optimized RBF neural network are transferred to the RBF neural network model.

[0174] Step 516, determine whether the number of iteration cycles reaches the preset value, if so, stop optimizing the prediction accuracy parameters to obtain the optimal parameters of the RBF neural network model; if not, continue iterative updating.

[0175] In this embodiment, the RBF neural network, namely the radial basis function neural network (RBF), is a three-layer feedforward network proposed by J. Moody and C. Darken based on the principle that biological neurons have local responses. It has the advantages of simple structure, strong generalization ability, and the ability to approximate any nonlinear function. It has been widely used in pattern recognition, signal processing, time series prediction and other fields. The structure of the RBF neural network is shown in the attached figure. Figure 6 As shown, it includes input layer, hidden layer and output layer.

[0176] The action function of the hidden layer unit is generally a Gaussian basis function, and the output of neuron j in the hidden layer is:

[0177]

[0178] j=1,2,...,m

[0179] X=(x1,x2,...,x n )

[0180] C j =(c j1 ,cj2 ,...c jn )

[0181] Where: m is the number of neurons in the hidden layer; X is the input vector of the network; x n is the characteristic variable value of the nth dimension of the sample; n is the number of sample characteristic variables; C j is the center vector of the jth neuron node; b j is the base width parameter of the jth node; c jn is the nth component of the center vector of the jth neuron node.

[0182] The expression of the output layer of the RBF neural network is:

[0183]

[0184] in, is the predicted value of the sample; w j is the connection weight from the jth node in the hidden layer to the output layer; h j is the output of the jth node in the hidden layer; m is the number of neurons in the hidden layer.

[0185] In the present invention, it can be seen from the training process of the RBF neural network that the key parameters that need to be trained for the RBF neural network include the center vector C j 、Base width vector B=(b1,b2,...,b m ) and weight vector W = (w1,w2,...,w m ).

[0186] As a new type of swarm intelligence optimization algorithm, the Tree Seed Algorithm (TSA) is based on the relationship between trees and seeds in nature: seeds can grow into new trees over time, and trees spread through seeds. Assuming that the land where the trees grow is used as the search space for the optimization problem, the locations of the trees and seeds are considered to be possible solutions to the optimization problem.

[0187] Each tree in the group will generate a random number of seeds in each iteration, and these seeds update their positions according to the position of their parent tree or the best tree and the position of another random tree. The updating method is determined by the control parameter search tendency (ST). Compared with general swarm intelligence optimization algorithms (such as Artificial Bee Colony (ABC) and Firefly Algorithm), the local search mode of this algorithm is more intense and sophisticated, so it has better global optimization ability.

[0188] In this embodiment, the seed position update formula of the TSA algorithm is:

[0189]

[0190] Where i2 is the i2th tree or seed, i2=1,2,...,N2, N2 is the population size; j2 is the j2th dimension, j2=1,2,...,d, d is the dimension of the tree or seed; ST is the search trend, and its value range is [0,1]; The value of the j2th dimension of the i2th seed generated for the i-th tree; is the value of the j2th dimension of the i-th tree; is the value of the j2th dimension of the best tree currently obtained; is the value of the j2th dimension of the rth tree randomly selected from the population; is a randomly generated scaling factor in the range [-1,1].

[0191] The seed position update is controlled by the search trend ST. When the randomly generated value in the range [0,1] is less than the search trend ST, it is updated using formula (1); otherwise, it is updated using formula (2). The number of seeds generated by each tree in the population is random and its range is [0.1N2, 0.25N2].

[0192] The formula for generating the initial position of the tree is:

[0193]

[0194] in: is the lower limit of the search space; is the upper limit of the search space; is a random number generated in the range [0,1].

[0195] The formula for generating the position of the best tree in the population is:

[0196]

[0197] Where: B is the position of the best tree; f(·) is the fitness function; is the position of the i2th tree.

[0198] Step 52: Use the main influencing indicator data and high-frequency components of historical public building energy consumption to train the LSTM neural network to obtain a trained LSTM prediction model; wherein the main influencing indicator data of public buildings after feature extraction by the Lasso algorithm are respectively combined with the high-frequency components after decomposition of the building energy consumption data to construct a training set and a test set; use the training set and the test set to train the LSTM neural network to obtain a trained LSTM prediction model.

[0199] As a special type of recurrent neural network (RNN), Long Short-Term Memory (LSTM) neural network can learn long-term dependencies and overcome the gradient vanishing problem of recurrent neural networks. LSTM neural networks have the advantages of high accuracy, distributed storage, and strong learning ability. They are robust and fault-tolerant to noisy neural networks, can fully approximate complex nonlinear relationships, and have the function of associative memory, so they can effectively handle time series problems. Different from recurrent neural networks, LSTM neural networks add a structure called memory cell to the hidden layer of neural nodes to remember past information, and add three "gate" structures to control the use of historical information, namely forget gate, input gate, and output gate, as shown in the attached figure. Figure 7 shown.

[0200] In LSTM neural networks, for each time step t, the input x from the previous time step is used t And output h t-1 Calculate the current cell state c t And output h t ; The first layer of the LSTM neural network forget gate f controls the previous unit cell state c t-1 Degree of forgetfulness:

[0201] f t =sigmoid(b f +W f,h *h t-1 +W f,x *x t )

[0202] Among them, f t is the output of the forget gate at time t; sigmoid(·) is the activation function; b f , W f,h , W f,x is the coefficient corresponding to the forget gate; h t-1 is the output of the hidden layer at time t-1; x t is the input at time t; f is the output of the forget gate; h is the output of the hidden layer; x is the input.

[0203] The input gate i updates the cell state according to the new input information and transmits the updated cell state to the next unit:

[0204] i t =sigmoid(b i +W i,h *h t-1 +W i,x *Xt )

[0205] c t =tanh(b c +W c,x *X t +W c,h *h t-1 )

[0206] Among them, i t is the output of the input gate at time t; b i , W i,h , W i,x is the coefficient corresponding to the input gate; h t-1 is the output of the hidden layer at time t-1; x t is the input at time t; c t is the information used to update the cell state; tanh(·) is the activation function; b c , W c,x , W c,h is the coefficient corresponding to the cell state.

[0207] The output gate o filters the cell state according to the input data and produces the output result of this stage:

[0208] c t =c t-1 *f t +i t *c t

[0209] o t =sigmoid(b o +W o,h *h t-1 +W o,x *x t )

[0210] h t =o t *tanh(c t )

[0211] Among them, c t is the state of the cell at time t; c t-1 is the state of the cell at time t-1; o t is the output of the output gate at time t; sigmoid(·) is the activation function; b o , W o,h , W o,x is the coefficient corresponding to the output gate; h t is the output of the hidden layer at time t.

[0212] Step 53: superimpose the optimized RBF neural network model and the trained LSTM prediction model to obtain a TSA-RBF&LSTM combined neural network prediction model.

[0213] Step 6: Use the main influencing indicator data of the public building energy consumption to be predicted as the input of the TSA-RBF&LSTM combined neural network prediction model, and output the prediction result of the public building energy consumption.

[0214] Test results:

[0215] In this embodiment, the prediction model is learned and tested based on the building energy consumption related data collected by a public building energy consumption monitoring platform in a certain city.

[0216] As attached Figure 8 As shown, attached Figure 8 The energy consumption prediction results of a large public building in this embodiment are compared; Figure 8 It can be seen that compared with the three prediction models of radial basis function neural network RBF, TSA-RBF optimized by TSA algorithm, and long short-term memory neural network LSTM, the difference between the predicted value of energy consumption of a large public building obtained by TSA-RBF&LSTM described in this embodiment and the true value is the smallest, some values ​​are almost completely overlapped, and the fitting effect is better.

[0217] At the same time, in order to verify the prediction effect of the model, the root mean square error (RMSE) and the mean absolute percentage error (MAPE) are selected as the main evaluation indicators of the model prediction accuracy. The formula is as follows:

[0218]

[0219]

[0220] As attached Fig. 9 As shown, attached Fig. 9 The prediction accuracy comparison diagram in this embodiment is given in FIG. Fig. 9 It can be seen that for the large public building in the case, the RMSE and MAPE of the TSA-RBF&LSTM combined neural network prediction model described in this embodiment are 2.8735 and 0.2% respectively, which are the smallest compared with other methods, and its prediction accuracy is higher; in addition, from the attached Fig. 9 It can be seen that compared with the RBF model, the prediction accuracy of the TSA-RBF model is improved, indicating that the TSA algorithm can be used for parameter optimization of the RBF model.

[0221] In order to further verify the generalization performance of the TSA-RBF&LSTM combined neural network prediction model proposed in this embodiment, the TSA-RBF&LSTM combined neural network prediction model was used to predict the energy consumption of a large public building in the last four days of the six months from April to September. The energy consumption data of the remaining days of the month were used as training data for the model, and an experimental comparison was carried out with whether to use the Lasso algorithm and the CEEMDAN algorithm as independent variables.

[0222] As attached Fig.10 As shown, attached Fig.10 The generalization capability proof diagram in this embodiment is given; Fig.10 It can be seen that the TSA-RBF&LSTM combined neural network prediction model has a strong generalization ability, and has a good prediction effect on the energy consumption of different months, and the error range is stable in a very small range; at the same time, compared with the model without Lasso feature extraction, the prediction effect after feature extraction is better, indicating that it has successfully reduced the dimension of the original feature set, and the selected features can better establish the prediction model and effectively increase the prediction accuracy of the model; compared with the model without CEEMDAN decomposition, the prediction effect after CEEMDAN decomposition is better, indicating that it can effectively preprocess the nonlinear and non-stationary energy consumption data, thereby further improving the prediction accuracy.

[0223] The above embodiment is only one of the implementation methods that can realize the technical solution of the present invention. The scope of protection claimed by the present invention is not limited only to this embodiment, but also includes changes, replacements and other implementation methods that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention.

Claims

1. A method for predicting energy consumption of public buildings, characterized in that: The following steps are involved: Obtain data on the impact components of historical public building energy consumption; The Lasso algorithm is used to extract the features of the influencing factors of the energy consumption of historical public buildings, and the main influencing indicator data of the energy consumption of historical public buildings are obtained; The CEEMDAN algorithm is used to decompose the data of the influencing factors of energy consumption of historical public buildings, the intrinsic mode function components of linear steady state and the residual components; Using the approximate entropy theory, the intrinsic mode function components and residual components of the linear steady state are divided according to the complexity to obtain high-frequency components and low-frequency components. The RBF neural network is trained using the main influencing indicator data and low-frequency components of historical public building energy consumption, and the TSA algorithm is used to optimize the key parameters of the RBF neural network to obtain the optimized RBF neural network model. The main influencing indicator data and high-frequency components of historical public building energy consumption are used to train the LSTM neural network to obtain the trained LSTM prediction model; The optimized RBF neural network model and the trained LSTM prediction model are superimposed to obtain the TSA-RBF&LSTM combined neural network prediction model; The main influencing indicator data of the public building energy consumption to be predicted is used as the input of the TSA-RBF&LSTM combined neural network prediction model, and the prediction result of the public building energy consumption is obtained as the output; The process of optimizing the key parameters of RBF neural network using TSA algorithm. The key parameters of RBF neural network include center vector, base width vector and weight vector. The main influencing index data of historical public building energy consumption obtained by feature extraction using the Lasso algorithm are combined with the low-frequency components respectively, and a training set and a test set are constructed; the RBF neural network is trained and optimized using the training set and the test set; in the training optimization process, the TSA algorithm is used to optimize the center vector, base width vector and weight vector of the RBF neural network to obtain the optimized RBF model; The TSA algorithm is used for the optimization process, which specifically includes the following steps: Determine the key parameters of the RBF neural network and determine the parameter range of the key parameters of the RBF neural network; wherein, each component in the center vector is set to a random number within [-3,3]; each component of the base width vector is set to a random number within [0,3]; each component of the weight vector is set to a random number within [-1,1]; According to the parameter range of the key parameters of the RBF neural network, the optimization range of the influencing parameters of the TSA algorithm is set; the influencing parameters of the TSA algorithm include population size, search trend parameters, optimization dimension, maximum number of iterations, upper and lower limits of tree species population positions, positions of randomly initialized tree species populations, and the fitness function of the TSA algorithm; The fitness function of the TSA algorithm is set to the root mean square error; the root mean square error calculation formula is as follows: Among them, fitness is the fitness function; n1 is the number of test samples, is the true value of the response variable, is the predicted value of the response variable; The prediction value of the RBF neural network is obtained by using the low-frequency component training set and the tree species population information; and according to the prediction value of the RBF neural network, the fitness values ​​of all individuals in the current population are calculated, the individual with the current best fitness is selected, and the position of the individual is set as the current best; The control parameter search trend ST of the TSA algorithm is used to iteratively update the position of the individual seed. When the random number generated in [0,1] is less than ST, the position of the seed's mother tree, the current best tree, and another random tree is updated; otherwise, the position of the seed's mother tree and another random tree is updated. After each iterative update, the key parameters of the optimized RBF neural network are transferred to the RBF neural network model; Determine whether the number of iterative cycles reaches the preset value. If so, stop optimizing the prediction accuracy parameters to obtain the optimal parameters of the RBF neural network model; if not, continue iterative updating.

2. A method for predicting energy consumption of public buildings according to claim 1, characterized in that: The data on the influencing factors of public building energy consumption include indoor lighting usage data of buildings, indoor personnel data, outdoor dry-bulb temperature, outdoor wet-bulb temperature, outdoor relative humidity, outdoor dew point temperature, wind speed, wind direction, historical current solar irradiance, solar irradiance at the previous historical moment and public building energy consumption at the previous historical moment.

3. A method for predicting energy consumption of public buildings according to claim 1, characterized in that: The Lasso algorithm is used to extract features from the data of influencing components of historical public building energy consumption. In the process of obtaining the main influencing indicator data of historical public building energy consumption, the data of influencing components of historical public building energy consumption whose coefficient vector of the Lasso regression optimization target model is not zero are used as the main influencing indicator data of historical public building energy consumption.

4. A method for predicting energy consumption of public buildings according to claim 1, characterized in that: The main influencing indicator data of historical public building energy consumption include building indoor lighting usage data, indoor personnel data, outdoor dry-bulb temperature, outdoor relative humidity, wind speed, historical current solar irradiance, solar irradiance at the previous moment in history, and public building energy consumption at the previous moment in history.

5. A method for predicting energy consumption of public buildings according to claim 1, characterized in that: The expression of the optimized RBF neural network model is: in, is the output of the optimized RBF neural network model; j is the jth neuron; m is the number of neurons in the hidden layer; w j is the optimized output weight of the jth node in the hidden layer; X is the input vector; C j is the optimized center vector of the jth node in the hidden layer; b j is the optimized basis width parameter of the jth node in the hidden layer.

6. A method for predicting energy consumption of public buildings according to claim 1, characterized in that: The expression of the trained LSTM prediction model is: f t =sigmoid(b f +W f,h *h t-1 +W f,x *x t ) i t =sigmoid(b i +W i,h *h t-1 +W i,x *X t ) o t =sigmoid(b o +W o,h *h t-1 +W o,x *x t ) h t =o t *tanh(c t ) Where t is an arbitrary time step; f t is the output of the forget gate at time t; i t is the output of the input gate at time t; is information used to update the cell status; c t is the cell state at time t; o t is the output of the output gate at time t; h t are the outputs of the hidden layer at time t; b f , W f,h , W f,x are the coefficients corresponding to the forget gates; b i , W i,h , W i,x are the coefficients corresponding to the input gates; b c , W c,h , W c,x is the coefficient corresponding to the cell state; b o , W o,h , W o,x are the coefficients corresponding to the output gates; h t-1 is the output of the hidden layer at time t-1; x t is the input of the LSTM prediction model; sigmoid and tanh are activation functions respectively.

7. A public building energy consumption prediction system, characterized in that: include: A data acquisition module is used to obtain data on the influencing components of historical public building energy consumption; The feature extraction module is used to extract the features of the data of the influencing elements of the energy consumption of historical public buildings by using the Lasso algorithm, and obtain the main influencing indicator data of the energy consumption of historical public buildings; Decomposition module, used to decompose the data of the influencing elements of energy consumption of historical public buildings using CEEMDAN algorithm, the intrinsic mode function components of linear steady state and the residual components; A partitioning module is used to partition the intrinsic mode function components and residual components of the linear steady state according to the complexity by using the approximate entropy theory to obtain high-frequency components and low-frequency components; The RBF model module is used to train the RBF neural network using the main influencing indicator data and low-frequency components of historical public building energy consumption, and to optimize the key parameters of the RBF neural network using the TSA algorithm to obtain the optimized RBF neural network model; The LSTM model module is used to train the LSTM neural network using the main influencing indicator data and high-frequency components of historical public building energy consumption to obtain a trained LSTM prediction model; The model superposition module is used to superimpose the optimized RBF neural network model and the trained LSTM prediction model to obtain the TSA-RBF&LSTM combined neural network prediction model; A prediction output module is used to use the main influencing indicator data of the public building energy consumption to be predicted as the input of the TSA-RBF&LSTM combined neural network prediction model, and output the prediction result of the public building energy consumption; The process of optimizing the key parameters of RBF neural network using TSA algorithm. The key parameters of RBF neural network include center vector, base width vector and weight vector. The main influencing index data of historical public building energy consumption obtained by feature extraction using the Lasso algorithm are combined with the low-frequency components respectively, and a training set and a test set are constructed; the RBF neural network is trained and optimized using the training set and the test set; in the training optimization process, the TSA algorithm is used to optimize the center vector, base width vector and weight vector of the RBF neural network to obtain the optimized RBF model; The TSA algorithm is used for the optimization process, which specifically includes the following steps: Determine the key parameters of the RBF neural network and determine the parameter range of the key parameters of the RBF neural network; wherein, each component in the center vector is set to a random number within [-3,3]; each component of the base width vector is set to a random number within [0,3]; each component of the weight vector is set to a random number within [-1,1]; According to the parameter range of the key parameters of the RBF neural network, the optimization range of the influencing parameters of the TSA algorithm is set; the influencing parameters of the TSA algorithm include population size, search trend parameters, optimization dimension, maximum number of iterations, upper and lower limits of tree species population positions, positions of randomly initialized tree species populations, and the fitness function of the TSA algorithm; The fitness function of the TSA algorithm is set to the root mean square error; the root mean square error calculation formula is as follows: Among them, fitness is the fitness function; n1 is the number of test samples, is the true value of the response variable, is the predicted value of the response variable; The prediction value of the RBF neural network is obtained by using the low-frequency component training set and the tree species population information; and according to the prediction value of the RBF neural network, the fitness values ​​of all individuals in the current population are calculated, the individual with the current best fitness is selected, and the position of the individual is set as the current best; The control parameter search trend ST of the TSA algorithm is used to iteratively update the position of the individual seed. When the random number generated in [0,1] is less than ST, the position of the seed's mother tree, the current best tree, and another random tree is updated; otherwise, the position of the seed's mother tree and another random tree is updated. After each iterative update, the key parameters of the optimized RBF neural network are transferred to the RBF neural network model; Determine whether the number of iterative cycles reaches the preset value. If so, stop optimizing the prediction accuracy parameters to obtain the optimal parameters of the RBF neural network model; if not, continue iterative updating.

8. A public building energy consumption prediction device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of a method for predicting energy consumption of a public building as described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of a method for predicting energy consumption of a public building as described in any one of claims 1 to 6 are implemented.