A multi-element load prediction method and device for an integrated energy system

Abstract

The application relates to a comprehensive energy system multi-element load prediction method and device, which comprises the following steps: obtaining historical meteorological data and cold-heat-electricity load data of a comprehensive energy system, and performing data preprocessing; performing standardization processing on the data in a database, and dividing the data into a training set, a verification set and a test set according to a preset proportion; calculating the Spearman correlation coefficient between cold-heat-electricity loads in each season, and establishing a neural network load prediction model of a target season according to the Spearman correlation coefficient; and training the model based on the training set; and inputting the data in the test set into the trained neural network load prediction model, and outputting a cold-heat-electricity load prediction result. The comprehensive energy system multi-element load prediction method fully considers the seasonality, nonlinearity and coupling of multi-element loads, extracts and shares coupling information between loads from different scales, and establishes a more refined prediction model for each season.

Description

Technical Field

[0001] This invention relates to the field of integrated energy system technology, and in particular to a method and apparatus for predicting multiple loads in an integrated energy system. Background Technology

[0002] As a new energy supply model integrating multiple energy carriers such as electricity, cooling, and heat, integrated energy systems have become an effective solution to the problem of energy resource shortages. Through various energy conversion and storage devices, different energy flows can be mutually converted and supplemented, greatly improving energy utilization efficiency. Accurate load forecasting helps energy suppliers anticipate user demand, promotes the economical and low-carbon operation of integrated energy systems, and improves system response speed. However, compared with traditional single-energy systems, the more frequent and complex interactions of multiple energy sources in integrated energy systems bring significant difficulties and challenges to accurate load forecasting.

[0003] In recent years, with the development of artificial intelligence technology, deep learning has gradually become dominant in the field of load forecasting. Existing deep learning-based multi-load forecasting technology for integrated energy systems has the following defects: (1) Many forecasting models do not fully consider the coupling characteristics between multiple loads in the integrated energy system, cannot mine and share the coupling feature information between strongly correlated loads from multiple scales, and the information sharing layer in some forecasting models will reduce the computational efficiency of the system; (2) Many existing forecasting models ignore the seasonal characteristics of multiple loads in the integrated energy system and do not analyze the changes in the coupling relationship between multiple loads with the seasons, resulting in an insufficiently refined forecasting model; (3) Existing joint forecasting loss function weight optimization methods are difficult to balance the loss magnitude and training speed of each task during training. Summary of the Invention

[0004] In view of this, it is necessary to provide a multi-dimensional load forecasting method for integrated energy systems to address the aforementioned deficiencies of existing technologies.

[0005] To address the aforementioned problems, in a first aspect, embodiments of the present invention provide a multi-source load forecasting method for integrated energy systems, comprising:

[0006] Historical meteorological data and cooling, heating and power load data of the integrated energy system are acquired, data preprocessing is performed, and the preprocessed data is stored in the database;

[0007] The data in the database is standardized and divided into training set, validation set and test set according to a preset ratio;

[0008] Calculate the Spearman correlation coefficients between cooling, heating, and electrical loads in each season, and establish a neural network load prediction model for the target season based on the Spearman correlation coefficients. Train the model based on the training set.

[0009] The data from the test set is input into the trained neural network load prediction model, which outputs the prediction results of cooling, heating and electricity loads.

[0010] Preferably, the historical meteorological data includes temperature, humidity, air pressure, wind speed, dew point, and precipitation; the cooling, heating, and electrical load data includes cooling load, heating load, and electrical load.

[0011] The data preprocessing steps include:

[0012] Outliers were removed using the quartile method.

[0013] Missing data values ​​are filled using a linear interpolation method.

[0014] Preferably, the standardization process for the data in the database includes:

[0015] The data in the database is divided into four parts according to the season;

[0016] The data for each season are standardized separately; the expression for standardization is:

[0017]

[0018] In the formula, x and x' are the sample values ​​before and after standardization, respectively, and μ and σ are the mean and standard deviation of the samples in the training set, respectively.

[0019] Preferably, the step of calculating the Spearman correlation coefficients between cooling, heating, and electrical loads in each season, and establishing a neural network load prediction model for the target season based on the Spearman correlation coefficients, includes:

[0020] The formula for calculating the Spearman correlation coefficient is as follows:

[0021]

[0022] In the formula, d i =rg(x i )-rg(y i ); rg(x i ) and rg(y i ) are respectively x i and y i The rank of the sample; n is the total number of samples; ρ s The closer the absolute value of x is to 1, the stronger the correlation between x and y; i and yi Let x and y represent the i-th elements in sequences x and y, respectively.

[0023] Based on the calculation results of the Spearman correlation coefficient, neural network load prediction models were built for the four seasons:

[0024] If the Spearman correlation coefficient of two of the loads in a certain season is greater than the preset threshold of 0.7, then the interactive multi-scale convolution module is used as the core prediction unit for joint prediction.

[0025] If the Spearman correlation coefficients of one load with the other two loads in the season are both less than the preset threshold of 0.7, then the multi-scale feature fusion module is used as the core prediction unit for separate prediction.

[0026] Preferably, the neural network load prediction model includes an interactive multi-scale convolution module, which consists of a one-dimensional copy padding layer, a one-dimensional convolutional layer, a LeakyReLU activation function, a dropout layer, a one-dimensional convolutional layer, and a Tanh activation function.

[0027] The load prediction steps of the interactive multi-scale convolution module include:

[0028] For any two load sequences F from the input of heating, cooling, and electricity e With F c First, information sharing is coupled through multiple rounds of interactive learning: using two different one-dimensional convolutional modules ψ e and ψ c Load sequence F e With F c Mapped to F e1 and F c1 After exponential operation, F e1 and F c1 They interact through the Hadamard product. The above process can be represented by the following formula, where represents the Hadamard product operation;

[0029]

[0030] Then, through a second set of different one-dimensional convolutional modules φ e and φ c Load sequence and Mapped to F e2 and F c2 The second interaction is completed by adding or subtracting from each other:

[0031]

[0032] After completing multiple rounds of interactive learning, the third set of different one-dimensional convolutional modules η e and η c For the load sequence respectively and The process involves processing the input sequences; then, feature addition and self-attention mechanisms from the multi-scale feature fusion module are introduced to extract multi-scale features from the two load sequences; finally, a residual network is introduced to add the output features of the self-attention mechanism to the original input sequence to obtain the final output F' of the interactive multi-scale convolution module. e and F' c .

[0033] Preferably, when using an interactive multi-scale convolutional module as the core prediction unit for joint prediction, a loss function weight optimization method based on Laplace distribution is adopted to dynamically adjust the weights of the loss function of each task during training.

[0034] Preferably, the formula for calculating the total loss function L of the neural network load prediction model is:

[0035]

[0036] In the formula, Let be the average loss of task k in the t-th training round, K be the total number of tasks, W be the parameters in the neural network, T be an adjustable hyperparameter, and λ be the value of λ. k Output noise for the model of task k.

[0037] Preferably, after inputting the data from the test set into the trained neural network load prediction model, the method further includes:

[0038] The output data of the neural network load prediction model is denormalized to obtain the prediction results of cooling, heating and electricity loads.

[0039] Secondly, embodiments of the present invention provide a multi-source load forecasting device for an integrated energy system, comprising:

[0040] The acquisition module is used to acquire historical meteorological data and cooling, heating and power load data of the integrated energy system, perform data preprocessing, and store the preprocessed data in the database.

[0041] The standardization processing module is used to standardize the data in the database and divide it into training set, validation set and test set according to a preset ratio;

[0042] The model building module is used to calculate the Spearman correlation coefficient between cooling, heating and power loads in each season, and to build a neural network load prediction model for the target season based on the Spearman correlation coefficient, and to train the model based on the training set.

[0043] The load forecasting module is used to input data from the test set into the trained neural network load forecasting model and output the predicted results of cooling, heating and electricity loads.

[0044] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein,

[0045] The memory is used to store programs;

[0046] The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the integrated energy system multi-load forecasting method as described in the first aspect embodiment of the present invention.

[0047] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instructions, which, when executed by a processor, can implement the steps in the integrated energy system multi-load forecasting method as described in the first aspect embodiment of the present invention.

[0048] The integrated energy system multi-load forecasting method and apparatus provided by this invention fully considers the seasonality, nonlinearity, and coupling of multi-loads. It can extract and share coupling information between loads at different scales, establishing a more refined forecasting model for each season. For joint forecasting, the proposed Laplace distribution-based loss function weight optimization method can more robustly balance the loss magnitude and training speed of each task simultaneously, and avoid the occurrence of negative losses. This invention improves the accuracy of multi-load forecasting for integrated energy systems, compensates for the insufficient utilization of collected load big data, and provides data support for the subsequent economical and low-carbon operation of integrated energy systems. Attached Figure Description

[0049] Figure 1 Flowchart of the integrated energy system multi-element load forecasting method provided by the present invention;

[0050] Figure 2 Spearman correlation coefficient heatmap of the integrated energy system for different seasons provided by the present invention;

[0051] Figure 3 This is a structural diagram of the interactive multi-scale convolution module provided by the present invention;

[0052] Figure 4 The present invention provides four possible neural network load prediction model structure diagrams;

[0053] Figure 5 Structural block diagram of the integrated energy system multi-load forecasting device provided by the present invention;

[0054] Figure 6This is a structural block diagram of the electronic device provided by the present invention. Detailed Implementation

[0055] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0056] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0057] The existing deep learning-based multi-load forecasting technology for integrated energy systems has the following defects: (1) Many forecasting models do not fully consider the coupling characteristics between multi-loads in integrated energy systems, cannot mine and share the coupling feature information between strongly correlated loads from multiple scales, and the information sharing layer in some forecasting models will reduce the computational efficiency of the system; (2) Many existing forecasting models ignore the seasonal characteristics of multi-loads in integrated energy systems and do not analyze the changes in the coupling relationship between multi-loads with the seasons, resulting in the forecasting model being not refined enough; (3) Existing joint forecasting loss function weight optimization methods are difficult to balance the loss magnitude and training speed of each task during training.

[0058] In view of this, the present invention provides a multi-load forecasting method for integrated energy systems, which fully considers the seasonality, nonlinearity, and coupling of multi-loads, extracts and shares coupling information between loads from different scales, and establishes a more refined forecasting model for each season. The following will elaborate and describe this method through several embodiments.

[0059] Figure 1 The flowchart illustrates the multi-element load forecasting method for integrated energy systems provided by this invention. Figure 1 As shown, the integrated energy system multi-element load forecasting method includes steps S1 to S4, wherein:

[0060] Step S1: Obtain historical meteorological data and cooling, heating and power load data of the integrated energy system, perform data preprocessing, and store the preprocessed data in the database.

[0061] Specifically, historical meteorological data includes temperature, humidity, air pressure, wind speed, dew point, and precipitation. Cooling, heating, and electrical load data includes cooling load, heating load, and electrical load.

[0062] In this embodiment, the data preprocessing step includes:

[0063] Outliers were removed using the quartile method.

[0064] Missing data values ​​are filled using a linear interpolation method.

[0065] Step S2: Standardize the data in the database and divide it into training set, validation set and test set according to a preset ratio.

[0066] Specifically, in this embodiment, the data in the database is divided into four parts according to the season, which can be further divided into four seasonal databases. Among them, March to May is spring; June to August is summer; September to November is autumn; and December to February is winter.

[0067] The data for each seasonal database were standardized separately; the expression for standardization is:

[0068]

[0069] In the formula, x and x' are the sample values ​​before and after standardization, respectively, and μ and σ are the mean and standard deviation of the samples in the training set, respectively.

[0070] After standardizing the data, it is divided into training set, validation set and test set according to a preset ratio. Preferably, the ratio can be set to 8:1:1.

[0071] Step S3: Calculate the Spearman correlation coefficient between cooling, heating and electricity loads in each season, and establish a neural network load prediction model for the target season based on the Spearman correlation coefficient. Train the model based on the training set.

[0072] The Spearman correlation coefficient, named after Charles Edward Spearman, is a Spearman rank correlation coefficient used to measure the correlation between two variables without assuming a linear relationship or that the data follows any particular distribution. It is calculated based on the data rank (i.e., the position of the data after sorting), rather than on the original numerical value.

[0073] In step S3, the Spearman correlation coefficient between the cooling, heating, and electrical loads of each season is calculated. The formula for calculating the Spearman correlation coefficient is as follows:

[0074]

[0075] In the formula, d i =rg(x i )-rg(y i ); rg(x i ) and rg(yi ) are respectively x i and y i The rank of the sample; n is the total number of samples; ρ s The closer the absolute value is to 1, the stronger the correlation between x and y.

[0076] Figure 2 The Spearman correlation coefficient heatmap of the integrated energy system provided by this invention, relating to the cooling, heating, and electrical loads in different seasons, is derived from... Figure 2 It can be seen that in spring, the Spearman correlation coefficients between electrical load and cooling load, as well as between cooling load and heating load, exceeded the preset threshold of 0.7; in summer, only the Spearman correlation coefficient between electrical load and cooling load exceeded the preset threshold of 0.7; in autumn, the Spearman correlation coefficients between all three loads exceeded the preset threshold of 0.7; and in winter, only the Spearman correlation coefficient between cooling load and heating load exceeded the preset threshold of 0.7.

[0077] The neural network load prediction model established in this embodiment of the invention includes an interactive multi-scale convolution module. Figure 3 The diagram below shows the structure of the interactive multi-scale convolution module provided by this invention. Figure 3 The interactive multi-scale convolutional module consists of a one-dimensional copy padding layer, a one-dimensional convolutional layer, a LeakyReLU activation function, a dropout layer, another one-dimensional convolutional layer, and a Tanh activation function. The load prediction steps of the interactive multi-scale convolutional module include:

[0078] For any two load sequences F from the input of heating, cooling, and electricity e With F c First, information sharing is coupled through multiple rounds of interactive learning: using two different one-dimensional convolutional modules ψ e and ψ c Load sequence F e With F c Mapped to F e1 and F c1 After exponential calculation, F e1 and F c1 They interact through the Hadamard product. The above process can be represented by the following formula, where represents the Hadamard product operation;

[0079]

[0080] Then, through a second set of different one-dimensional convolutional modules φ e and φ c Load sequence and Mapped to F e2 and F c2The second interaction is completed by adding or subtracting from each other:

[0081]

[0082] After completing multiple rounds of interactive learning, the third set of different one-dimensional convolutional modules η e and η c For the load sequence respectively and The process involves several steps: first, feature addition and self-attention mechanisms from the multi-scale feature fusion module are introduced to extract multi-scale features from the two load sequences; finally, a residual network is introduced to add the output features of the self-attention mechanism to the original input sequence to obtain the final output F' of the interactive multi-scale convolution module. e and F' c .

[0083] In this embodiment, based on the calculated Spearman correlation coefficient, neural network load prediction models are built for each of the four seasons: if the Spearman correlation coefficient of two loads in a certain season is greater than a preset threshold of 0.7, then an interactive multi-scale convolutional module is used as the core prediction unit for joint prediction. If the Spearman correlation coefficient of one load with the other two loads in a season is less than the preset threshold of 0.7, then a multi-scale feature fusion module is used as the core prediction unit for individual prediction. This leads to the following... Figure 4 The diagram shows the structure of four possible neural network load prediction models. Figure 4 The diagram shows the structure of four possible neural network load prediction models provided by this invention.

[0084] Depend on Figure 2 The Spearman correlation coefficient heatmap showing the seasonal loads for cooling, heating, and electricity is presented in Table 1. The selection results for the integrated energy system's seasonal load forecasting model are shown in Table 1.

[0085]

[0086] In this embodiment, when an interactive multi-scale convolution module is used as the core prediction unit for joint prediction, the present invention adopts a loss function weight optimization method based on Laplace distribution to dynamically adjust the weights of the loss function of each task during training.

[0087] The derivation process of the loss function weight optimization method based on the Laplace distribution will be explained below.

[0088] Let f W Let f(x) be the output of the prediction model with x as input and W as model parameters. Then the likelihood between f(x) and the observed value y is defined by the following Laplace distribution:

[0089]

[0090] wherein λ represents noise, which corresponds to homoscedastic uncertainty of the current task. p(y丨f W (x)) denotes the conditional probability distribution of observed data y given the model f W (x), and Lp(y丨f W (x), λ) is the log-likelihood function.

[0091] According to maximum likelihood estimation, the parameters in the Laplace distribution are obtained by maximizing the log-likelihood of the model with respect to W and λ:

[0092]

[0093] Taking the joint prediction of three types of loads as an example, it is assumed that the three outputs y1, y2 and y3 of the model all follow the Laplace distribution:

[0094] p(y1,y2,y3|f W (x))

[0095] =p(y1|f W (x))·p(y2|f W (x))·p(y3|f W (x))

[0096] =Lp(y1|f W (x),λ1)·Lp(y2|f W (x),λ2)·Lp(y3|f W (x),λ3)

[0097] Then the loss function of the neural network load prediction model is equivalent to minimizing the following negative log-likelihood:

[0098]

[0099] wherein L1(W)=|y1-f W (x)| is the loss of the first task. λ k is the model output noise for task k, which is used to learn and determine the weight of L k (W).

[0100] In the above formula, the loss function of each task is derived from Laplace likelihood, which has a linear relationship with the absolute error, so that the model exhibits stronger robustness when facing outliers in the dataset. More generally, for K joint prediction tasks, we have:

[0101]

[0102] log2λk Its function is to act as a regularization term to avoid noise λ. K Excessive growth, however, when λ k When the value is less than 0.5, it may lead to a negative loss value. Therefore, this invention uses log2λ... k With slight modifications, the following formula is obtained:

[0103]

[0104] Furthermore, this invention also introduces another weight. To balance the training speed of each task, the total loss function L of the neural network load prediction model is calculated by the following formula:

[0105]

[0106] In the formula, Let be the average loss of task k in the t-th training round, K be the total number of tasks, W be the parameters in the neural network, T be an adjustable hyperparameter, and λ be the value of λ. k Output noise for the model of task k. Let be the ratio of the average loss of task k in the (t-1)th and (t-2)th training rounds. Let be the ratio of the average loss of task i in the (t-1)th and (t-2)th training rounds, i∈[1,K]; As weight.

[0107] By learning and calculating the homoscedastic uncertainty and average loss variation related to the task, the Laplace distribution-based loss function weight optimization method proposed in this invention can simultaneously balance the loss magnitude and training speed for each task. Furthermore, the L1-norm loss function, which is linearly related to the absolute error, can reduce the impact of outliers on the estimation results.

[0108] Step S4: Input the data from the test set into the trained neural network load prediction model and output the cooling, heating and electrical load prediction results.

[0109] Specifically, after inputting the data from the test set into the trained neural network load prediction model, the output data of the neural network load prediction model is destandardized to obtain the prediction results of cooling, heating and electricity loads.

[0110] In this embodiment, in order to verify the superiority and effectiveness of the multi-dimensional load forecasting method for the integrated energy system, the present invention was compared with two classic machine learning models: Gradient Boosting Regression (GBR) and Artificial Neural Network (ANN), as well as two commonly used deep learning models based on recurrent neural network architecture: Long Short-Term Memory Recurrent Neural Network (LSTM) and Bidirectional Long Short-Term Memory Recurrent Neural Network (BiLSTM) on the PyTorch platform. The MAPE performance index of each model on the test set was calculated as shown in Table 2.

[0111] Table 2

[0112]

[0113]

[0114] The formula for calculating the MAPE performance metric is as follows:

[0115]

[0116] In the formula, n is the total number of samples, and y′ i and y i These are the predicted and actual load values, respectively.

[0117] Referring to Table 2, this invention performs best in 11 out of 12 load forecasting tasks, and achieves the greatest improvement in forecast accuracy during the autumn season when load fluctuations are strongest. Further calculations show that this invention improves average accuracy by 7.82% compared to BiLSTM, 6.28% compared to LSTM, 15.79% compared to ANN, and 22.49% compared to GBR.

[0118] The multi-load forecasting method for integrated energy systems provided by this invention fully considers the seasonality, nonlinearity, and coupling of multiple loads. It can extract and share coupling information between loads from different scales, establishing a more refined forecasting model for each season. For joint forecasting, the proposed Laplace distribution-based loss function weight optimization method can more robustly balance the loss magnitude and training speed of each task simultaneously, and avoid the occurrence of negative losses. This invention improves the accuracy of multi-load forecasting for integrated energy systems, compensates for the deficiency in utilizing the collected load big data, and provides data support for the subsequent economical and low-carbon operation of integrated energy systems.

[0119] Figure 5 The structural block diagram of the integrated energy system multi-element load forecasting device provided by the present invention is shown below. Figure 5 The integrated energy system multi-load forecasting device includes:

[0120] The acquisition module 501 is used to acquire historical meteorological data and cooling, heating and power load data of the integrated energy system, perform data preprocessing, and store the preprocessed data in the database.

[0121] The standardization processing module 502 is used to standardize the data in the database and divide it into training set, validation set and test set according to a preset ratio;

[0122] The model building module 503 is used to calculate the Spearman correlation coefficient between cooling, heating and power loads in each season, and to build a neural network load prediction model for the target season based on the Spearman correlation coefficient, and to train the model based on the training set.

[0123] The load prediction module 504 is used to input the data from the test set into the trained neural network load prediction model and output the prediction results of cooling, heating and electricity loads.

[0124] The integrated energy system multi-load forecasting device provided by the present invention executes the integrated energy system multi-load forecasting method provided in the above embodiments through the above modules. The integrated energy system multi-load forecasting method has been described in detail in the above embodiments, and will not be repeated here.

[0125] The integrated energy system multi-load forecasting device provided by this invention fully considers the seasonality, nonlinearity, and coupling of multi-loads, and can extract and share coupling information between loads from different scales to establish a more refined forecasting model for each season. For joint forecasting, the proposed loss function weight optimization method based on Laplace distribution can more robustly balance the loss magnitude and training speed of each task simultaneously, and avoid the occurrence of negative loss values.

[0126] Figure 6 A structural block diagram of the electronic device provided by the present invention, such as Figure 6 As shown, the present invention also provides an electronic device 600, which can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing device. The electronic device 600 includes a processor 601 and a memory 602, wherein the memory 602 stores a multi-load forecasting program 606 for an integrated energy system.

[0127] In some embodiments, memory 602 may be an internal storage unit of a computer device, such as a hard disk or memory. In other embodiments, memory 602 may be an external storage device of a computer device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, memory 602 may include both internal and external storage units of the computer device. Memory 602 is used to store application software and various types of data installed on the computer device, such as program code installed on the computer device. Memory 602 can also be used to temporarily store data that has been output or will be output. In one embodiment, when the integrated energy system multi-load forecasting program 606 is executed by processor 601, the following steps are implemented:

[0128] Historical meteorological data and cooling, heating and power load data of the integrated energy system are acquired, data preprocessing is performed, and the preprocessed data is stored in the database;

[0129] The data in the database is standardized and divided into training set, validation set and test set according to a preset ratio;

[0130] Calculate the Spearman correlation coefficients between cooling, heating, and electrical loads in each season, and establish a neural network load prediction model for the target season based on the Spearman correlation coefficients. Train the model based on the training set.

[0131] The data from the test set is input into the trained neural network load prediction model, which outputs the prediction results of cooling, heating and electricity loads.

[0132] In some embodiments, processor 601 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 602 or process data, such as executing a multi-load forecasting program for an integrated energy system.

[0133] This embodiment also provides a computer-readable storage medium storing a multi-load forecasting program for an integrated energy system. When executed by a processor, the multi-load forecasting program for an integrated energy system performs the following steps:

[0134] Historical meteorological data and cooling, heating and power load data of the integrated energy system are acquired, data preprocessing is performed, and the preprocessed data is stored in the database;

[0135] The data in the database is standardized and divided into training set, validation set and test set according to a preset ratio;

[0136] Calculate the Spearman correlation coefficients between cooling, heating, and electrical loads in each season, and establish a neural network load prediction model for the target season based on the Spearman correlation coefficients. Train the model based on the training set.

[0137] The data from the test set is input into the trained neural network load prediction model, which outputs the prediction results of cooling, heating and electricity loads.

[0138] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-element load forecasting method for integrated energy systems, characterized in that, include: Historical meteorological data and cooling, heating and power load data of the integrated energy system are acquired, data preprocessing is performed, and the preprocessed data is stored in the database; The data in the database is standardized and divided into training set, validation set and test set according to a preset ratio; Calculate the Spearman correlation coefficients between cooling, heating, and electrical loads in each season, and establish a neural network load prediction model for the target season based on the Spearman correlation coefficients. Train the model based on the training set. If the Spearman correlation coefficient of two loads in a certain season is greater than a preset threshold, then an interactive multi-scale convolution module is used as the core prediction unit for joint prediction. The load prediction steps of the interactive multi-scale convolution module include: For any two load sequences from the input cooling, heating, and electricity... and First, information sharing is coupled through multiple rounds of interactive learning: using two different one-dimensional convolutional modules. and Load sequence and Mapped to and After exponential calculation, and They interact through the Hadamard product; the above process can be represented by the following formula, where... This represents the Hadamard product operation; ； ； Then, through a second set of different one-dimensional convolutional modules and Load sequence and Mapped to and The second interaction is completed by adding or subtracting from each other: ； ； After completing multiple rounds of interactive learning, a third set of different one-dimensional convolutional modules was used. and For the load sequence respectively and The process involves several steps: first, feature addition and self-attention mechanisms from the multi-scale feature fusion module are introduced to extract multi-scale features from the two load sequences; finally, a residual network is introduced to add the output features of the self-attention mechanism to the original input sequence to obtain the final output of the interactive multi-scale convolution module. and ; The data from the test set is input into the trained neural network load prediction model, which outputs the prediction results of cooling, heating and electricity loads.

2. The multi-source load forecasting method for integrated energy systems according to claim 1, characterized in that, The historical meteorological data includes temperature, humidity, air pressure, wind speed, dew point, and precipitation; the cooling, heating, and electrical load data includes cooling load, heating load, and electrical load. The data preprocessing steps include: Outliers were removed using the quartile method. Missing data values ​​are filled using a linear interpolation method.

3. The integrated energy system multi-source load forecasting method according to claim 1, characterized in that, The standardization process for the data in the database includes: The data in the database is divided into four parts according to the season; The data for each season are standardized separately; the expression for standardization is: ； In the formula, and These are the sample values ​​before and after standardization, respectively. and These are the mean and standard deviation of the samples in the training set, respectively.

4. The integrated energy system multi-source load forecasting method according to claim 2, characterized in that, The calculation of the Spearman correlation coefficients between cooling, heating, and electrical loads in each season, and the establishment of a neural network load prediction model for the target season based on the Spearman correlation coefficients, includes: The formula for calculating the Spearman correlation coefficient is as follows: ； In the formula, ; and They are respectively and rank; It is the total number of samples; The closer the absolute value is to 1, the stronger it indicates. and The stronger the correlation; Based on the calculation results of the Spearman correlation coefficient, neural network load prediction models were built for the four seasons: If the Spearman correlation coefficient of two of the loads in a certain season is greater than a preset threshold, then an interactive multi-scale convolution module is used as the core prediction unit for joint prediction. If the Spearman correlation coefficients of one load with the other two loads in the season are both less than a preset threshold, then the multi-scale feature fusion module is used as the core prediction unit for separate prediction.

5. The multi-source load forecasting method for integrated energy systems according to claim 4, characterized in that, The neural network load prediction model includes an interactive multi-scale convolution module, which consists of a one-dimensional copy padding layer, a one-dimensional convolutional layer, a LeakyReLU activation function, a dropout layer, a one-dimensional convolutional layer, and a Tanh activation function.

6. The multi-source load forecasting method for integrated energy systems according to claim 4, characterized in that, When using interactive multi-scale convolutional modules as core prediction units for joint prediction, a loss function weight optimization method based on Laplace distribution is adopted to dynamically adjust the weights of the loss function of each task during training. The total loss function of the neural network load prediction model The calculation formula is: ； ； ； In the formula, For the task In the Average loss over training rounds For the total number of tasks, For parameters in a neural network, These are adjustable hyperparameters. For the task The model output noise, As weight.

7. The integrated energy system multi-source load forecasting method according to claim 1, characterized in that, After inputting the data from the test set into the trained neural network load prediction model, the method further includes: The output data of the neural network load prediction model is denormalized to obtain the prediction results of cooling, heating and electricity loads.

8. A multi-element load forecasting device for an integrated energy system, characterized in that, include: The acquisition module is used to acquire historical meteorological data and cooling, heating and power load data of the integrated energy system, perform data preprocessing, and store the preprocessed data in the database. The standardization processing module is used to standardize the data in the database and divide it into training set, validation set and test set according to a preset ratio; The model building module is used to calculate the Spearman correlation coefficient between cooling, heating and power loads in each season, and to build a neural network load prediction model for the target season based on the Spearman correlation coefficient, and to train the model based on the training set. If the Spearman correlation coefficient of two loads in a certain season is greater than a preset threshold, then an interactive multi-scale convolution module is used as the core prediction unit for joint prediction. The load prediction steps of the interactive multi-scale convolution module include: For any two load sequences from the input cooling, heating, and electricity... and First, information sharing is coupled through multiple rounds of interactive learning: using two different one-dimensional convolutional modules. and Load sequence and Mapped to and After exponential calculation, and They interact through the Hadamard product; the above process can be represented by the following formula, where... This represents the Hadamard product operation; ； ； Then, through a second set of different one-dimensional convolutional modules and Load sequence and Mapped to and The second interaction is completed by adding or subtracting from each other: ； ； After completing multiple rounds of interactive learning, a third set of different one-dimensional convolutional modules was used. and For the load sequence respectively and The process involves several steps: first, feature addition and self-attention mechanisms from the multi-scale feature fusion module are introduced to extract multi-scale features from the two load sequences; finally, a residual network is introduced to add the output features of the self-attention mechanism to the original input sequence to obtain the final output of the interactive multi-scale convolution module. and ; The load forecasting module is used to input data from the test set into the trained neural network load forecasting model and output the predicted results of cooling, heating and electricity loads.

9. An electronic device, Its features are, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the integrated energy system multi-load forecasting method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the integrated energy system multi-load forecasting method according to any one of claims 1 to 7.

Images