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

Abstract

The application discloses a kind of comprehensive energy system multivariate load prediction method, comprising: obtaining comprehensive energy system data, according to the correlation analysis of each meteorological factor and load to comprehensive energy system data, and strong correlation history meteorological factor data is screened out;Build multivariate load coupling feature matrix;According to meteorological factor data, multivariate load coupling feature matrix is clustered using adaptive K-means clustering algorithm;The multivariate load coupling feature matrix after clustering is input into the Attention-BiGRU prediction model pre-trained to predict, and the load prediction result is obtained.Considering meteorological factors and the cross-coupling characteristics of multivariate load, the model complexity and prediction time are reduced, so that the final prediction result takes into account stability and accuracy.

Description

Technical Field

[0001] This invention belongs to the field of integrated energy system forecasting technology, specifically relating to a multi-element load forecasting method for integrated energy systems. Background Technology

[0002] Against the backdrop of the energy internet and low-carbon living, the Integrated Energy System (IES) breaks away from the traditional energy system where each energy source is independent during design and operation. It achieves the integration of energy production, supply and sales from generation to consumption, and realizes the goal of mutual conversion and unified management of multiple energy networks through energy conversion equipment. Among these, accurate and efficient load forecasting is an important prerequisite for the operation planning and optimized scheduling of the entire integrated energy system.

[0003] As an integrated energy system encompassing production, supply, and sales, changes in meteorological factors can alter users' energy consumption behavior, thereby affecting diverse load curves. Besides common factors like temperature and humidity, other meteorological factors cannot be ignored.

[0004] Furthermore, an integrated energy system is not simply a superposition of multiple energy supply systems, but rather a collaborative control and optimization of various energy supply units. Its internal electrical, thermal, and cooling loads have strong coupling. Traditional single load forecasting methods treat each load forecasting task as an independent and simple sub-problem, failing to take into account the mutual conversion and influence relationships between different energy demands, resulting in a decrease in forecasting accuracy. Summary of the Invention

[0005] This invention analyzes the correlation between different meteorological factors and electricity, heat, and cooling loads, and selects meteorological factors with strong correlations as input features for load forecasting to more effectively improve the accuracy of load forecasting.

[0006] To address the decline in prediction accuracy, this invention constructs a multi-variable load coupling feature matrix to achieve multi-variable load collaborative prediction, thereby effectively improving the prediction accuracy and model stability of the system.

[0007] Objective: To address the aforementioned problems, this invention provides a multi-element load forecasting method for integrated energy systems based on K-means and Attention-BiGRU. This method considers the impact of meteorological factors on load, employs the grey relational analysis method to analyze the correlation between various meteorological factors and load, and selects strongly correlated meteorological factors as forecast inputs. Considering the coupling between multiple loads within the system, a multi-element load feature matrix reflecting the interactive influence of electricity, heat, and cooling loads is constructed. K-means clustering and Attention-BiGRU networks are used for multi-element load forecasting to improve the forecasting accuracy of multi-element loads.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] Firstly, a multi-source load forecasting method for integrated energy systems is provided, including:

[0010] Step S1: Obtain integrated energy system data, which includes historical electricity load data, historical heat load data, historical cooling load data, and meteorological data within the system.

[0011] Step S2: Based on the integrated energy system data, perform correlation analysis on various meteorological factors and loads, and screen out historical meteorological factor data with strong correlations from the meteorological data within the system;

[0012] Step S3: Based on the historical meteorological data, historical electricity load data, historical heat load data, and historical cooling load data, construct a multi-dimensional load coupling characteristic matrix;

[0013] Step S4: Based on meteorological factor data, the multivariate load coupling feature matrix is ​​clustered using the adaptive K-means clustering algorithm to obtain the clustered multivariate load coupling feature matrix;

[0014] Step S5: Input the clustered multivariate load coupling feature matrix into the pre-trained Attention-BiGRU (Attention Mechanism-Bidirectional Gated Recurrent Unit) prediction model for prediction to obtain the load prediction results, which include the prediction results of electrical load, thermal load, and cooling load.

[0015] In some embodiments, in step S2, the correlation analysis uses grey relational analysis to determine the degree of correlation between the two samples based on the similarity of their geometric shapes:

[0016] ;

[0017] ;

[0018] in, The correlation coefficient, For the degree of correlation, x i x0(k) is a sample sequence, ρ is the resolution coefficient, which is set to 0.5, and n is the number of sequences.

[0019] In some embodiments, in step S3, the multi-element load coupling characteristic matrix consists of three parts: the first part is the historical electrical load, heat load, and cooling load matrix; the second part is the multi-element load product matrix obtained by multiplying the electrical load, heat load, and cooling load; and the third part is the selected historical meteorological factor matrix. The multi-element load coupling characteristic matrix X is represented as follows:

[0020]

[0021] In the formula, X e For electrical load, X h For heat load, X c For cooling load, X o X is the product of the electrical load, heat load, and cooling load at the previous moment. w This is due to meteorological factors.

[0022] In some embodiments, in step S4, the adaptive K-means clustering algorithm iteratively aggregates M unlabeled samples into k clusters. Each value is assigned to the cluster with the smallest distance between the cluster centers. The value of k is selected by comprehensively considering two indicators: the sum of squared errors and the silhouette coefficient.

[0023] Furthermore, in step S4, the adaptive K-means clustering algorithm includes:

[0024] Initialize k cluster centers C k =[C1,C2,…,C k ], where the Euclidean distance dis from each value to each cluster center is calculated, and the expression is:

[0025]

[0026] Among them, X i Let C represent the i-th object. j Let X represent the j-th cluster center, n represent the number of attributes for each object, and X represent the number of cluster centers. it Let C represent the t-th attribute of the i-th object. jt The t-th attribute of the j-th cluster center;

[0027] The formulas for calculating the squared error SSE and the profile coefficient S are as follows:

[0028]

[0029]

[0030] Where p is the cluster C i The sample points in the middle, m i It is C i The cluster center; a is the average distance between the sample and other samples in the same cluster, and b is the average distance between the sample and all samples in the nearest cluster.

[0031] In some embodiments, in step S5, the Attention-BiGRU prediction model sequentially includes an input layer, a two-layer bidirectional gated recurrent unit (BiGRU) layer, an attention layer, a dropout layer, a fully connected layer, and an output layer.

[0032] The clustered multivariate load coupling feature matrix is ​​input into a bidirectional gated recurrent unit (BiGRU) through an input layer.

[0033] The bidirectional gated recurrent unit (BiGRU) layer is used to extract time feature information;

[0034] The attention layer is used to assign weights to the temporal feature information extracted by the BiGRU layer during the prediction process using an attention mechanism.

[0035] The dropout layer prevents overfitting by appropriately discarding neurons.

[0036] The fully connected layer maps the relationships between extracted features to the output layer.

[0037] The output layer is used to output load prediction results.

[0038] Furthermore, the BiGRU is a neural network model constructed on the basis of the GRU, consisting of two unidirectional GRUs with opposite directions, and its output is jointly determined by the states of the two GRUs.

[0039] GRU uses the status h from the previous transmission t-1 The input x of the current node's BiGRU layer t The gate status is obtained, namely the reset gate and the update gate. The reset gate and the update gate are calculated by controlling the forgotten and updated hidden states of the control unit.

[0040] ;

[0041] ;

[0042] Where, r t To reset the door, z t To update the door, W is the sigmoid function. ir W hr To reset the weight matrix in the gate, W iz W hz To update the weight matrix in the gate, b r ,b z These are the bias vectors in the reset gate and update gate, respectively;

[0043] The input data after reset and update is h. t-1 The current hidden state h obtained after the forgetting and selective memory processes. t The candidate hidden state is passed to the next node to obtain the hidden state, r t with h t-1 Element-wise dot product, when resetting the gate rt When the number of elements in the middle is close to 0, the hidden state from the previous time step is forgotten, and the hidden state h is forgotten. t By updating gate z t Compared with the hidden state h in the previous time step t-1 With candidate hidden state The hidden state h obtained by combining the results t As the output of the current layer, it is passed to the hidden layer at the next time step. The calculation formula is as follows:

[0044] ;

[0045] ;

[0046] in, For candidate hidden states, h t As the current hidden state, W is the output of the BiGRU layer. xh W hh Let b be the weight matrix of the candidate hidden states. h is the bias vector of the candidate hidden state.

[0047] Furthermore, the processing procedure of the attention layer includes:

[0048] ;

[0049] ;

[0050] ;

[0051] Among them, h t Let e ​​be the output of the BiGRU layer at time t. t This represents the output vector of the neural network at time t. The weight matrix, u,w is the weight matrix, b is the bias, α t denoted as the attention weights of the hidden layer states of the historical inputs to the current input, and v is the output of the attention layer at time t.

[0052] Secondly, the present invention provides a multi-load forecasting device for an integrated energy system, including a processor and a storage medium;

[0053] The storage medium is used to store instructions;

[0054] The processor is configured to operate according to the instructions to perform the steps of the method according to the first aspect.

[0055] Thirdly, the present invention provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0056] Fourthly, the present invention provides a computer device, including a processor and a storage medium;

[0057] The storage medium is used to store instructions;

[0058] The processor is configured to operate according to the instructions to perform the steps of the method according to the first aspect.

[0059] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides a multi-element load forecasting method for integrated energy systems based on K-means and Attention-BiGRU, which takes into account meteorological factors and the cross-coupling characteristics of multi-element loads, and solves the problem of low prediction accuracy caused by the increased uncertainty of load changes and the mutual influence between loads due to the related factors; and the use of K-means and Attention-BiGRU prediction models reduces model complexity and prediction time, so that the final prediction results can balance stability and accuracy. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the integrated energy system structure according to an embodiment of the present invention.

[0061] Figure 2 This is a schematic diagram of the BiGRU layer structure in an embodiment of the present invention.

[0062] Figure 3 This is a schematic diagram of the Attention layer structure in an embodiment of the present invention.

[0063] Figure 4 This is a structural diagram of the Attention-BiGRU prediction model in an embodiment of the present invention;

[0064] Figure 5 This is a flowchart of the multi-element load forecasting method for integrated energy systems according to an embodiment of the present invention. Detailed Implementation

[0065] The present invention will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be used to limit the scope of protection of the present invention.

[0066] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0067] In the description of this invention, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0068] Example 1

[0069] like Figure 1 As shown, a multi-element load forecasting method for integrated energy systems includes:

[0070] Step S1: Obtain integrated energy system data, which includes historical electricity load data, historical heat load data, historical cooling load data, and meteorological data within the system.

[0071] Step S2: Based on the integrated energy system data, perform correlation analysis on various meteorological factors and loads, and screen out historical meteorological factor data with strong correlations from the meteorological data within the system;

[0072] Step S3: Based on the historical meteorological data, historical electricity load data, historical heat load data, and historical cooling load data, construct a multi-dimensional load coupling characteristic matrix;

[0073] Step S4: Based on meteorological factor data, the multivariate load coupling feature matrix is ​​clustered using the adaptive K-means clustering algorithm to obtain the clustered multivariate load coupling feature matrix;

[0074] Step S5: Input the clustered multivariate load coupling feature matrix into the pre-trained Attention-BiGRU prediction model for prediction to obtain load prediction results, which include electrical load, heat load, and cooling load prediction results.

[0075] In some embodiments, a multivariate load forecasting method for an integrated energy system based on K-means and Attention-BiGRU includes: first, using grey relational analysis to calculate the correlation between meteorological factors and loads, and selecting strongly correlated meteorological factors as inputs to the forecasting model; second, considering the cross-coupling characteristics among multivariate loads within the system, constructing a multivariate load coupling feature matrix, and proposing a short-term multivariate load forecasting model based on K-means and Attention-BiGRU, using the mean absolute percentage error and root mean square error as forecasting result analysis and evaluation indicators to verify the effectiveness of the model.

[0076] The prediction process of this invention mainly includes data preparation, correlation analysis, coupling matrix construction, and prediction evaluation modules.

[0077] S1. Data preparation and correlation analysis:

[0078] The data preparation module is used to acquire raw data from the integrated energy system, including historical electricity load data, historical heat load data, historical cooling load data, and meteorological data within the system. Relevant meteorological information and load data for 24 hours daily over a year are selected from the integrated energy system. The training and test sets are divided in an 8:2 ratio. Before the experiment, the data is checked to identify and remove outliers, and the mean value is used to replace outliers to ensure relative data integrity.

[0079] The correlation analysis described is a model that uses grey relational analysis to quantitatively analyze the relevant factors affecting the target sequence. Its basic principle is to perform dimensionless processing on the known target sequence data and the comparison sequence data, and then determine the influence (weight) of each factor on the target sequence by comparing the similarity or dissimilarity of the data growth or decrease of each sequence on the target sequence. The correlation degree is determined by the similarity of the geometric shapes of the sample sequences. The calculation formula is as follows:

[0080] ;

[0081] ;

[0082] in, The correlation coefficient, For the degree of correlation, x i xk(k) represents the sample sequence, ρ is the resolution coefficient (set to 0.5), and n is the number of sequences. (Association value) The larger the value, the stronger the correlation between the two variables. By calculating the correlation between different meteorological factors and electricity, heat and cooling loads, meteorological information with strong correlation to the load to be predicted is selected to establish a meteorological factor matrix. The meteorological factor matrix is ​​used as part of the input matrix for multivariate load prediction.

[0083] In some embodiments, the integrated energy system of a university in Nanjing is selected as an example for analysis, and the grey relational degree values ​​are shown in Table 1. The table shows that the electrical load is highly correlated with dew point, wind speed, precipitation, wind direction, and temperature; the heat load is highly correlated with cloud type, dew point, wind speed, precipitation, relative humidity, and air pressure; and the cooling load is highly correlated with dew point, surface albedo, wind speed, precipitation, wind direction, and temperature.

[0084] Table 1. Grey Relational Degree of Different Meteorological Factors on Load

[0085]

[0086] S2. The coupling matrix construction module: Considering the cross-coupling characteristics of multi-energy loads, where electrical, thermal, and cooling loads influence each other, traditional single-load forecasting methods treat each load forecasting task as an independent and simple sub-problem, failing to account for the mutual conversion and influence relationships between different energy demands, leading to decreased forecasting accuracy. A multi-energy load coupling feature matrix reflecting the cross-influence of electrical, thermal, and cooling loads is constructed as input to the forecasting model, breaking the independence between different forms of energy. The multi-energy load coupling feature matrix consists of three parts: a historical load matrix, a multi-energy load product matrix obtained through a three-term polynomial, and a historical meteorological factor matrix filtered through grey relational analysis. Concatenating these three matrices yields the following coupling feature matrix:

[0087] ;

[0088] In the formula, X e For electrical load, X h For heat load, X c For cooling load, X o X is the load coupling product of the previous time step. w This is due to meteorological factors.

[0089] S3. The prediction module first uses adaptive K-means to cluster the original data into three categories, which are then fed into the Attention-BiGRU prediction model to obtain the prediction results.

[0090] The adaptive K-means clustering algorithm iteratively aggregates M unlabeled samples into k clusters. Each value is assigned to the cluster with the smallest distance between cluster centers, maximizing data similarity within each cluster. The algorithm uses two metrics, the sum of squared errors and the silhouette coefficient, to select the value of k, ultimately dividing the original data into meaningful arrays. For K-means, k cluster centers C are first initialized. k =[C1,C2,…,C k The Euclidean distance from each value to each cluster center is calculated using the following expression:

[0091] ;

[0092] Among them, X i Let C represent the i-th object. j Let X represent the j-th cluster center, n represent the number of attributes for each object, and X represent the number of cluster centers. it Let C represent the t-th attribute of the i-th object. jt Let t be the attribute of the j-th cluster center.

[0093] By comparing the Euclidean distance between each object and each cluster center in turn, the object with the smallest Euclidean distance is selected as the cluster, resulting in k clusters.

[0094] The value of k is determined by comprehensively considering two indicators: the sum of squared errors and the profile coefficient. The calculation formula is as follows:

[0095] ;

[0096] ;

[0097] Where p is the cluster C i The sample points in the middle, m i It is C i The cluster center; a is the average distance between the sample and other samples in the same cluster, and b is the average distance between the sample and all samples in the nearest cluster.

[0098] By using two parameters to jointly determine the value of k and dividing the original data into three categories for prediction, the complexity of the prediction model is reduced, prediction time is saved, and the accuracy of load prediction is improved by analyzing the inherent relationships of data within each load scenario.

[0099] The Attention-BiGRU prediction model mainly consists of an input layer, a two-layer BiGRU layer, an attention layer, a dropout layer, a fully connected layer, and an output layer.

[0100] The input layer takes the clustered multivariate load coupling feature matrix as input and performs normalization processing before inputting it into the prediction model.

[0101] The normalization process maps the data to the range of 0 to 1, as shown in the following formula:

[0102] ;

[0103] Where, x n For normalized data, x is the original data. max x represents the maximum value of the data. min This represents the minimum value of the data.

[0104] The BiGRU layer is a neural network model built on top of the GRU, consisting of two unidirectional GRUs with opposite directions, and its output is determined by the states of both GRUs. The GRU uses the previously transmitted state h... t-1 With the input x of the current node t The gate status is obtained, namely the reset gate and the update gate. The reset gate and the update gate are calculated by controlling the forgotten and updated hidden states of the control unit.

[0105] ;

[0106] ;

[0107] Where, r t To reset the door, z t To update the door, W is the sigmoid function. ir W hr To reset the weight matrix in the gate, W iz W hz To update the weight matrix in the gate, b r ,b z These are the bias vectors in the reset gate and update gate, respectively.

[0108] The input data after reset and update is h. t-1 The current state h obtained after the forgetting and selective memory processes t The candidate hidden state is passed to the next node to obtain the hidden state, r t with h t-1 Element-wise dot product, when resetting the gate r t When an element in the middle is close to 0, the hidden state from the previous time step is forgotten. The hidden state is determined by z. t Compared with the hidden state h in the previous time step t-1 With candidate hidden state The hidden state h obtained by combining the results t As the output of the current layer, it is passed to the hidden layer at the next time step. The calculation formula is as follows:

[0109] ;

[0110] ;

[0111] in, For candidate hidden states, h t W is currently hidden. xh W hh Let b be the weight matrix of the candidate hidden states. h is the bias vector of the candidate hidden state.

[0112] The attention layer utilizes an attention mechanism during prediction to assign weights to the temporal feature information extracted by the BiGRU hidden layer. This differentiated weight allocation improves data processing efficiency and prediction accuracy. Where x... t h represents the input of the BiGRU layer. t α represents the output of the hidden layer of the BiGRU neural network. t denoted as the probability distribution value of AM’s attention to the output of the BiGRU hidden layer, and y represents the BiGRU output optimized by the attention mechanism.

[0113] The input to the attention layer is the output vector h processed by the BiGRU hidden layer. t The weight calculation formula is as follows:

[0114] ;

[0115] ;

[0116] ;

[0117] Among them, e t This represents the output vector of the neural network at time t. The weight matrix, u,w is the weight matrix, b is the bias, α t denoted as the attention weights of the hidden layer states of the historical inputs to the current input, and v is the output of the attention layer at time t.

[0118] The dropout layer is used during training to appropriately discard hidden layer neurons, thus avoiding overfitting caused by the complexity of the neural network structure.

[0119] Each node in the fully connected layer is connected to all nodes in the previous layer, used to synthesize the features extracted earlier. At the end of a convolutional neural network, there are often one or two fully connected layers. The fully connected layers transform the two-dimensional feature map output by the convolution into a one-dimensional vector, extract the correlation between features and map it to the output layer.

[0120] The output layer is connected to the fully connected layer and is used to output load prediction result data.

[0121] S4. Evaluation module: The effectiveness of the proposed model is verified by analyzing numerical examples and selecting two evaluation indicators: mean absolute percentage error and root mean square error.

[0122] This invention uses mean absolute percentage error (MAPE) and root mean square error (RMSE) as model prediction and analysis indicators. The effectiveness of the proposed model is verified by comparing the MAPE and RMSE of different models, as shown in the following formula:

[0123] ;

[0124] ;

[0125] in, For the true value, The value is the predicted value, and n represents the number of data points.

[0126] This invention provides a method for predicting multiple loads in an integrated energy system. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

[0127] Example 2

[0128] Secondly, this embodiment provides a multi-load forecasting device for an integrated energy system, including a processor and a storage medium;

[0129] The storage medium is used to store instructions;

[0130] The processor is configured to operate according to the instructions to perform the steps of the method according to Embodiment 1.

[0131] Example 3

[0132] Thirdly, this embodiment provides a storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0133] Example 4

[0134] Fourthly, this embodiment provides a computer device, including a processor and a storage medium;

[0135] The storage medium is used to store instructions;

[0136] The processor is configured to operate according to the instructions to perform the steps of the method according to Embodiment 1.

[0137] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0138] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0139] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0140] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0141] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-element load forecasting method for integrated energy systems, characterized in that, include: Step S1: Obtain integrated energy system data, which includes historical electricity load data, historical heat load data, historical cooling load data, and meteorological data within the system. Step S2: Based on the integrated energy system data, perform correlation analysis on various meteorological factors and loads, and screen out historical meteorological factor data with strong correlations from the meteorological data within the system; the correlation analysis uses grey relational analysis to determine the degree of correlation between the two based on the similarity of the geometric shapes of the sample sequences. ； ； in, The correlation coefficient, For the degree of correlation, x i x0(k) is a sample sequence, ρ is the resolution coefficient, which is set to 0.5, and n is the number of sequences. Step S3: Based on the historical meteorological data, historical electricity load data, historical heat load data, and historical cooling load data, construct a multi-dimensional load coupling feature matrix. The multi-dimensional load coupling feature matrix consists of three parts: the first part is the historical electricity load, heat load, and cooling load matrix; the second part is the multi-dimensional load product matrix obtained by multiplying the electricity load, heat load, and cooling load; and the third part is the selected historical meteorological factor matrix. The multi-dimensional load coupling feature matrix X is represented as follows: ； In the formula, X e For electrical load, X h For heat load, X c For cooling load, X o X is the product of the electrical load, heat load, and cooling load at the previous moment. w Meteorological factors; Step S4: Based on meteorological factor data, the multivariate load coupling feature matrix is ​​clustered using the adaptive K-means clustering algorithm to obtain the clustered multivariate load coupling feature matrix; Step S5: Input the clustered multivariate load coupling feature matrix into the pre-trained Attention-BiGRU prediction model for prediction to obtain load prediction results, including electrical load, heat load, and cooling load prediction results. The Attention-BiGRU prediction model sequentially comprises an input layer, a bi-dimensional gated recurrent unit (BiGRU) layer, an attention layer, a dropout layer, a fully connected layer, and an output layer. The clustered multivariate load coupling feature matrix is ​​input into the BiGRU layer through the input layer. The BiGRU layer extracts temporal feature information. The attention layer assigns weights to the temporal feature information extracted by the BiGRU layer during prediction using an attention mechanism. The dropout layer prevents overfitting by appropriately discarding neurons. The fully connected layer maps the relationships between extracted features to the output layer. The output layer outputs the load prediction result.

2. The multi-source load forecasting method for integrated energy systems according to claim 1, characterized in that, In step S4, the adaptive K-means clustering algorithm iteratively aggregates the M unlabeled samples into k clusters. Each value is assigned to the cluster with the smallest distance between the cluster centers. The value of k is selected by comprehensively considering two indicators: the sum of squared errors and the silhouette coefficient.

3. The multi-source load forecasting method for integrated energy systems according to claim 2, characterized in that, In step S4, the adaptive K-means clustering algorithm includes: Initialize k cluster centers C k =[C1,C2,…,C k ], where the Euclidean distance dis from each value to each cluster center is calculated, and the expression is: ； Among them, X i Let C represent the i-th object. j Let X represent the j-th cluster center, n represent the number of attributes for each object, and X represent the number of cluster centers. it Let C represent the t-th attribute of the i-th object. jt The t-th attribute of the j-th cluster center; The formulas for calculating the squared error SSE and the profile coefficient S are as follows: ； ； Where p is the cluster C i The sample points in the middle, m i It is C i The cluster center; a is the average distance between the sample and other samples in the same cluster, and b is the average distance between the sample and all samples in the nearest cluster.

4. The integrated energy system multi-source load forecasting method according to claim 1, characterized in that, The BiGRU is a neural network model constructed on the basis of the GRU, consisting of two unidirectional GRUs with opposite directions, and its output is determined by the states of the two GRUs. GRU uses the status h from the previous transmission t-1 The input x of the current node's BiGRU layer t The gate status is obtained, namely the reset gate and the update gate. The reset gate and the update gate are calculated by controlling the forgotten and updated hidden states of the control unit. ； ； Where, r t To reset the door, z t To update the door, W is the sigmoid function. ir W hr To reset the weight matrix in the gate, W iz W hz To update the weight matrix in the gate, b r ,b z These are the bias vectors in the reset gate and update gate, respectively; The input data after reset and update is h. t-1 The current hidden state h obtained after the forgetting and selective memory processes. t The candidate hidden state is passed to the next node to obtain the hidden state, r t with h t-1 Element-wise dot product, when resetting the gate r t When the number of elements in the middle is close to 0, the hidden state from the previous time step is forgotten, and the hidden state h is forgotten. t By updating gate z t Compared with the hidden state h in the previous time step t-1 With candidate hidden state The hidden state h obtained by combining the results t As the output of the current layer, it is passed to the hidden layer at the next time step. The calculation formula is as follows: ； ； in, For candidate hidden states, h t As the current hidden state, W is the output of the BiGRU layer. xh W hh Let b be the weight matrix of the candidate hidden states. h is the bias vector of the candidate hidden state.

5. The multi-source load forecasting method for integrated energy systems according to claim 1, characterized in that, The processing procedure of the attention layer includes: ； ； ； Among them, h t Let e ​​be the output of the BiGRU layer at time t. t This represents the output vector of the neural network at time t. The weight matrix, u,w is the weight matrix, b is the bias, α t denoted as the attention weights of the hidden layer states of the historical inputs to the current input, and v is the output of the attention layer at time t.

6. A multi-element load forecasting device for an integrated energy system, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1 to 5.

7. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images