A combined prediction method and system for wind power, load and electricity price facing multiple regions

Abstract

The present invention relates to a method and system for jointly predicting wind power, load, and electricity price for multiple regions. First, wind power, load, and electricity price data of multiple regions are collected, relevant meteorological and geographical location information is obtained, and the obtained data is processed to form a feature input channel. Then, through frequency decomposition of the Laplacian matrix of the graph, a multi-scale graph convolution is constructed to capture multi-scale information in the feature channel. A new loss function and parameter update rule are defined to optimize model training. Subsequently, the shared intermediate feature layer forms a shared feature matrix by integrating the intermediate features of all target variables and inputs it into the model for training. Finally, the wind power, load, and electricity price target variables are output simultaneously. The present invention can extract the unique features of each region and capture the interdependencies between different regions and target variables, significantly improving the prediction accuracy and being applicable to scenarios such as wind power output, load demand, and electricity price prediction for multiple regions.

Description

Technical Field

[0001] The present invention relates to the technical field of integrated energy system prediction, and more specifically, to a method and system for jointly predicting wind power, load, and electricity price for multiple regions. Background Art

[0002] With the transformation of the global energy structure, integrated energy systems have gradually become an important research direction in the energy field. However, the complexity and uncertainty of integrated energy systems have brought many technical challenges. Integrated energy systems cover links such as energy production, transmission, storage, and consumption. In this process, the prediction of key factors such as wind power, load, and electricity price becomes particularly important. Wind power has obvious volatility and uncertainty due to factors such as meteorology and geographical environment; load demand is driven by various factors such as seasonal changes, climate conditions, and social and economic activities, making it difficult to accurately predict; electricity price is affected by the combined effects of market supply and demand, policy adjustments, and external factors, with a high degree of uncertainty. These factors are intertwined, resulting in the uncertainty of integrated energy systems, making it difficult for traditional methods of predicting each factor separately to meet actual needs. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiency of the poor joint prediction accuracy of wind power - load - electricity price for multiple regions in the prior art, and provide a method and system for jointly predicting wind power, load, and electricity price for multiple regions, which is applicable to scenarios such as wind power output, load demand, and electricity price prediction for multiple regions, and improves the prediction accuracy of wind power - load - electricity price joint prediction.

[0004] To solve the above - mentioned technical problems, the technical solution adopted by the present invention is:

[0005] Provide a method for jointly predicting wind power, load, and electricity price for multiple regions, including the following steps:

[0006] S1. Data collection: Collect wind power, load demand, and electricity price data of multiple regions, and obtain relevant meteorological data and geographical location information, and pre - process the collected data;

[0007] S2. Construct a feature input channel matrix: Construct a feature input channel matrix from the processed wind power, load demand, electricity price data, and relevant meteorological data and geographical location information; and divide the feature input channel matrix into a training set and a test set;

[0008] S3. Construct a graph convolutional neural network model based on spatial topological structure: Construct a graph convolutional neural network based on spatial topological structure, and construct multi - scale graph convolution through Laplacian frequency decomposition to capture multi - scale information in the feature input channel, and obtain an output intermediate feature matrix;

[0009] S4. Establish a shared intermediate feature layer: Integrate the output intermediate feature matrices of each target variable after multi-scale graph convolution into a shared matrix to generate a joint feature representation, where the target variables include wind power, load demand, and electricity price data;

[0010] S5. Optimize the model: Combine the intra-region loss and cross-region interaction to construct a loss function, and design a parameter update rule, introducing a task coordination gradient term to reduce the probability of the loss function falling into a local optimum;

[0011] S6. Model training: Use the training set to train the graph convolutional neural network model based on the spatial topological structure;

[0012] S7. Prediction: Input the test set into the trained graph convolutional neural network model based on the spatial topological structure to jointly predict the wind power-load-electricity price in multiple regions and output the predicted values.

[0013] A joint prediction method for wind power, load, and electricity price for multiple regions proposed by the present invention first collects wind power, load demand, and electricity price data of multiple regions, and obtains relevant meteorological data and geographical location information, processes the obtained data to form a feature input channel; then, through frequency decomposition of the Laplacian matrix of the graph, constructs a multi-scale graph convolution to capture multi-scale information in the feature channel; secondly, defines a new loss function and a parameter update rule to optimize model training; subsequently, the shared intermediate feature layer forms a shared feature matrix by integrating the intermediate features of all target variables and inputs it into the model for training, and finally outputs the wind power, load, and electricity price target variables simultaneously. The present invention can extract the unique features of each region and capture the interdependencies between different regions and target variables, significantly improving the prediction accuracy, applicable to scenarios such as wind power output, load demand, and electricity price prediction in multiple regions, and having extremely high practical application value and promotion potential.

[0014] Preferably, in step S1, the wind power sequence, load sequence, electricity price sequence, wind speed sequence, and wind direction sequence are processed by min-max normalization to obtain the processed power sequence P, load sequence L, electricity price sequence EP, and wind speed sequence WS, where the wind direction sequence is processed by sine and cosine to obtain the wind direction sine SWD and wind direction cosine CWD.

[0015] Preferably, the step S2 includes:

[0016] S21. After processing the data collected in step S1, obtain the feature input channel matrices of R regions, and the feature input channel matrices are represented as follows:

[0017]

[0018]

[0019]

[0020] wherein, H ws represents the wind speed input channel matrix, H SWD represents the wind direction sine input channel matrix, H CWD represents the wind direction cosine input channel matrix, H P represents the wind power input channel matrix, H L represents the load demand input channel matrix, H EP represents the electricity price input channel matrix; , , , , and respectively represent the wind power, wind speed, wind direction sine, wind direction cosine, load demand and electricity price at the r th moment in the t-n th area;

[0021] S22. Construct a complex weight matrix according to the collected geographical location information for simultaneously capturing distance and direction information, and the specific expression is:

[0022]

[0023] wherein, is the direction angle from node i to node j , is the standard deviation parameter of the Gaussian kernel function, is the threshold value. When the weight mapped by the distance between regions is less than the threshold value, it is considered that the two relationships are small and are selected to be ignored; r represents the r th region, r' represents the r' th region, drr' represents the distance between regions;

[0024] S23. Divide the data processed in step S1 into a training set and a test set.

[0025] Preferably, step S3 includes:

[0026] S31. Establish the graph convolution formula, which is expressed as follows:

[0027]

[0028] Wherein, is the graph convolution kernel function, is the graph convolution operation, is the Laplacian matrix of the graph, and its normalized form and eigenvalue decomposition form are respectively and , is a diagonal matrix, U is the Laplacian matrix L the normalized eigenvectors, is the complex weight matrix, is the identity matrix, The degree matrix is a diagonal matrix, composed of the node degrees ;

[0029] S32. Perform eigenvalue decomposition on the Laplacian matrix L of the graph to project the node features into different frequency ranges to capture the multi-scale characteristics of the node information:

[0030]

[0031] Wherein, is a diagonal matrix containing L the eigenvalues of;

[0032] S33. For the L eigenvalues in step S32, use a clustering algorithm to cluster the eigenvalues, cluster the spectrum into 3 clusters, and the eigenvalue range of each cluster is used as a frequency range to form 3 intervals, which respectively correspond to low frequency, medium frequency and high frequency;

[0033] S34. Design a corresponding graph convolution kernel for each frequency band and apply it to the feature input channel matrix on each channel to obtain the output intermediate feature matrix. The specific formula is:

[0034]

[0035]

[0036]

[0037] Wherein, represents the feature input channel matrix of the l layer, which contains the multi-dimensional input features corresponding to each target variable; Denote the output intermediate feature matrix of the wind power target variable at the l +1-th layer. The wind power target variable includes wind speed, wind direction, and wind power generation; Denote the output intermediate feature matrix of the load target variable at the l +1-th layer; Denote the output intermediate feature matrix of the electricity price target variable at the l +1-th layer.

[0038] Preferably, the step S5 includes:

[0039] S51. The loss function combines the loss within the region and the loss of cross-region interaction to achieve joint optimization. The formula is as follows:

[0040]

[0041]

[0042]

[0043] In the formula, is the weight, respectively represent the loss within the region and the cross-region interaction loss, is the r -th region's i -th variable's t predicted value at the -th r region's i -th variable's t true value at the is a constant close to 0, used to prevent the zero value problem in the logarithmic function; is used to control the mutual influence between regions, is the distance between regions, is the distance-sensitive value, used to adjust the sensitivity of the distance weighting term;

[0044] S52. The parameter update and optimization process includes:

[0045]

[0046]

[0047] In the formula, is the update amount of the model parameters, and are the regularization gradient term and the coordination gradient term respectively, is the Euclidean distance between the intermediate features of the i -th and j -th targets;η represents the learning rate; λ represents the coordinated gradient term coefficient, β represents the regularization gradient term coefficient.

[0048] Preferably, the step S4 includes: for wind power, load and electricity price, integrating the output intermediate feature matrices generated in each frequency band to respectively obtain the output feature matrices corresponding to the target variables , and passing the three output feature matrices through Concat concatenating to obtain: .

[0049] Preferably, in step S7, inputting the test set into the trained graph convolutional neural network model based on the spatial topological structure to obtain the predicted values :

[0050]

[0051] wherein, represents the predicted value of the wind power output using the test set; represents the predicted value of the load output using the test set; represents the predicted value of the electricity price output using the test set.

[0052] The present invention also provides a joint prediction system for wind power, load and electricity price for multiple regions, including:

[0053] Data acquisition module: used to collect wind power, load demand and electricity price data of multiple regions, and obtain relevant meteorological data and geographical location information, and preprocess the collected data;

[0054] Data processing module: used to construct a feature input channel matrix, and construct the processed wind power, load demand, electricity price data and relevant meteorological data and geographical location information into a feature input channel matrix; and divide the feature input channel matrix into a training set and a test set;

[0055] Graph convolutional neural network module: used to construct a graph convolutional neural network model based on the spatial topological structure, and construct multi-scale graph convolution through Laplace frequency decomposition to capture multi-scale information in the feature input channel to obtain an output intermediate feature matrix;

[0056] Shared intermediate feature layer module: used to establish a shared intermediate feature layer, and integrate the output intermediate feature matrices of each target variable after multi-scale graph convolution into a shared matrix to generate a joint feature representation, wherein the target variables include wind power, load demand and electricity price data;

[0057] Model optimization module: used to construct a loss function by combining the internal loss of regions and cross-region interactions, design a parameter update rule, and introduce a task coordination gradient term to reduce the probability of the loss function falling into a local optimum;

[0058] Model training module: used to train the graph convolutional neural network model based on the spatial topological structure using the training set;

[0059] Prediction output module: used to input the test set into the trained graph convolutional neural network model based on the spatial topological structure, jointly predict the wind power - load - electricity price in multiple regions, and output the predicted values.

[0060] The present invention also provides a computer device, including a processor, a memory, and a computer program stored in the memory, the computer program being configured to be executed by the processor, and when the computer program is executed by the processor, it implements the above-mentioned method for jointly predicting wind power, load, and electricity price for multiple regions.

[0061] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the above-mentioned method for jointly predicting wind power, load, and electricity price for multiple regions.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] The method and system for jointly predicting wind power, load, and electricity price for multiple regions of the present invention can extract the unique features of each region and capture the mutual dependencies between different regions and target variables, significantly improving the prediction accuracy, being applicable to scenarios such as wind power output, load demand, and electricity price prediction in multiple regions, and having extremely high practical application value and promotion potential. Brief Description of the Drawings

[0064] Figure 1 It is a schematic diagram of the method flow in Embodiment 1;

[0065] Figures 2 to 4 It is a schematic diagram of the joint prediction effect of wind power, load, and electricity price for Region 1 in Embodiment 6;

[0066] Figures 5 to 7 The figure is a schematic diagram of the joint prediction effect of wind power, load, and electricity price for Region 2 in Embodiment 6;

[0067] Figures 8 to 10 It is a schematic diagram of the joint prediction effect of wind power, load, and electricity price for Region 3 in Embodiment 6. Detailed Embodiments

[0068] The present invention will be further described below in conjunction with specific embodiments. Among them, the attached drawings are only for illustrative purposes, showing only schematic diagrams, rather than physical diagrams, and should not be construed as a limitation to the present invention; in order to better illustrate the embodiments of the present invention, some components in the attached drawings will be omitted, enlarged or reduced, which does not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the attached drawings may be omitted.

[0069] In the attached drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", etc. indicating the orientation or positional relationship, it is based on the orientation or positional relationship shown in the attached drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the attached drawings are only for illustrative purposes and should not be construed as a limitation to the present invention. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0070] Embodiment 1

[0071] This embodiment is the first embodiment of a combined prediction method for wind power, load and electricity price facing multiple regions, including the following steps:

[0072] S1. Data collection: Collect wind power, load demand and electricity price data of multiple regions, and obtain relevant meteorological data and geographical location information, and preprocess the collected data;

[0073] S2. Construct a feature input channel matrix: Construct the processed wind power, load demand, electricity price data and relevant meteorological data and geographical location information into a feature input channel matrix; and divide the feature input channel matrix into a training set and a test set;

[0074] S3. Construct a graph convolutional neural network model based on the spatial topological structure: Construct a graph convolutional neural network based on the spatial topological structure, and construct multi-scale graph convolution through Laplacian frequency decomposition to capture multi-scale information in the feature input channel, and obtain an output intermediate feature matrix;

[0075] S4. Establish a shared intermediate feature layer: Integrate the output intermediate feature matrices of each target variable after multi-scale graph convolution into a shared matrix to generate a joint feature representation, where the target variables include wind power, load demand and electricity price data;

[0076] S5. Optimize the model: Combine the intra-regional loss and cross-regional interaction to construct a loss function, and design a parameter update rule, introduce a task coordination gradient term to reduce the probability that the loss function falls into a local optimum;

[0077] S6. Model training: Use the training set to train the graph convolutional neural network model based on the spatial topological structure;

[0078] S7. Prediction: Input the test set into the trained graph convolutional neural network model based on the spatial topological structure, perform joint prediction on wind power-load-electricity price in multiple regions, and output the predicted values.

[0079] A joint prediction method for wind power, load and electricity price for multiple regions proposed in this embodiment first collects wind power, load demand and electricity price data of multiple regions, obtains relevant meteorological data and geographical location information, processes the obtained data to form a feature input channel; then, through frequency decomposition of the Laplacian matrix of the graph, constructs a multi-scale graph convolution to capture multi-scale information in the feature channel; secondly, defines a new loss function and parameter update rule to optimize model training; subsequently, the shared intermediate feature layer forms a shared feature matrix by integrating the intermediate features of all target variables and inputs it into the model for training, and finally outputs the wind power, load and electricity price target variables simultaneously. The present invention can refine the unique features of each region, capture the interdependencies between different regions and target variables, significantly improve the prediction accuracy, is applicable to scenarios such as wind power output, load demand and electricity price prediction in multiple regions, and has extremely high practical application value and promotion potential.

[0080] Embodiment Two

[0081] This embodiment is the second embodiment of a joint prediction method for wind power, load and electricity price for multiple regions. This embodiment is similar to the embodiment, the difference is that in this embodiment, the preprocessing of the collected data includes: performing min-max normalization processing on the wind power sequence, load sequence, electricity price sequence, wind speed sequence and wind direction sequence to obtain the processed power sequence P, load sequence L, electricity price sequence EP and wind speed sequence WS, and the wind direction sequence is processed by sine and cosine to obtain the wind direction sine SWD and wind direction cosine CWD.

[0082] Embodiment Three

[0083] This embodiment is the third embodiment of a joint prediction method for wind power, load and electricity price for multiple regions. This embodiment is similar to the embodiment, the difference is that in this embodiment, the detailed steps for constructing the feature input channel are proposed, specifically including:

[0084] S21. After processing the data collected in step S1, obtain the feature input channel matrix of R regions, and the feature input channel matrix is expressed as follows:

[0085]

[0086]

[0087]

[0088] wherein, H ws represents the wind speed input channel matrix, H SWD represents the wind direction sine input channel matrix, H CWD represents the wind direction cosine input channel matrix, H P represents the wind power input channel matrix, H L represents the load demand input channel matrix, H EP represents the electricity price input channel matrix; , , , , and respectively represent the wind power, wind speed, wind direction sine, wind direction cosine, load demand and electricity price at the r th moment in the t-n th area;

[0089] S22. Construct a complex weight matrix according to the collected geographical location information to capture distance and direction information simultaneously. The specific expression is:

[0090]

[0091] wherein, is the direction angle from node i to node j , is the standard deviation parameter of the Gaussian kernel function, is the threshold. When the weight mapped by the distance between regions is less than the threshold, it is considered that the two relationships are small and can be ignored; r represents the r th region, r' represents the r' th region, drr' represents the distance between regions;

[0092] S23. Divide the data processed in step S1 into a training set and a test set.

[0093] Example 4

[0094] This embodiment is the fourth embodiment of a combined prediction method for wind power, load, and electricity price for multiple regions. This embodiment is similar to the embodiment, except that in this embodiment, detailed steps for constructing a graph convolutional neural network model based on a spatial topological structure are proposed, specifically including:

[0095] S31. Establish a graph convolution formula, expressed as follows:

[0096]

[0097] In the formula, is the graph convolution kernel function, is the graph convolution operation, is the Laplacian matrix of the graph, and its normalized form and eigenvalue decomposition form are respectively and , is a diagonal matrix, U is the Laplacian matrix L The normalized eigenvector, is the complex weight matrix, is the identity matrix, The degree matrix is a diagonal matrix, composed of the node degrees ;

[0098] S32. Perform eigenvalue decomposition on the Laplacian matrix L of the graph to project the node features into different frequency ranges to capture the multi-scale characteristics of the node information:

[0099]

[0100] In the formula, is a diagonal matrix, containing L The eigenvalues of ;

[0101] S33. For the L eigenvalues in step S32, use a clustering algorithm to cluster the eigenvalues, cluster the spectrum into 3 clusters, and the eigenvalue range of each cluster is used as a frequency range to form 3 intervals, and the 3 intervals respectively correspond to low frequency, medium frequency, and high frequency;

[0102] S34. Design a corresponding graph convolution kernel for each frequency band and apply it to the feature input channel matrix on each channel to obtain the output intermediate feature matrix. The specific formula is:

[0103]

[0104]

[0105]

[0106] In the formula, represents the characteristic input channel matrix of the l -th layer, which contains multi-dimensional input features corresponding to each target variable; represents the output intermediate feature matrix of the wind power target variable in the l +1-th layer. The wind power target variables include wind speed, wind direction, and wind power; represents the output intermediate feature matrix of the load target variable in the l +1-th layer; represents the output intermediate feature matrix of the electricity price target variable in the l +1-th layer.

[0107] In step S4, for wind power, load, and electricity price, the output intermediate feature matrices generated in each frequency band are integrated to obtain the output feature matrices corresponding to the target variables , and the three output feature matrices are Concat concatenated to obtain: . The test set is input into the trained graph convolutional neural network model based on the spatial topology structure to obtain the predicted values :

[0108]

[0109] In the formula, represents the predicted value of the wind power output using the test set; represents the predicted value of the load output using the test set; represents the predicted value of the electricity price output using the test set.

[0110] In this embodiment, is actually the characteristic input channel matrix corresponding to Embodiment 3. For example, for the 4 factors affecting Wind (wind power, wind speed, sine of wind direction, cosine of wind direction), there are correspondingly 4 characteristic input channel matrices ( H ws , H SWD , H CWD , and H P ), then has 4; similarly, there is only one for Load and Price. The meaning of the output intermediate feature matrix is actually: since there are three intervals in the present invention, there will be three values of k. Then, for Wind, there will be 3 * 4 = 12 output intermediate feature matrices, and the 12 output intermediate feature matrices regarding Wind are integrated to obtain the output feature matrix Hwind , similarly, there will be 3 output intermediate feature matrices for Load and Price, and then they are integrated respectively to obtain H load and H price . Finally, they are concatenated for simultaneous output of the target variables of wind power, load, and electricity price, denoted as .

[0111] Embodiment Five

[0112] This embodiment is the fifth embodiment of a joint prediction method for wind power, load, and electricity price for multiple regions. This embodiment is similar to the embodiment, but the difference is that in this embodiment, detailed steps for optimizing the model are proposed, specifically including:

[0113] S51. The loss function combines the loss within the region and the loss of cross-region interaction to achieve joint optimization. The formula is as follows:

[0114]

[0115]

[0116]

[0117] In the formula, is the weight, respectively represent the loss within the region and the cross-region interaction loss, is the predicted value of the r th variable at the i th time in the t th region, the r th variable at the i th time in the t th region; is a constant close to 0, used to prevent the zero value problem in the logarithmic function; is used to control the mutual influence between regions, the distance between regions, is the distance-sensitive value, used to adjust the sensitivity of the distance weighting term;

[0118] S52. The parameter update and optimization process includes:

[0119]

[0120]

[0121] In the formula, is the update amount of the model parameter, and They are the regularization gradient term and the coordinated gradient term respectively, is the Euclidean distance between the intermediate features of the i and the j th targets; η represents the learning rate; λ represents the coordinated gradient term coefficient, β represents the regularization gradient term coefficient.

[0122] Example Six

[0123] In this example, to verify the effectiveness of the proposed joint prediction method for wind power, load and electricity price for multiple regions, first, the wind power sequence, load sequence, electricity price sequence, wind speed sequence, wind direction sequence and geographical location information of 30 regions in a certain area from 00:00 on January 1, 2020 to 23:00 on December 31, 2020 are obtained in step S1. A geographical location matrix containing direction information is generated through the geographical location information. Then, a graph convolutional neural network based on the spatial topological structure is constructed, and multi-scale information in the feature channels is captured by multi-scale graph convolution of Laplacian frequency decomposition. A shared intermediate feature layer is designed. Finally, error coordination is performed through the shared gradient and loss function to jointly predict wind power-load-electricity price. Since the number of regions is large, the prediction results of 3 regions are taken for verification in this case, where Figures 2 to 10 represents the example prediction effect diagrams of the joint prediction method for wind power, load and electricity price applied to the graph convolutional neural network and individual prediction. It can be clearly seen that the joint prediction method of the present invention can effectively improve the prediction accuracy of wind power-load-electricity price.

[0124] Example Seven

[0125] This example is an example of a joint prediction system for wind power, load and electricity price for multiple regions. Based on the prediction methods provided in Examples One to Five, a joint prediction system for wind power, load and electricity price for multiple regions is provided, including:

[0126] Data acquisition module: used to collect wind power, load demand and electricity price data of multiple regions, obtain relevant meteorological data and geographical location information, and preprocess the collected data;

[0127] Data processing module: used to construct feature input channels and geographical location matrix, construct the processed wind power, load demand, electricity price data and relevant meteorological data and geographical location information into feature input channels; and divide the feature input channels into training sets and test sets;

[0128] Graph Convolutional Neural Network Module: It is used to construct a graph convolutional neural network model based on the spatial topology structure, and construct multi-scale graph convolutions through Laplacian frequency decomposition to capture multi-scale information in the feature input channels, and obtain the output intermediate feature matrix;

[0129] Shared Intermediate Feature Layer Module: It is used to establish a shared intermediate feature layer, integrate the output intermediate feature matrices of each target variable after multi-scale graph convolution into the shared matrix, and generate a joint feature representation, where the target variables include wind power, load demand, and electricity price data;

[0130] Model Optimization Module: It is used to construct a loss function by combining the intra-region loss and cross-region interaction, and design a parameter update rule, introducing a task coordination gradient term to reduce the probability of the loss function falling into a local optimum;

[0131] Model Training Module: It is used to train the graph convolutional neural network model based on the spatial topology structure using the training set;

[0132] Prediction Output Module: It is used to input the test set into the trained graph convolutional neural network model based on the spatial topology structure, jointly predict the wind power-load-electricity price in multiple regions, and output the predicted values.

[0133] Example Eight

[0134] This example is a computer device, including a processor, a memory, and a computer program stored in the memory. The computer program is configured to be executed by the processor, and when the computer program is executed by the processor, it implements the method for jointly predicting wind power, load, and electricity price for multiple regions described in any one of Embodiments 1 to 5 above.

[0135] Example Nine

[0136] This example is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for jointly predicting wind power, load, and electricity price for multiple regions described in any one of Embodiments 1 to 5 above.

[0137] In the specific content of the above specific implementation manner, each technical feature can be combined arbitrarily without contradiction. For the sake of concise description, not all possible combinations of the above technical features are described. However, as long as the combination of these technical features does not exist in contradiction, it should be considered as the scope recorded in this specification.

[0138] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation manners here. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. A method for joint forecasting of wind power, load and electricity price for multiple regions, characterized in that: The following steps are involved: S1. Data collection: Collect wind power, load demand and electricity price data from multiple regions, obtain relevant meteorological data and geographic location information, and pre-process the collected data; S2. Constructing a feature input channel matrix: constructing the processed wind power, load demand, electricity price data, related meteorological data and geographic location information into a feature input channel matrix; and dividing the feature input channel matrix into a training set and a test set; S3. Construct a graph convolutional neural network model based on spatial topological structure: Construct a graph convolutional neural network based on spatial topological structure, and construct a multi-scale graph convolution through Laplace frequency decomposition to capture the multi-scale information in the feature input channel, and obtain the output intermediate feature matrix; S4. Optimizing the model: constructing a loss function by combining the intra-regional loss and the cross-regional interaction, designing a parameter update rule, and introducing a task coordination gradient term to reduce the probability of the loss function falling into a local optimum; the step S4 includes: S41. The loss function combines the intra-region loss and the cross-region interaction loss to achieve joint optimization. The formula is as follows: In the formula, is the weight, Represent the loss within the region and the interaction loss between regions, respectively. For the r Region No. i variables t The predicted value at time, No. r Region No. i variables t The true value of the moment; is a constant close to 0, used to prevent the zero value problem in the logarithmic function; Used to control the mutual influence between regions. d rr' is the distance between regions, is the distance-sensitive value, used to adjust the sensitivity of the distance-weighted term; R represents the region; S42. The parameter updating and optimization process includes: In the formula, is the update amount of the model parameters, and are the regularization gradient term and the coordination gradient term, respectively. It is i and j The Euclidean distance between the intermediate features of the targets; η represents the learning rate; λ represents the coordination gradient term coefficient, β represents the regularization gradient term coefficient; m =3; S5. Establish a shared intermediate feature layer: Integrate the output intermediate feature matrix of each target variable after multi-scale graph convolution into a shared matrix to generate a joint feature representation, where the target variables include wind power, load demand, and electricity price data; S6. Model training: Use the training set to train the graph convolutional neural network model based on spatial topology structure; S7. Prediction: The test set is input into the trained graph convolutional neural network model based on spatial topology structure to make a joint prediction of wind power, load and electricity price in multiple regions and output the predicted value.

2. The multi-regional wind power, load and electricity price joint forecasting method according to claim 1 is characterized in that: In step S1, the wind power sequence, load sequence, electricity price sequence, wind speed sequence and wind direction sequence are normalized by min-max to obtain the processed power sequence P, load sequence L, electricity price sequence EP and wind speed sequence WS, among which the wind direction sequence is processed by sine and cosine to obtain wind direction sine SWD and wind direction cosine CWD.

3. The multi-regional wind power, load and electricity price joint forecasting method according to claim 1 is characterized in that: The step S2 includes: S21. After processing the data collected in step S1, a feature input channel matrix of R regions is obtained. The feature input channel matrix is ​​expressed as follows: In the formula, H ws represents the wind speed input channel matrix, H SWD represents the wind direction sinusoidal input channel matrix, H CWD represents the wind direction cosine input channel matrix, H P represents the wind power input channel matrix, H L represents the load demand input channel matrix, H EP represents the electricity price input channel matrix; , , , , and Respectively represent r In the area tn Wind power, wind speed, wind direction sine, wind direction cosine, load demand and electricity price at the moment; S22. Construct a complex weight matrix based on the collected geographic location information , used to capture distance and direction information at the same time, the specific expression is: In the formula, For Node i To Node j The direction angle, is the standard deviation parameter of the Gaussian kernel function, is the threshold. When the weight of the distance mapping between regions is less than the threshold, the two relationships are considered small and are ignored. r Indicates r Regions, r' Indicates r' Regions, d rr' Indicates the distance between regions; S23. Divide the data processed in step S1 into a training set and a test set.

4. The method for joint forecasting of wind power, load and electricity price for multiple regions according to claim 3 is characterized in that: The step S3 comprises: S31. Establish the graph convolution formula, expressed as follows: In the formula, is the graph convolution kernel function, is the graph convolution operation, is the Laplace matrix of the graph, and its normalized form and eigenvalue decomposition form are and , is a diagonal matrix, U is the Laplace matrix L The normalized feature vector, is the complex weight matrix, is the identity matrix, The degree matrix is ​​a diagonal matrix composed of node degrees composition; S32. Laplacian matrix of graph L Perform eigenvalue decomposition and project node features into different frequency ranges to capture the multi-scale characteristics of node information: In the formula, is a diagonal matrix containing L The eigenvalue of ; S33. For step S32 L The eigenvalues ​​of are clustered using a clustering algorithm to cluster the eigenvalues ​​into three clusters. The eigenvalue range of each cluster is used as a frequency range to form three intervals, which correspond to low frequency, medium frequency and high frequency respectively. S34. Design a corresponding graph convolution kernel for each frequency band , and applied to the feature input channel matrix on each channel to obtain the output intermediate feature matrix. The specific formula is: In the formula, Indicates l The feature input channel matrix of the layer, which contains the multi-dimensional input features corresponding to each target variable; Indicates that the wind power target variable is l +1 layer output intermediate feature matrix, wind power target variables include wind speed, wind direction, and wind power; Indicates that the load target variable is l +1 layer output intermediate feature matrix; Indicates that the target variable of electricity price is l The output intermediate feature matrix of layer +1.

5. The method for joint forecasting of wind power, load and electricity price for multiple regions according to claim 4 is characterized in that: The step S5 includes: integrating the output intermediate feature matrices generated by each frequency band for wind power, load and electricity price, and obtaining the output feature matrices corresponding to the target variables respectively. , the three output feature matrices are passed through Concat Splicing to get: .

6. The multi-regional wind power, load and electricity price joint forecasting method according to claim 1 is characterized in that: In step S7, the test set is input into the trained graph convolutional neural network model based on spatial topology structure to obtain the predicted value : In the formula, Represents the predicted value of wind power output using the test set; Represents the predicted value of the load output using the test set; Represents the predicted value of the electricity price output using the test set.

7. A wind power, load and electricity price joint forecasting system for multiple regions, characterized in that: include: Data acquisition module: used to collect wind power, load demand and electricity price data in multiple regions, obtain relevant meteorological data and geographical location information, and pre-process the collected data; Data processing module: used to construct a feature input channel matrix, which is composed of the processed wind power, load demand, electricity price data, related meteorological data and geographic location information; And divide the feature input channel matrix into training set and test set; Graph convolutional neural network module: used to build a graph convolutional neural network model based on spatial topological structure, and construct multi-scale graph convolution through Laplace frequency decomposition to capture multi-scale information in the feature input channel, and obtain the output intermediate feature matrix; Shared intermediate feature layer module: used to establish a shared intermediate feature layer, integrate the output intermediate feature matrix of each target variable after multi-scale graph convolution into the shared matrix, and generate a joint feature representation, where the target variables include wind power, load demand and electricity price data; Model optimization module: It is used to construct the loss function by combining the internal loss of the region and the cross-region interaction, and to design the parameter update rules and introduce the task coordination gradient term to reduce the probability of the loss function falling into the local optimum. The loss function combines the internal loss of the region and the loss of the cross-region interaction to achieve joint optimization. The formula is as follows: In the formula, is the weight, Represent the loss within the region and the interaction loss between regions, respectively. For the r Region No. i variables t The predicted value at time, No. r Region No. i variables t The true value of the moment; is a constant close to 0, used to prevent the zero value problem in the logarithmic function; Used to control the mutual influence between regions. is the distance between regions, It is a distance-sensitive value, used to adjust the sensitivity of the distance-weighted term; R Indicates region; The parameter updating and optimization process includes: In the formula, is the update amount of the model parameters, and are the regularization gradient term and the coordination gradient term, respectively. It is i and j The Euclidean distance between the intermediate features of the targets; η represents the learning rate; λ represents the coordination gradient term coefficient, β represents the regularization gradient term coefficient; m =3; Model training module: used to train the graph convolutional neural network model based on spatial topology structure using the training set; Prediction output module: used to input the test set into the trained graph convolutional neural network model based on spatial topology structure, conduct joint prediction of wind power, load and electricity price in multiple regions, and output the predicted value.

8. A computer device, characterized in that: The invention comprises a processor, a memory and a computer program stored in the memory, wherein the computer program is configured to be executed by the processor, and when the computer program is executed by the processor, the method for joint forecasting of wind power, load and electricity price for multiple regions as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-regional wind power, load and electricity price joint forecasting method as described in any one of claims 1 to 6 is implemented.

Images