Long sequence load forecasting method and device of power system, electronic equipment and medium

Abstract

The application relates to a long sequence load prediction method and device of a power system, electronic equipment and a medium, wherein the method comprises: performing multistage operation on an input matrix by using a Kronecker product layer to construct a high-dimensional matrix; arranging the high-dimensional matrix by using a rotation matrix mask, and deleting high-order repeated items from the high-dimensional matrix by using the high-dimensional matrix satisfying a preset condition; applying a preset kernel function to the input matrix of different orders; processing an output matrix according to a preset multilayer perception machine; evaluating a preset long sequence load prediction model of the power system by using a mean square error and a mean absolute error; analyzing the relationship and influence between internal variables of the preset long sequence load prediction model of the power system; and determining an optimization scheme of the preset long sequence load prediction model of the power system. The application can realize accurate long sequence load prediction by combining a large model with a Kronecker product, a rotation matrix mask and the like, and can provide effective system optimization strategies, and has significant improvement in feature extraction capability, model robustness, model interpretability and long sequence prediction.

Description

Technical Field

[0001] This application relates to the field of time series forecasting technology, and in particular to a method, apparatus, electronic device and medium for long-sequence load forecasting of power systems. Background Technology

[0002] Long sequence prediction is a new field that extends from short sequence prediction. Early methods for long sequence prediction included recurrent neural networks (RNNs) and their variants LSTM and GRU, with researchers aiming to use RNN-like networks to extract long-term load features and generate predictions. Recently, the Transformer framework, based on the self-attention mechanism, has demonstrated powerful capabilities in sequence data. The self-attention mechanism excels at one-time encoding of long-term time series, avoiding the loss of long-term information. Furthermore, the encoder in the Transformer can encode features from long-term sequences, which are then decoded in the decoder for prediction.

[0003] Current load forecasting research primarily focuses on point forecasting. However, with the increasing number of distributed renewable energy generators and new responsive loads in smart grids, accurate load forecasting is becoming increasingly challenging. The variability and uncertainty of electricity demand have become a challenge for the utility industry. The nonlinear characteristics of existing large-scale models are mainly expressed through activation functions. While these activation functions can capture complex nonlinear relationships in the data, they also make the internal mechanisms of the model black-box, difficult to interpret and understand intuitively. Users often cannot clearly understand how each input variable affects the forecast results, leading to a lack of transparency and reliability in practical applications.

[0004] In addition, existing large models are difficult to fully utilize and describe cross-variable relationships. Although they can handle multivariate inputs, they are difficult to intuitively show the mutual influence and relationships between variables. The parameters of the models are usually distributed in a high-dimensional space, which cannot clearly reflect the interaction between variables. This is particularly inadequate for application scenarios that require understanding the relationship between variables and optimizing the system. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and medium for long-sequence load forecasting of power systems to address the problems in existing large-scale models for time series forecasting, such as the large amount of data involved, the black box effect caused by expressing nonlinear characteristics through complex activation functions, and the difficulty in intuitively expressing cross-variable relationships. These problems limit the effectiveness and interpretability of the models in practical applications and affect users' understanding and trust in the model's forecast results.

[0006] The first aspect of this application provides a long-sequence load forecasting method for power systems, comprising the following steps: performing multi-level operations on an input matrix using a Kronecker product layer to construct a high-dimensional matrix; rearranging the high-dimensional matrix using a rotation matrix mask to generate a high-dimensional matrix that meets preset conditions, and deleting high-order duplicate terms from the high-dimensional matrix that meets the preset conditions to generate deleted data; applying a preset kernel function to input matrices of different orders based on the deleted data to convert the input matrices into output matrices of corresponding orders; processing the output matrix according to a preset multilayer perceptron to generate a final forecast result; evaluating a preset long-sequence load forecasting model for power systems using mean squared error and mean absolute error based on the final forecast result to generate an evaluation result; analyzing the relationships and influences between the internal variables of the preset long-sequence load forecasting model for power systems based on the evaluation result to generate an analysis result, and determining an optimization scheme for the preset long-sequence load forecasting model for power systems based on the analysis result.

[0007] Optionally, in one embodiment of this application, before using the Kronecker product layer to perform multi-level operations on the input matrix to construct a high-dimensional matrix, the method further includes: performing data cleaning, missing value imputation, and outlier processing on the load sequence dataset of the power system to remove erroneous information from the load sequence dataset and obtain the removed data; based on the removed data, dividing the load sequence dataset into a training set and a test set, and normalizing the training set and the test set to obtain the processed data.

[0008] Optionally, in one embodiment of this application, the formula for calculating the high-dimensional matrix is:

[0009] X N (i,j,…,n)=X e (i)×X e (j)×…×X e (n)'

[0010] Among them, X e (n) is obtained by transforming the dimension of the input matrix X. N A matrix consists of N X e The result obtained from the Kronecker product of matrices.

[0011] Optionally, in one embodiment of this application, the formula for calculating the elements in the rotation matrix mask is:

[0012]

[0013] Where i, j, n are the position codes of the corresponding dimensions of the mask matrix, and m represents the mask matrix.

[0014] Optionally, in one embodiment of this application, the formula for calculating the mean square error is:

[0015]

[0016] Among them, y i For the predicted target, This is a predicted value;

[0017] The formula for calculating the mean absolute error is:

[0018]

[0019] Among them, y i For the predicted target, The predicted value is denoted as .

[0020] A second aspect of this application provides a long-sequence load forecasting device for a power system, comprising: a construction module for performing multi-level operations on an input matrix using a Kronecker product layer to construct a high-dimensional matrix; a sorting module for sorting the high-dimensional matrix using a rotation matrix mask to generate a high-dimensional matrix that meets preset conditions, and deleting high-order duplicate terms from the high-dimensional matrix that meets the preset conditions to generate deleted data; an input module for applying a preset kernel function to input matrices of different orders based on the deleted data to convert the input matrices into output matrices of corresponding orders; a processing module for processing the output matrix according to a preset multilayer perceptron to generate a final prediction result; an evaluation module for evaluating a preset long-sequence load forecasting model for a power system using mean squared error and mean absolute error based on the final prediction result to generate an evaluation result; and a prediction module for analyzing the relationships and influences between internal variables of the preset long-sequence load forecasting model for a power system based on the evaluation result to generate analysis results, and determining an optimization scheme for the preset long-sequence load forecasting model for a power system based on the analysis results.

[0021] Optionally, in one embodiment of this application, it further includes: a removal module, configured to perform data cleaning, missing value imputation, and outlier processing on the load sequence dataset of the power system before using the Kronecker product layer to perform multi-level operations on the input matrix to construct a high-dimensional matrix, so as to remove error information in the load sequence dataset and obtain the removed data; and a generation module, configured to divide the load sequence dataset into a training set and a test set based on the removed data, and normalize the training set and the test set to obtain the processed data.

[0022] Optionally, in one embodiment of this application, the formula for calculating the high-dimensional matrix is:

[0023] X N (i,j,…,n)=X e (i)×X e (j)×…×X e (n)'

[0024] Among them, X e (n) is obtained by transforming the dimension of the input matrix X. N A matrix consists of N X e The result obtained from the Kronecker product of matrices.

[0025] Optionally, in one embodiment of this application, the formula for calculating the elements in the rotation matrix mask is:

[0026]

[0027] Where i, j, n are the position codes of the corresponding dimensions of the mask matrix, and m represents the mask matrix.

[0028] Optionally, in one embodiment of this application, the formula for calculating the mean square error is:

[0029]

[0030] Among them, y i For the predicted target, This is a predicted value;

[0031] The formula for calculating the mean absolute error is:

[0032]

[0033] Among them, y i For the predicted target, The predicted value is denoted as .

[0034] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the long-sequence load forecasting method for a power system as described in the above embodiments.

[0035] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the long-sequence load forecasting method for a power system as described above.

[0036] A fifth aspect of this application provides a computer program product that stores a computer program that, when executed by a processor, implements the long-sequence load forecasting method for a power system as described above.

[0037] This application's embodiments achieve accurate long-series load forecasting by combining large models with techniques such as Kronecker product and rotation matrix masking. It intuitively expresses variable relationships and provides effective system optimization strategies. Furthermore, it significantly improves feature extraction capabilities, model robustness, model interpretability, and long-series forecasting, offering a more effective and reliable method for power system load forecasting. This solves the problems in related technologies where existing large models in time series forecasting suffer from massive data volume, black-box representation of nonlinear characteristics through complex activation functions, and difficulty in intuitively expressing cross-variable relationships. These limitations restrict the effectiveness and interpretability of the models in practical applications, affecting users' understanding and trust in the model's prediction results.

[0038] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0039] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0040] Figure 1 This is a flowchart of a long-sequence load forecasting method for a power system according to an embodiment of this application;

[0041] Figure 2 This is a schematic diagram of the prediction model structure of a long-sequence load forecasting method for a power system according to an embodiment of this application;

[0042] Figure 3 This is a schematic diagram of the overall algorithm of the prediction model for a long-sequence load prediction method for a power system according to an embodiment of this application;

[0043] Figure 4This is a schematic diagram of the overall process of a long-sequence load forecasting method for a power system according to an embodiment of this application;

[0044] Figure 5 This is a schematic diagram of the structure of a long-sequence load forecasting device for a power system according to an embodiment of this application;

[0045] Figure 6 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0046] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0047] The following description, with reference to the accompanying drawings, outlines a method, apparatus, electronic device, and medium for long-sequence load forecasting of power systems according to embodiments of this application. Addressing the issues raised in the background section regarding existing large-scale models in time series forecasting, such as massive data sets, black-box representation of nonlinear characteristics through complex activation functions, and difficulty in intuitively expressing cross-variable relationships, this method limits the effectiveness and interpretability of the models in practical applications, impacting user understanding and trust in the model's prediction results. This application provides a method for long-sequence load forecasting of power systems. This method combines large-scale models with techniques such as Kronecker product and rotation matrix masking to achieve accurate long-sequence load forecasting, intuitively express variable relationships, perform detailed variable impact analysis, and provide effective system optimization strategies, thereby improving the overall performance of the power system and ensuring its safe and efficient operation. This solves the problems of existing large-scale models in time series forecasting, such as massive data sets, black-box representation of nonlinear characteristics through complex activation functions, and difficulty in intuitively expressing cross-variable relationships, which limit the effectiveness and interpretability of the models in practical applications and affect user understanding and trust in the model's prediction results.

[0048] Before introducing the long-sequence load forecasting method for power systems proposed in the embodiments of this application, let's briefly introduce the background technology involved in this method.

[0049] When making deterministic point predictions of power system loads, the core process involves the model learning the linear and nonlinear characteristics of the load sequence and fitting it to the future situation to be predicted. However, due to the limited load sequence data, the prediction model needs to be able to extract truly effective linear and nonlinear characteristics from the limited energy dataset. This places high demands on the performance of the prediction model, and the existing methods still lack the ability to extract load sequence features and urgently need improvement.

[0050] In short-term load sequence forecasting research, most studies focus on short-sequence forecasting, which only predicts one or a few steps into the future. Few studies predict more than a hundred steps into the future, i.e., long-sequence forecasting. However, long-sequence forecasting plays an important role in the medium-term planning and safe operation of loads in power systems. Therefore, a high-performance long-sequence load forecasting model is urgently needed.

[0051] Specifically, Figure 1 This is a flowchart illustrating a long-sequence load forecasting method for a power system provided in an embodiment of this application.

[0052] like Figure 1 As shown, the long-sequence load forecasting method for this power system includes the following steps:

[0053] In step S101, the input matrix is ​​subjected to multi-level operations using the Kronecker product layer to construct a high-dimensional matrix.

[0054] It is understood that the Kronecker product layer in the embodiments of this application can perform multi-order Kronecker product operations on the input matrix to form an N-order input matrix, thereby better capturing and expressing the nonlinear relationship of the system.

[0055] In actual implementation, the input data in this embodiment is transformed into an input matrix by the input layer and then passed through the Kronecker product layer. The Kronecker product layer can represent the higher-order nonlinear relationship of the system. By passing the input matrix through an N-order Kronecker product, an N-order input matrix can be generated. For any input matrix X, the index mapping relationship before and after the expansion is as follows: X(i,j) = X e (m), where m = (i-1)*N+j. X ​​after the Nth order Kronecker product is denoted as X in this application. N X N The shape is (length,…,length), and the dimension is N.

[0056] The high-dimensional representation in the embodiments of this application can effectively capture the complex nonlinear relationships within the system and intuitively demonstrate the interaction between variables, helping users to understand the prediction mechanism of the model more deeply, thereby enhancing their confidence in the prediction results.

[0057] In one embodiment of this application, the formula for calculating the high-dimensional matrix is:

[0058] X N (i,j,…,n)=X e (i)×X e (j)×…×X e (n)'

[0059] Among them, X e (n) is obtained by transforming the dimension of the input matrix X. N A matrix consists of N X e The result obtained from the matrix Kronecker product therefore has a large number of repeated higher-order terms.

[0060] X here N A matrix consists of N X e The result is obtained by matrix Kronecker product, and therefore has a large number of repeated higher-order terms.

[0061] Optionally, in one embodiment of this application, before using the Kronecker product layer to perform multi-level operations on the input matrix to construct a high-dimensional matrix, the method further includes: performing data cleaning, missing value imputation, and outlier processing on the load sequence dataset of the power system to remove erroneous information from the load sequence dataset and obtain the removed data; based on the removed data, dividing the load sequence dataset into a training set and a test set, and normalizing the training set and the test set to obtain the processed data.

[0062] It is understood that the embodiments of this application may use linear interpolation to impute missing values ​​in the collected load sequence dataset; use global threshold detection, box plots, and other methods to handle outliers in the collected load sequence dataset; and use Min-Max standardization to normalize the collected load sequence dataset.

[0063] In actual implementation, firstly, in terms of data preprocessing, the embodiments of this application can clean the data, handle outliers, and handle missing values ​​to remove erroneous information in the load sequence dataset, ensuring data consistency, accuracy, and quality. The dataset is divided as follows: the first 80% of the dataset is selected as the training set, and the last 20% is selected as the test set. The training and test sets are then normalized to generate processed data, making the input data have a more stable gradient and reducing the magnitude difference between features, thereby improving the accuracy of the model, ensuring that the dynamic characteristics of the power system load are fully captured, and evaluating the predictive performance of the model on unseen datasets, providing a reliable foundation for model training and validation.

[0064] In step S102, the high-dimensional matrix is ​​rearranged using a rotation matrix mask to generate a high-dimensional matrix that meets preset conditions, and high-order duplicate terms are deleted from the high-dimensional matrix that meets the preset conditions to generate the deleted data.

[0065] It is understood that the high-dimensional matrix satisfying the preset conditions in this application embodiment can be an N-order high-dimensional matrix. This application embodiment can solve the problem of repeated elements in high-order models by using a rotation matrix mask, creating an N-order rotation matrix mask, which is formally equivalent to the N-order input matrix X. N They have the exact same form.

[0066] Specifically, the rotation matrix mask in this application embodiment can mask X after N-order Kronecker product. N The data is then processed to generate an N-order high-dimensional matrix, which ignores high-order duplicates to generate the deleted data.

[0067] This application's embodiments introduce a rotation matrix masking technique, which effectively eliminates the redundancy caused by the Kronecker product, reduces the complexity of the model, and improves the model efficiency.

[0068] It should be noted that the preset conditions can be set by those skilled in the art according to the actual situation, and no specific restrictions are imposed here.

[0069] In one embodiment of this application, the formula for calculating the elements in the rotation matrix mask is as follows:

[0070]

[0071] Where i, j, n are the position codes of the corresponding dimensions of the mask matrix, and m represents the mask matrix. For an element at a certain position, the value of the element at that position is 1 if and only if i≤j≤…≤n, and 0 otherwise.

[0072] In step S103, based on the deleted data, a preset kernel function is applied to input matrices of different orders to convert the input matrices into output matrices of the corresponding orders.

[0073] It is understood that the preset kernel function in the embodiments of this application can be N kernel functions.

[0074] In actual implementation, the embodiments of this application can define N kernel functions at the kernel function layer, apply the kernel functions to the input matrices of corresponding orders, effectively extract the key features in the input matrices, and convert them into output matrices of corresponding orders through nonlinear mapping.

[0075] In step S104, the output matrix of the preset multilayer perceptron is processed to generate the final prediction result.

[0076] It is understood that the output matrix in the embodiments of this application is concatenated and then further processed in the MLP. Unlike the traditional MLP, the MLP used in this application does not contain a non-linear activation function.

[0077] In practical implementation, this embodiment of the application concatenates the output matrix through a multilayer perceptron layer and outputs it through a purely linear multilayer perceptron to obtain the final classification result. That is, the output matrix is ​​directly fused through a linear layer using a linear superposition method. After linear superposition of the corresponding order, the final prediction result is obtained and output. This embodiment of the application uses a linear multilayer perceptron, which avoids the black box effect caused by traditional activation functions, improves the interpretability of the model, and makes it easier for users to understand the working principle of the model.

[0078] The following is the mathematical expression of the model structure: y(k)=b+k1×y (1) (k)+…+k n ×y (n) (k), where for any first-order output, y (m) (k) is obtained through the aforementioned operations in this study: y (m) (k)=h m ×X N It is worth noting that the × here is redefined through the einsum operation, and its einsum description is: a…,b…→ab.

[0079] Accurate prediction of long-sequence loads in power systems requires models capable of handling varying lengths of long-sequence load input data. This necessitates testing the performance of the proposed model under different prediction step sizes and comprehensively capturing the dynamic characteristics of the power system. By effectively utilizing these variables, the model should achieve high-precision prediction results, ensuring reliable prediction outcomes under diverse operating conditions.

[0080] It should be noted that, due to the omission of the nonlinear activation function, the prediction process in this application has a clear mathematical expression, which can intuitively show the impact of each input variable on the prediction result, making the model more transparent and reliable, and convenient for users to understand and apply.

[0081] In step S105, based on the final prediction results, the preset long-sequence load prediction model of the power system is evaluated using mean square error and mean absolute error to generate evaluation results.

[0082] It is understood that, in order to evaluate the point prediction performance of this model, the embodiments of this application introduce two error indices, MAE and MSE, to evaluate the prediction accuracy of the model. MAE is the mean of absolute error, which is a general error measurement index. MSE is the mean of the squares of the error between the predicted value and the actual observed value, which is more sensitive to sample points with high error.

[0083] In one embodiment of this application, the formula for calculating the mean square error is:

[0084]

[0085] Among them, y i For the predicted target, This is a predicted value;

[0086] The formula for calculating the mean absolute error is:

[0087]

[0088] Among them, y i For the predicted target, These are predicted values.

[0089] In step S106, based on the evaluation results, the relationships and influences between the internal variables of the preset power system long-sequence load forecasting model are analyzed to generate analysis results, and the optimization scheme of the preset power system long-sequence load forecasting model is determined according to the analysis results.

[0090] As one possible approach, embodiments of this application include interpretability analysis capabilities, which can reveal the relationships between internal variables of a pre-defined long-sequence load forecasting model for a power system and their impact on the forecasting results. Based on these analysis results, users can formulate corresponding optimization schemes to further improve the model's performance.

[0091] It should be noted that the interpretability of the model in this application embodiment is as follows: it not only makes accurate predictions but also intuitively expresses the relationships between variables. The model parameters can reflect the specific contribution and mutual influence of each variable to the prediction results, helping users understand the internal mechanism of the model. It can perform detailed impact analysis on variables and quantify the degree of influence of different variables on the predicted future power system load. Through techniques such as feature importance analysis and partial dependency graphs, the role and function of each variable in the prediction are clarified.

[0092] Furthermore, the optimization strategies for power system load systems are proposed and validated: Utilizing the variable relationships and impact analyses provided by the model, specific optimization schemes are formulated, such as adjusting the supply and demand relationship of the power system. Through simulation and practical application, the effectiveness of these optimization strategies is verified, improving the efficiency and stability of the power system and ensuring that the system's equipment operates stably under optimal load conditions.

[0093] The embodiments of this application can outperform existing best models on multiple datasets and significantly reduce the computational requirements for fitting.

[0094] Specifically, it can be combined with Figures 2 to 4 As shown, the working principle of the long-sequence load forecasting method for power systems in this application is explained in detail with a specific embodiment.

[0095] like Figure 2 As shown, the core prediction model in this application embodiment includes: Multilayer Perceptron (MLP), Kronecker product layer, kernel function, and rotation matrix mask. It aims to achieve accurate prediction of power system load, intuitively express the relationship between variables, conduct detailed variable impact analysis, and provide effective system optimization strategies. This will further help the power system achieve supply and demand balance and help power companies achieve more reasonable power allocation through more accurate load prediction.

[0096] Furthermore, Figure 3 This is a schematic diagram of the overall algorithm of the prediction model. Figure 4 The power system load dataset is first preprocessed by a preprocessing module before being input into the prediction model. The model then processes the data to predict and output the power system load parameters for future times.

[0097] The long-series load forecasting method for power systems proposed in this application combines large models with techniques such as Kronecker product and rotation matrix masking. This achieves accurate power system load forecasting, intuitively expresses variable relationships, performs detailed variable impact analysis, and provides effective system optimization strategies. Furthermore, it helps power systems achieve supply-demand balance and assists power companies in achieving more rational power allocation through more accurate load forecasting. This solves the problems in related technologies, such as the massive data volume, the black-box nature of nonlinear characteristics expressed through complex activation functions, and the difficulty in intuitively expressing cross-variable relationships in existing large models for time series forecasting. These limitations restrict the effectiveness and interpretability of the models in practical applications, affecting users' understanding and trust in the model's forecast results.

[0098] Next, with reference to the accompanying drawings, a long-sequence load forecasting device for a power system according to an embodiment of this application is described.

[0099] Figure 5 This is a schematic diagram of the structure of a long-sequence load forecasting device for a power system according to an embodiment of this application.

[0100] like Figure 5 As shown, the long-sequence load forecasting device 10 for this power system includes:

[0101] Specifically, module 100 is used to perform multi-level operations on the input matrix using the Kronecker product layer to construct a high-dimensional matrix.

[0102] The sorting module 200 is used to sort the high-dimensional matrix using a rotation matrix mask to generate a high-dimensional matrix that meets preset conditions, and to delete high-order duplicate terms from the high-dimensional matrix that meets the preset conditions to generate the deleted data.

[0103] The input module 300 is used to apply a preset kernel function to input matrices of different orders based on the deleted data, so as to convert the input matrix into an output matrix of the corresponding order.

[0104] The processing module 400 is used to process the output matrix according to the preset multilayer perceptron to generate the final prediction result.

[0105] Evaluation module 500 is used to evaluate a preset long-sequence load forecasting model for the power system based on the final forecast results, using mean square error and mean absolute error, to generate evaluation results.

[0106] The prediction module 600 is used to analyze the relationships and influences between internal variables of the preset long-sequence load prediction model of the power system based on the evaluation results, so as to generate analysis results and determine the optimization scheme of the preset long-sequence load prediction model of the power system based on the analysis results.

[0107] Optionally, in one embodiment of this application, the long-sequence load forecasting device 10 for a power system further includes a processing module and a generation module.

[0108] The removal module is used to perform data cleaning, missing value imputation, and outlier processing on the load sequence dataset of the power system before using the Kronecker product layer to perform multi-level operations on the input matrix to construct a high-dimensional matrix, so as to remove erroneous information in the load sequence dataset and obtain the removed data.

[0109] The generation module is used to divide the load sequence dataset into training and test sets based on the removed data, and to normalize the training and test sets to obtain the processed data.

[0110] Optionally, in one embodiment of this application, the formula for calculating the high-dimensional matrix is:

[0111] X N (i,j,…,n)=X e (i)×X e (j)×…×X e (n)'

[0112] Among them, X e (n) is obtained by transforming the dimension of the input matrix X. N A matrix consists of N X e The result obtained from the Kronecker product of matrices.

[0113] Optionally, in one embodiment of this application, the formula for calculating the elements in the rotation matrix mask is:

[0114]

[0115] Where i, j, n are the positional codes of the corresponding dimensions of the mask matrix, and m represents the mask matrix.

[0116] Optionally, in one embodiment of this application, the formula for calculating the mean square error is:

[0117]

[0118] Among them, y i For the predicted target, This is a predicted value;

[0119] The formula for calculating the mean absolute error is:

[0120]

[0121] Among them, y i For the predicted target, These are predicted values.

[0122] It should be noted that the foregoing explanation of the long-sequence load forecasting method embodiment for power systems also applies to the long-sequence load forecasting device for power systems in this embodiment, and will not be repeated here.

[0123] The long-sequence load forecasting device for power systems proposed in this application combines large models with techniques such as Kronecker product and rotation matrix masking to achieve accurate long-sequence load forecasting, intuitively express variable relationships, and provide effective system optimization strategies. It also significantly improves feature extraction capabilities, model robustness, model interpretability, and long-sequence forecasting, providing a more effective and reliable method for power system load forecasting. This solves the problems in related technologies, such as the large volume of existing large models in time series forecasting, the black-box nature of nonlinear characteristics expressed through complex activation functions, and the difficulty in intuitively expressing cross-variable relationships. These limitations restrict the effectiveness and interpretability of the models in practical applications, affecting users' understanding and trust in the model's prediction results.

[0124] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0125] The memory 601, the processor 602, and the computer program stored on the memory 601 and capable of running on the processor 602.

[0126] When the processor 602 executes the program, it implements the long-sequence load forecasting method for the power system provided in the above embodiments.

[0127] Furthermore, electronic devices also include:

[0128] Communication interface 603 is used for communication between memory 601 and processor 602.

[0129] The memory 601 is used to store computer programs that can run on the processor 602.

[0130] The memory 601 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0131] If the memory 601, processor 602, and communication interface 603 are implemented independently, then the communication interface 603, memory 601, and processor 602 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0132] Optionally, in a specific implementation, if the memory 601, processor 602, and communication interface 603 are integrated on a single chip, then the memory 601, processor 602, and communication interface 603 can communicate with each other through an internal interface.

[0133] The processor 602 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0134] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described long-sequence load forecasting method for a power system.

[0135] This application also provides a computer program product storing a computer program that, when executed by a processor, implements the above-described long-sequence load forecasting method for power systems.

[0136] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. 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. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0137] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0138] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0139] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0140] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0141] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0142] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0143] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A long-sequence load forecasting method for power systems, characterized in that, Includes the following steps: The input matrix is ​​processed through multiple stages of Kronecker multiplication layers to construct a high-dimensional matrix. The high-dimensional matrix is ​​rearranged using a rotation matrix mask to generate a high-dimensional matrix that meets preset conditions. Then, high-order duplicate terms are removed from the high-dimensional matrix that meets the preset conditions to generate the deleted data. Based on the deleted data, a preset kernel function is applied to input matrices of different orders to convert the input matrices into output matrices of the corresponding orders; The output matrix is ​​processed by a preset multilayer perceptron to generate the final prediction result; Based on the final prediction results, the preset long-sequence load prediction model of the power system is evaluated using mean square error and mean absolute error to generate evaluation results; Based on the evaluation results, the relationships and influences between the internal variables of the preset power system long-sequence load forecasting model are analyzed to generate analysis results, and the optimization scheme of the preset power system long-sequence load forecasting model is determined according to the analysis results. Before using the Kronecker product layer to perform multi-level operations on the input matrix to construct a high-dimensional matrix, the process further includes: performing data cleaning, missing value imputation, and outlier processing on the load sequence dataset of the power system to remove erroneous information from the load sequence dataset and obtain the removed data; based on the removed data, dividing the load sequence dataset into a training set and a test set, and normalizing the training set and the test set to obtain the processed data; The output matrices are concatenated and then further processed in the MLP, where the MLP does not contain a non-linear activation function. The output matrix is ​​concatenated through a multilayer perceptron layer and output through a purely linear multilayer perceptron to obtain the final classification result. That is, the output matrix is ​​directly fused through a linear layer using a linear superposition method. After linear superposition of the corresponding order, the final prediction result is obtained and output.

2. The method according to claim 1, characterized in that, The formula for calculating the high-dimensional matrix is: ， in, For the input matrix Obtained through dimensional transformation A matrix is ​​composed of indivual The result obtained from the Kronecker product of matrices.

3. The method according to claim 1, characterized in that, The formula for calculating the elements in the rotation matrix mask is: ， in, All of these are positional codes corresponding to the dimensions of the mask matrix. This represents the mask matrix; The formula for calculating the mean square error is: ， in, For the predicted target, This is a predicted value; The formula for calculating the mean absolute error is: ， in, For the predicted target, The predicted value; This also includes: the Kronecker product layer represents the higher-order nonlinear relationships of the system, which generates an N-order input matrix by passing the input matrix through an N-order Kronecker product; for any input matrix... The following discussion establishes the sequence number mapping relationship before and after `expand`: ,in: .

4. A long-sequence load forecasting device for a power system, characterized in that, The long-sequence load forecasting method for power systems as described in any one of claims 1-3 includes: The building blocks are used to perform multi-level operations on the input matrix using Kronecker multiplication layers to construct high-dimensional matrices; The sorting module is used to sort the high-dimensional matrix using a rotation matrix mask to generate a high-dimensional matrix that meets preset conditions, and to delete high-order duplicates from the high-dimensional matrix that meets the preset conditions to generate the deleted data. The input module is used to apply a preset kernel function to input matrices of different orders based on the deleted data, so as to convert the input matrices into output matrices of corresponding orders; The processing module is used to process the output matrix according to a preset multilayer perceptron to generate the final prediction result; The evaluation module is used to evaluate the preset long-sequence load forecasting model of the power system based on the final prediction results, using mean square error and mean absolute error, so as to generate evaluation results; The prediction module is used to analyze the relationships and influences between the internal variables of the preset power system long-sequence load prediction model based on the evaluation results, so as to generate analysis results and determine the optimization scheme of the preset power system long-sequence load prediction model according to the analysis results. The system further includes: a removal module, used to perform data cleaning, missing value imputation, and outlier processing on the load sequence dataset of the power system before using the Kronecker product layer to perform multi-level operations on the input matrix to construct a high-dimensional matrix, in order to remove erroneous information from the load sequence dataset and obtain the removed data; and a generation module, used to divide the load sequence dataset into a training set and a test set based on the removed data, and to normalize the training set and the test set to obtain the processed data. The output matrices are concatenated and then further processed in the MLP, where the MLP does not contain a non-linear activation function. The output matrix is ​​concatenated through a multilayer perceptron layer and output through a purely linear multilayer perceptron to obtain the final classification result. That is, the output matrix is ​​directly fused through a linear layer using a linear superposition method. After linear superposition of the corresponding order, the final prediction result is obtained and output.

5. The apparatus according to claim 4, characterized in that, The formula for calculating the elements in the rotation matrix mask is: ， in, All of these are positional codes corresponding to the dimensions of the mask matrix. This represents the mask matrix; The formula for calculating the mean square error is: ， in, For the predicted target, This is a predicted value; The formula for calculating the mean absolute error is: ， in, For the predicted target, The predicted value; This also includes: the Kronecker product layer represents the higher-order nonlinear relationships of the system, which generates an N-order input matrix by passing the input matrix through an N-order Kronecker product; for any input matrix... The following discussion establishes the sequence number mapping relationship before and after `expand`: ,in: .

6. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the long-sequence load forecasting method for a power system as described in any one of claims 1-3.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by a processor to implement the long-sequence load forecasting method for power systems as described in any one of claims 1-3.

8. A computer program product, comprising a computer program, characterized in that, The computer program is executed to implement the long-sequence load forecasting method for power systems as described in any one of claims 1-3.

Images