A method and apparatus for power load forecasting based on multi-head attention mechanism fusion graph convolutional network and time series data.

Abstract

This application provides a method and apparatus for power load forecasting based on a multi-head attention mechanism fusion graph convolutional network and time series data. The method includes: constructing a power system topology model based on the topological relationships between devices in a power system; obtaining the node features of each node in the power system topology model and the node features of its neighboring nodes, and using a multi-head attention mechanism fusion graph convolutional network to perform feature aggregation and concatenation of the node features of each node and the node features of its neighboring nodes using multiple attention heads to obtain updated node features for each node; obtaining the historical time series features of each node, and fusing the historical time series features of each node with the updated node features to obtain fused feature data; and inputting the fused feature data into a trained deep neural network prediction model to obtain the power load forecast result for each node.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and apparatus for predicting power load based on a multi-head attention mechanism that fuses graph convolutional networks and time series data. Background Technology

[0002] In modern power systems, numerous methods exist for power load forecasting. Traditional methods, such as regression analysis, time series analysis, and grey models, offer some effectiveness and stability when handling simple, linear load changes. However, these methods often struggle to accurately capture the complex nonlinear characteristics and uncertainties of power loads. Intelligent forecasting methods, such as artificial neural networks, support vector machines, and fuzzy logic, while advantageous in handling complex relationships, also have limitations. For instance, neural networks are prone to getting trapped in local optima, leading to inaccurate predictions. Furthermore, current technologies rely solely on historical load and equipment power data, lacking a comprehensive depiction of the overall plant operation system and missing correlations between different components, thus affecting forecast accuracy. Summary of the Invention

[0003] In view of this, embodiments of this application provide a power load forecasting method based on a multi-head attention mechanism fusing graph convolutional networks and time series data. One or more embodiments of this application also relate to a power load forecasting apparatus based on a multi-head attention mechanism fusing graph convolutional networks and time series data, a computing device, a computer-readable storage medium, and a computer program, to address the technical deficiencies existing in the prior art.

[0004] According to a first aspect of the embodiments of this application, a power load forecasting method based on a multi-head attention mechanism fusing graph convolutional networks and time series data is provided, comprising:

[0005] Obtain the topological relationships between devices in a power system, and construct a power system topology model based on these relationships.

[0006] The node features of each node and its neighboring nodes in the power system topology model are obtained, and the node features of each node and its neighboring nodes are aggregated and concatenated by multiple attention heads using a multi-head attention mechanism to obtain the updated node features of each node.

[0007] The historical time series features of each node in the power system topology model are obtained, and the historical time series features of each node are fused with the updated node features to obtain the fused feature data of each node.

[0008] By inputting the fused feature data of each node into a trained deep neural network prediction model, the power load prediction result of each node is obtained.

[0009] Preferably, constructing a power system topology model based on the topological relationships between devices in the power system includes:

[0010] Based on the topological relationship between devices in the power system, each device node in the power system is modeled as a node, the current flow relationship or load dependency relationship between device nodes is modeled as an edge, and the device node that has a current flow relationship or load dependency relationship with the device node is modeled as a neighbor node.

[0011] Construct a power system topology model based on all nodes, edges, and neighboring nodes in the power system.

[0012] Preferably, the multi-head attention mechanism fusion graph convolutional network performs feature aggregation and concatenation of multiple attention heads on the node features of each node in the power system topology model and the node features of its neighboring nodes to obtain the updated node features of each node, including:

[0013] By performing a linear transformation of the node features of each node and the node features of its neighboring nodes using the kth attention head, the linear transformation features of the kth attention head of each node and the linear transformation features of the kth attention head of its neighboring nodes are obtained.

[0014] Using the linear transformation characteristics of the kth attention head of each node and the linear transformation characteristics of the kth attention head of the neighboring nodes, the attention coefficient of the kth attention head between each node and its neighboring nodes is calculated.

[0015] Using the attention coefficients of the kth attention head between each node and its neighboring nodes and the linear transformation characteristics of the kth attention head of its neighboring nodes, the aggregation characteristics of the kth attention head of each node are calculated;

[0016] The updated node features of each node are obtained by concatenating the aggregated features of all heads of each node.

[0017] Preferably, the step of concatenating the aggregated features of all heads of each node to obtain the updated node features of each node includes:

[0018]

[0019] in, Represents a node Updated node features; Represents a node The aggregated features of the k-th attention head; || represents the concatenation operation; k represents the k-th attention head; K is the number of multi-head attention mechanisms.

[0020] Preferably, the historical time series characteristics of each node are historical load time series characteristics or historical weather data time series characteristics.

[0021] Preferably, the step of obtaining the power load prediction result for each node by inputting the fused feature data of each node into a trained deep neural network prediction model includes:

[0022]

[0023] in: Z represents the power load forecast result for each node; Z represents the fused feature data for each node. It is the weight matrix of the output layer of a deep neural network prediction model; It is a bias term.

[0024] Preferably, it further includes:

[0025] Obtain the topological relationships between devices in multiple power systems and the actual power load results of each device, and construct a power system topology model based on the topological relationships between devices in each power system;

[0026] The node features of each node and its neighboring nodes in the power system topology model are obtained, and the node features of each node and its neighboring nodes are aggregated and concatenated by multiple attention heads using a multi-head attention mechanism to obtain the updated node features of each node.

[0027] The historical time series features of each node in the power system topology model are obtained, and the historical time series features of each node are fused with the updated node features to obtain the fused feature data of each node.

[0028] The deep neural network prediction model is trained based on the fused feature data of each node and the actual power load results of each device to obtain the trained deep neural network prediction model.

[0029] According to a second aspect of the embodiments of this application, a power load forecasting device based on a multi-head attention mechanism fusing graph convolutional networks and time series data is provided, comprising:

[0030] The acquisition module is configured to acquire the topological relationships between devices in the power system and construct a power system topology model based on the topological relationships between devices in the power system.

[0031] The aggregation and splicing module is configured to obtain the node features of each node in the power system topology model and the node features of its neighboring nodes, and to use a multi-head attention mechanism to fuse the graph convolutional network to perform feature aggregation and splicing of the node features of each node in the power system topology model and the node features of its neighboring nodes with multiple attention heads to obtain the updated node features of each node.

[0032] The fusion processing module is configured to acquire the historical time series features of each node in the power system topology model, and to obtain the fused feature data of each node by fusing the historical time series features of each node with the updated node features.

[0033] The prediction module is configured to obtain the power load prediction result of each node by inputting the fused feature data of each node into a trained deep neural network prediction model.

[0034] According to a third aspect of the embodiments of this application, a computing device is provided, comprising:

[0035] Memory and processor;

[0036] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement any of the steps of the power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series.

[0037] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of any one of the power load forecasting methods based on a multi-head attention mechanism fusion graph convolutional network and time series.

[0038] According to a fifth aspect of the present application, a computer program is provided, wherein when the computer program is executed in a computer, the computer is instructed to perform the steps of the above-described power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series.

[0039] This application provides a power load forecasting scheme based on a multi-head attention mechanism fusion graph convolutional network and time series data. The scheme acquires the topological relationships between devices in a power system and constructs a power system topology model based on these relationships. It then acquires the node features of each node in the power system topology model and the node features of its neighboring nodes. A multi-head attention mechanism fusion graph convolutional network is used to aggregate and concatenate the node features of each node and its neighboring nodes using multiple attention heads to obtain updated node features for each node. Finally, the scheme acquires the historical time series features of each node in the power system topology model and fuses these features with the updated node features to obtain fused feature data for each node. This fused feature data is then input into a trained deep neural network prediction model to obtain the power load forecasting result for each node. This application dynamically selects the importance of neighboring nodes, more accurately capturing complex relationships between nodes and improving the model's understanding of power network structural features. By introducing a multi-head attention mechanism, the model's expressive power and robustness are enhanced, comprehensively considering the influence of multiple neighboring nodes and avoiding information loss and single attention bias. By combining the advantages of graph convolutional networks and time-series data, and fully utilizing the topological relationships between devices and historical load variation patterns, this method achieves high accuracy and stability in load forecasting. Adaptable to complex power demand scenarios, this approach provides energy storage companies with more scientific load forecasting and decision support, promoting efficient operation and optimized management of the power system. Attached Figure Description

[0040] Figure 1 This is a flowchart of a power load forecasting method based on a multi-head attention mechanism that fuses graph convolutional networks and time series data, provided in one embodiment of this application.

[0041] Figure 2 This is a flowchart of a deep neural network prediction model training method provided in one embodiment of this application;

[0042] Figure 3 This is a schematic diagram of an embodiment of the power load forecasting device based on a multi-head attention mechanism that fuses a graph convolutional network and a time series, provided in this application.

[0043] Figure 4 This is a structural block diagram of a computing device provided in one embodiment of this application. Detailed Implementation

[0044] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0045] The terminology used in one or more embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of one or more embodiments of this application. The singular forms “a,” “the,” and “the” used in one or more embodiments of this application and in the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” used in one or more embodiments of this application refers to and includes any or all possible combinations of one or more associated listed items.

[0046] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this application, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0047] This application provides a power load forecasting method based on a multi-head attention mechanism that fuses a graph convolutional network and a time series. This application also relates to a power load forecasting device based on a multi-head attention mechanism that fuses a graph convolutional network and a time series, a computing device, a computer-readable storage medium, and a computer program, which will be described in detail in the following embodiments.

[0048] Figure 1 The flowchart illustrates a power load forecasting method based on a multi-head attention mechanism that fuses graph convolutional networks and time series data according to an embodiment of this application, specifically including the following steps.

[0049] Step S101: Obtain the topological relationships between devices in the power system, and construct a power system topology model based on the topological relationships between devices in the power system;

[0050] In one optional implementation, constructing a power system topology model based on the topological relationships between devices in the power system includes: modeling each device node in the power system as a node, modeling the current flow relationship or load dependency relationship between device nodes as an edge, and modeling device nodes that have a current flow relationship or load dependency relationship with the device nodes as neighbor nodes; and constructing a power system topology model based on all the nodes, edges, and neighbor nodes in the power system.

[0051] Step S102: Obtain the node features of each node in the power system topology model and the node features of its neighboring nodes, and use a multi-head attention mechanism to fuse the graph convolutional network to perform feature aggregation and splicing of the node features of each node in the power system topology model and the node features of its neighboring nodes with multiple attention heads to obtain the updated node features of each node.

[0052] In one optional implementation, the multi-head attention mechanism fusion graph convolutional network performs feature aggregation and concatenation of multiple attention heads on the node features of each node in the power system topology model and the node features of its neighboring nodes to obtain the updated node features of each node. This includes: performing a linear transformation of the k-th attention head on the node features of each node and the node features of its neighboring nodes respectively to obtain the linear transformation features of the k-th attention head of each node and the linear transformation features of the k-th attention head of its neighboring nodes; using the linear transformation features of the k-th attention head of each node and the linear transformation features of the k-th attention head of its neighboring nodes, calculating the attention coefficient of the k-th attention head between each node and its neighboring nodes; using the attention coefficient of the k-th attention head between each node and its neighboring nodes and the linear transformation features of the k-th attention head of its neighboring nodes, calculating the aggregated features of the k-th attention head of each node; and concatenating the aggregated features of all heads of each node to obtain the updated node features of each node.

[0053] In one optional implementation, the step of concatenating the aggregated features of all heads of each node to obtain the updated node features of each node includes:

[0054]

[0055] in, Represents a node Updated node features; Represents a node The aggregated features of the k-th attention head; || represents the concatenation operation; k represents the k-th attention head; K is the number of multi-head attention mechanisms.

[0056] Step S103: Obtain the historical time series features of each node in the power system topology model, and obtain the fused feature data of each node by fusing the historical time series features of each node with the updated node features;

[0057] In one alternative implementation, the historical time series characteristics of each node are historical load time series characteristics or historical weather data time series characteristics.

[0058] Step S104: By inputting the fused feature data of each node into the trained deep neural network prediction model, the power load prediction result of each node is obtained.

[0059] In one optional implementation, obtaining the power load prediction result for each node by inputting the fused feature data of each node into a trained deep neural network prediction model includes:

[0060]

[0061] in: Z represents the power load forecast result for each node; Z represents the fused feature data for each node. It is the weight matrix of the output layer of a deep neural network prediction model; It is a bias term.

[0062] In an optional implementation, the method further includes: acquiring the topological relationships between devices in multiple power systems and the actual power load results of each device, and constructing a power system topology model based on the topological relationships between devices in each power system; acquiring the node features of each node in the power system topology model and the node features of its neighboring nodes, and using a multi-head attention mechanism to fuse the node features of each node in the power system topology model and the node features of its neighboring nodes to perform feature aggregation and splicing of multiple attention heads to obtain the updated node features of each node; acquiring the historical time series features of each node in the power system topology model, and fusing the historical time series features of each node with the updated node features to obtain fused feature data of each node; and training a deep neural network prediction model based on the fused feature data of each node and the actual power load results of each device to obtain the trained deep neural network prediction model.

[0063] The power load forecasting scheme based on a multi-head attention mechanism, fusing graph convolutional networks and time series data, provided in this application aims to more accurately capture complex relationships between nodes and improve the model's understanding of power network structural characteristics by dynamically selecting the importance of neighboring nodes. By introducing a multi-head attention mechanism, the model's expressive power and robustness are enhanced, comprehensively considering the influence of multiple neighboring nodes and avoiding information loss and single attention bias. Combining the advantages of graph convolutional networks and time series data, and fully utilizing the topological relationships between devices and historical load variation patterns, high accuracy and stability in load forecasting are achieved. This method is adaptable to complex power demand scenarios, providing energy storage companies with more scientific load forecasting and decision support, and promoting efficient operation and optimized management of the power system.

[0064] The technical solution of the present invention will be further described below.

[0065] The definition of nodes and edges: each device (such as a machine, sensor, etc.) will be treated as a node. Indicates. Side Indicates equipment and neighboring equipment The current flow relationship or load dependency between the devices. The weight of the edges can be based on the current flow, power consumption, or other relevant metrics between the devices.

[0066] Example: In a production system, energy storage devices are included. Production equipment In addition, other non-electrical equipment (such as trucks, forklifts, and other auxiliary equipment). The connection weights between these devices can be determined based on the combined ratio of their electricity consumption and power consumption, and adjusted empirically in conjunction with their actual importance in the production process.

[0067] The diagram can be represented as ,in It is a set of nodes. It is a set of edges. Each node has a feature vector. , can be represented as: [Load, current, voltage, equipment status, ...]

[0068] Graph attention networks learn node features by introducing an attention mechanism, enabling them to dynamically adjust the feature aggregation process based on the influence of neighboring nodes. The attention coefficient is calculated for each node. neighboring nodes Calculate the attention coefficient : in: It is a node eigenvectors; Representing neighboring nodes eigenvectors; It is a weight matrix of linear transformation, with parameters configured based on random assignment or historical experience; Ⅱ represents the weight vector used to calculate attention; Ⅱ represents the feature concatenation operation; softmax is the normalization function, and LeakyReLU is the activation function; each node The features are updated as follows: in Represents a node The set of neighbors; Represents a node Updated node features; σ represents the activation function.

[0069] In power grid load forecasting, load changes on some devices have a greater impact on the target device, while those on others have a smaller impact. Introducing an attention mechanism allows the model to automatically learn and identify the importance of different neighboring nodes to the current node. Attention coefficient. These are dynamic weights learned based on node features, representing the actual contribution of neighboring node information to the target node under the current model. The attention mechanism can be viewed as assigning a dynamic, data-driven weight to each edge based on existing edges (current flow relationships). This weight not only considers the existence of the edge but also comprehensively considers the current characteristics of the connected devices, reflecting their varying importance to load forecasting. This is the attention coefficient. This involves the model dynamically calculating the attention given to neighboring devices based on node characteristics (dynamic edge weights), enhancing the model's flexibility in identifying the influence of different neighbors. Through this mechanism, the load impact of neighboring devices can be combined more accurately, improving the overall load prediction performance.

[0070] To enhance the model's expressiveness, different attention weights are multiplied for different node rows, as shown in the following formula: ,in It represents the number of attention heads, where U indicates that the outputs of multiple attention heads are concatenated in sequence. It is a node The set of neighboring nodes. It is a node and his neighbors The attention coefficients between nodes are typically calculated using the following steps: First, obtain node features using a linear transformation. Then, for each pair of nodes compute nodes and his neighbors Attention coefficient between ,here, It is a learnable weight vector used to compute attention, and II represents the concatenation operation. Finally, features are aggregated, and each attention head... The features of neighboring nodes are weighted and aggregated based on the calculated attention coefficients. Specifically, for each attention head, the features of neighboring nodes are... First, linearly transform to Then, based on the calculated attention coefficient Weighted summation, processed by activation function σ, yields the nodes. The output characteristics of this attention head Finally, the outputs of all the heads are concatenated to obtain the node. Update features.

[0071] Specifically, the steps include the following:

[0072] 1) Linear transformation:

[0073] For the current node and its neighboring nodes eigenvectors , Perform linear transformations on the k-th attention head respectively:

[0074]

[0075] 2) Calculate attention energy (unnormalized attention coefficient)

[0076] For the current node Its neighboring nodes Features after concatenation and transformation, calculation:

[0077]

[0078] 3) Normalized attention coefficient

[0079] For all neighboring nodes Perform softmax:

[0080]

[0081] 4) Weighted aggregation neighbor node characteristics

[0082] Feature aggregation is performed as a weighted sum, and an activation function is added:

[0083]

[0084] 5) Multi-point attention splicing

[0085] Concatenate the aggregated outputs of all attention points:

[0086]

[0087] The set of neighboring nodes N(i) is defined based on the set of edges in the graph structure G=(V,E). Specifically:

[0088] Furthermore, for the current node Its neighbor set N(i) is the set of all nodes that are related to node i. via edge A set of directly connected nodes. In a power network scenario, nodes represent devices, and edges represent current flow relationships or load dependencies between devices. Therefore, neighboring nodes are those that are connected to the devices in terms of power flow or dependency. There are directly connected device nodes. This neighbor relationship can be quickly indexed using an adjacency matrix or adjacency list in the graph. For example, if the element Aij in the i-th row and j-th column of adjacency matrix A is not equal to 0, then the node... It is a node The neighbors.

[0089] Time series data processing: Time series data (such as historical load data) can be represented as a matrix. Each row corresponds to a feature of a time step (e.g., load of the past 24 hours, weather data, etc.).

[0090] Feature fusion: Combining the node features aggregated by the above model with time series features: in It is the node feature of the last layer, that is, when When it is L-1. Where, the symbol... This represents the concatenation operation of the feature matrices.

[0091] The prediction layer takes the combined features and inputs them into a fully connected layer for load prediction. in: This is the predicted load value. It is the weight matrix of the output layer. This is the bias term. The mean squared error (MSE) is used as the loss function for model training. in It is the sample size. This is the actual load value.

[0092] The deep neural network prediction model training method used in this invention is based on the backpropagation algorithm and gradient descent optimization technique. By iteratively adjusting network parameters, it improves the model's ability to fit and predict input data. Figure 2 As shown, the specific steps include:

[0093] Step S201: Data preprocessing: Normalize and standardize the node feature data and time series load data aggregated by the input model to ensure the stability of data distribution and improve training efficiency.

[0094] Step S202: Forward propagation: Input the preprocessed data into the deep neural network prediction model, and calculate the output prediction value after linear transformation and nonlinear activation function of each layer of neurons.

[0095] Step S203: Error Calculation: Calculate the difference between the predicted output and the actual load value using a defined loss function (such as mean squared error MSE) to quantify the prediction error of the model.

[0096] Step S204: Backpropagation: Based on the error value, the gradient of each layer's parameters is calculated using the chain rule, and the error information is propagated forward layer by layer to guide the parameter update direction.

[0097] Step S205: Parameter update: Use an optimization algorithm (such as Adam optimizer or stochastic gradient descent SGD) to adjust the network weights and biases according to the calculated gradient, and gradually reduce the prediction error.

[0098] Step S206: Training Iteration: Repeat the forward propagation, error calculation, back propagation and parameter update steps until the preset stopping conditions are met (such as reaching the maximum number of training rounds or the error convergence threshold).

[0099] Step S207: Model Validation and Tuning: Use an independent validation set to evaluate the performance of the trained model, and adjust hyperparameters (such as learning rate, batch size, number of network layers, etc.) to optimize the model performance and effectively prevent overfitting.

[0100] This application combines graph convolutional networks and time-series data-based load forecasting algorithms to fully leverage relationships between devices and historical load patterns, thereby improving the accuracy and robustness of load forecasting. This model design can adapt to complex electricity demand scenarios, providing better decision support for energy storage companies.

[0101] This application also provides an embodiment of a power load forecasting device based on a multi-head attention mechanism that fuses graph convolutional networks and time series data. Figure 3 This diagram illustrates the structure of a power load forecasting device based on a multi-head attention mechanism that fuses a graph convolutional network with time series data, according to an embodiment of this application. Figure 3As shown, the device includes: an acquisition module configured to acquire the topological relationships between devices in a power system and construct a power system topology model based on these relationships; an aggregation and splicing module configured to acquire the node features of each node in the power system topology model and the node features of its neighboring nodes, and to perform feature aggregation and splicing of the node features of each node in the power system topology model and the node features of its neighboring nodes using a multi-head attention mechanism to fuse the graph convolutional network, thereby obtaining updated node features for each node; a fusion processing module configured to acquire the historical time series features of each node in the power system topology model, and to obtain fused feature data for each node by fusing the historical time series features of each node with the updated node features; and a prediction module configured to obtain the power load prediction result for each node by inputting the fused feature data of each node into a trained deep neural network prediction model.

[0102] The above is a schematic scheme of an electricity load forecasting device based on a multi-head attention mechanism fusion graph convolutional network and time series in this embodiment. It should be noted that the technical solution of this electricity load forecasting device based on a multi-head attention mechanism fusion graph convolutional network and time series belongs to the same concept as the above-described electricity load forecasting method based on a multi-head attention mechanism fusion graph convolutional network and time series. Details not described in detail in the technical solution of the electricity load forecasting device based on a multi-head attention mechanism fusion graph convolutional network and time series can be found in the description of the above-described electricity load forecasting method based on a multi-head attention mechanism fusion graph convolutional network and time series.

[0103] Figure 4 A structural block diagram of a computing device 400 according to an embodiment of this application is shown. The components of the computing device 400 include, but are not limited to, a memory 410 and a processor 420. The processor 420 is connected to the memory 410 via a bus 430, and a database 450 is used to store data.

[0104] The computing device 400 also includes an access device 440, which enables the computing device 400 to communicate via one or more networks 460. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 440 may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0105] In one embodiment of this application, the aforementioned components of the computing device 400 and Figure 4 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 4 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this application. Those skilled in the art can add or replace other components as needed.

[0106] The computing device 400 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or PCs. The computing device 400 can also be a mobile or stationary server.

[0107] The processor 420 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-described power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series.

[0108] The above is a schematic scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device belongs to the same concept as the above-described power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series. For details not described in detail in the technical solution of the computing device, please refer to the description of the above-described power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series.

[0109] An embodiment of this application also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described power load forecasting method based on a multi-head attention mechanism fusion graph convolutional network and time series.

[0110] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the above-described power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series. For details not described in detail in the technical solution of the storage medium, please refer to the description of the above-described power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series.

[0111] An embodiment of this application also provides a computer program, wherein when the computer program is executed in a computer, the computer is instructed to perform the steps of the above-described power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series.

[0112] The above is an illustrative scheme of a computer program according to this embodiment. It should be noted that the technical solution of this computer program belongs to the same concept as the above-described power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series. For details not described in detail in the technical solution of the computer program, please refer to the description of the above-described power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series.

[0113] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0114] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0115] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of this application.

[0116] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0117] The preferred embodiments disclosed above are merely illustrative of this application. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments of this application. These embodiments are selected and specifically described in this application to better explain the principles and practical applications of the embodiments of this application, thereby enabling those skilled in the art to better understand and utilize this application. This application is limited only by the claims and their full scope and equivalents.

Claims

1. A power load forecasting method based on a multi-head attention mechanism that fuses graph convolutional networks and time series data, characterized in that, include: Obtain the topological relationships between devices in a power system, and construct a power system topology model based on these relationships. This includes: modeling each device node in the power system as a node, modeling the current flow relationships or load dependencies between device nodes as edges, and modeling device nodes with current flow relationships or load dependencies as neighbor nodes; and constructing a power system topology model based on all the nodes, edges, and neighbor nodes in the power system. The process involves obtaining the node features of each node and its neighboring nodes in the power system topology model, and then using a multi-head attention mechanism to fuse the graph convolutional network to aggregate and concatenate the node features of each node and its neighboring nodes using multiple attention heads. This yields the updated node features for each node. The process includes: performing a linear transformation of the k-th attention head on each node's node features and the node features of its neighboring nodes, resulting in linear transformation features of the k-th attention head for each node and its neighboring nodes; calculating the attention coefficient between each node and its neighboring nodes using the linear transformation features of the k-th attention head and the linear transformation features of the k-th attention head; calculating the aggregated feature of the k-th attention head for each node using the attention coefficient and the linear transformation feature of the k-th attention head; and concatenating the aggregated features of all heads for each node to obtain the updated node features for each node. The historical time series features of each node in the power system topology model are obtained, and the historical time series features of each node are fused with the updated node features to obtain the fused feature data of each node. By inputting the fused feature data of each node into a trained deep neural network prediction model, the power load prediction result of each node is obtained, including: ； in: Z represents the power load forecast result for each node; Z represents the fused feature data for each node. It is the weight matrix of the output layer of a deep neural network prediction model; It is a bias term.

2. The method according to claim 1, characterized in that, The updated node features of each node are obtained by concatenating the aggregated features of all heads of each node. ； in, Represents a node Updated node features; Represents a node The aggregated features of the k-th attention head; || represents the concatenation operation; k represents the k-th attention head; K is the number of multi-head attention mechanisms.

3. The method according to claim 1, characterized in that, The historical time series characteristics of each node are either historical load time series characteristics or historical weather data time series characteristics.

4. The method according to claim 1, characterized in that, Also includes: Obtain the topological relationships between devices in multiple power systems and the actual power load results of each device, and construct a power system topology model based on the topological relationships between devices in each power system; The node features of each node and its neighboring nodes in the power system topology model are obtained, and the node features of each node and its neighboring nodes are aggregated and concatenated by multiple attention heads using a multi-head attention mechanism to obtain the updated node features of each node. The historical time series features of each node in the power system topology model are obtained, and the historical time series features of each node are fused with the updated node features to obtain the fused feature data of each node. The deep neural network prediction model is trained based on the fused feature data of each node and the actual power load results of each device to obtain the trained deep neural network prediction model.

5. A power load forecasting device based on a multi-head attention mechanism fusing graph convolutional networks and time series data, the device being used to implement the method as described in any one of claims 1-4, characterized in that, include: The acquisition module is configured to acquire the topological relationships between devices in the power system and construct a power system topology model based on the topological relationships between devices in the power system. The aggregation and splicing module is configured to obtain the node features of each node in the power system topology model and the node features of its neighboring nodes, and to use a multi-head attention mechanism to fuse the graph convolutional network to perform feature aggregation and splicing of the node features of each node in the power system topology model and the node features of its neighboring nodes with multiple attention heads to obtain the updated node features of each node. The fusion processing module is configured to acquire the historical time series features of each node in the power system topology model, and to obtain the fused feature data of each node by fusing the historical time series features of each node with the updated node features. The prediction module is configured to obtain the power load prediction result of each node by inputting the fused feature data of each node into a trained deep neural network prediction model.

6. A computing device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the power load forecasting method based on multi-head attention mechanism fusion graph convolutional network and time series as described in any one of claims 1 to 4.

Images