Regional electrical load prediction method based on graph structure and related equipment

Abstract

The invention relates to the technical field of power systems, and discloses a regional electrical load prediction method based on a graph structure and related equipment, and the method comprises the steps: constructing a multivariate feature data set; taking each reference area as a node, and constructing a target graph structure according to the geographic information corresponding to each reference area; training a preset basic prediction model based on the multivariate feature data set and the target graph structure to obtain an electrical load prediction model; and when an electrical load prediction request for a target area is received, predicting the electrical load of the target area based on the electrical load prediction request and an electrical load prediction model. The method can solve the problem that a current regional power load prediction method is poor in prediction precision and robustness in rural and high-speed service area scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a regional power load prediction method based on a graph structure and related equipment. BACKGROUND

[0002] With the increasing popularity of new energy vehicles, the charging load in tourist areas and business areas shows significant volatility and uncertainty. Accurate load prediction is crucial for ensuring regional power supply stability, optimizing charging pile utilization, and achieving efficient energy scheduling.

[0003] Current power load prediction schemes usually rely on time series analysis (such as ARIMA) or classical machine learning (such as SVR) for prediction. However, relying solely on historical patterns of load has weak adaptability to sudden and strongly fluctuating loads, and does not consider spatial dependencies, making it difficult to capture the coupling characteristics of regional multi-node loads. In rural and highway service area scenarios, prediction accuracy and robustness are insufficient. SUMMARY

[0004] The embodiments of the present application provide a regional power load prediction method based on a graph structure and related equipment, which can solve the problem of insufficient prediction accuracy and robustness of current regional power load prediction methods in rural and highway service area scenarios.

[0005] In a first aspect, the embodiments of the present application provide a regional power load prediction method based on a graph structure, comprising: constructing a multi-element feature dataset; constructing a target graph structure based on the geographical information of each reference region; training a preset basic prediction model based on the multi-element feature dataset and the target graph structure to obtain a power load prediction model; when receiving a power load prediction request for a target region, predicting the power load of the target region based on the power load prediction request and the power load prediction model.

[0006] Optionally, in some embodiments of the present application, the target graph structure is constructed based on the geographical information of each reference region, comprising: determining a plurality of reference regions in the region library according to a preset power load prediction strategy; constructing an original adjacency matrix based on the geographical information of each reference region; adjusting the original adjacency matrix based on the region information of each reference region to obtain a target graph structure.

[0007] Optionally, in some embodiments of the present application, the original adjacency matrix is constructed based on the geographical information of each reference region, comprising: Construct a basic matrix using each reference region as a node; Based on the geographic information corresponding to each reference area, calculate the geographic distance corresponding to each reference area; The base matrix is ​​adjusted based on the geographical distance to obtain the original adjacency matrix.

[0008] Optionally, in some embodiments of this application, adjusting the original adjacency matrix according to the region information corresponding to each reference region to obtain the target graph structure includes: The element values ​​of the original adjacency matrix are adjusted according to the region information corresponding to each reference region; The region information corresponding to the reference region is added to the adjusted original adjacency matrix, and the original adjacency matrix after adding the information is normalized to obtain the normalized adjacency matrix. The attribute information of the reference region is fused with the normalized adjacency matrix to obtain the target graph structure.

[0009] Optionally, in some embodiments of this application, the step of training a preset basic prediction model based on the multivariate feature dataset and the target graph structure to obtain an electricity load prediction model may specifically include: The multivariate feature set is divided to obtain a training set, a validation set, and a test set. The preset basic prediction model is trained based on the training set and the target graph structure, and the trained prediction model is validated using the validation set. When the trained prediction model meets the preset conditions, the trained prediction model is tested based on the test set; When the trained prediction model meets the testing criteria, the prediction model that passes the test is determined as the electricity load prediction model.

[0010] Optionally, in some embodiments of this application, the step of predicting the electricity load of the target area based on the electricity load prediction request and the electricity load prediction model when a power load prediction request for the target area is received includes: When a power load forecasting request for the target area is received, the forecasting time range and the target area identifier are obtained from the power load forecasting request. Based on the target area identifier, the target electricity load prediction model corresponding to the target area is determined in the electricity load prediction model corresponding to each area; Based on the predicted time range, real-time characteristic data of the target area, and the target electricity load prediction model, the electricity load of the target area is predicted.

[0011] Optionally, in some embodiments of this application, constructing a multivariate feature dataset includes: Obtain regional data from multiple reference areas; Divide multiple reference regions into regions based on the region type of each reference region; Based on the segmentation results and regional data from multiple reference regions, a multivariate feature dataset corresponding to different region types is constructed.

[0012] Secondly, embodiments of this application provide a graph-based regional electricity load forecasting device, comprising: The first building module is used to construct a multivariate feature dataset; The second construction module is used to construct the target map structure based on the geographic information corresponding to each reference area, with each reference area as a node. The training module is used to train the preset basic prediction model based on the multivariate feature dataset and the target graph structure to obtain the electricity load prediction model. The prediction module is used to predict the electricity load of the target area based on the electricity load prediction request and the electricity load prediction model when a power load prediction request for the target area is received.

[0013] Thirdly, embodiments of this application provide 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 computer program to implement the steps of the graph-based regional power load forecasting method as described in the first aspect.

[0014] Fourthly, embodiments of this application provide a storage medium storing a computer program capable of being loaded by a processor and executing the graph-based regional electricity load forecasting method as described in the first aspect.

[0015] This application provides a graph-based method, apparatus, electronic device, and storage medium for regional electricity load forecasting. After constructing a multivariate feature dataset, a target graph structure is built using reference regions as nodes and based on the corresponding geographical information of each reference region. Then, a preset basic prediction model is trained based on the multivariate feature dataset and the target graph structure to obtain an electricity load forecasting model. When a request for electricity load forecasting for a target region is received, the electricity load of the target region is predicted based on the request and the electricity load forecasting model. In the graph-based regional electricity load forecasting scheme provided in this application, constructing a target graph structure using reference regions as nodes and combining geographical information can capture the spatial dependence and coupling characteristics of multiple nodes in a region, overcoming the shortcomings of traditional methods that lack spatial correlation. The model training integrates both methods, taking into account spatiotemporal characteristics, and improving the prediction accuracy and robustness in scenarios such as rural areas and highway service areas. Finally, the system adapts to the target region in response to prediction requests, providing support for power supply stability, charging pile optimization, and energy dispatch. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is an application environment diagram of the regional electricity load prediction method based on graph structure provided in the embodiments of this application; Figure 2 This is a flowchart illustrating the regional electricity load forecasting method based on graph structure provided in this application embodiment; Figure 3 This is a simulation diagram of rural load forecasting provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of the graph-based regional power load prediction device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of systems and methods consistent with those detailed in the appended claims or with some aspects of this application.

[0019] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover descriptions such as non-exclusive inclusion, so that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, components, features, and elements with the same names in different embodiments of this application may have the same meaning or different meanings, the specific meaning of which must be determined by its interpretation in that specific embodiment or further in conjunction with the context of that specific embodiment.

[0020] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0021] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, "module," "part," or "unit" may be used interchangeably.

[0022] To address the aforementioned technical problems and overcome the shortcomings of existing technologies, this application provides a graph-based method for predicting regional power load and related equipment, which can accurately determine and quickly protect fault areas in active distribution networks, thereby improving the reliability and adaptability of differential protection in active distribution networks.

[0023] Figure 1 This is an application environment diagram of a graph-based regional electricity load forecasting method in one embodiment. (Refer to...) Figure 1 This graph-based regional electricity load forecasting method is applied to a graph-based regional electricity load forecasting system. The graph-based regional electricity load forecasting system includes a terminal 110 and a server 120. The terminal 110 and server 120 are connected via a network. The terminal 110 can be a desktop terminal or a mobile terminal, and the mobile terminal can be at least one of a mobile phone, tablet computer, or laptop computer. The server 120 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0024] Server 120 is configured to execute the above-mentioned graph-based regional electricity load prediction method, including: constructing a multivariate feature dataset; constructing a target graph structure with each reference region as a node and based on the corresponding geographical information of each reference region; training a preset basic prediction model based on the multivariate feature dataset and the target graph structure to obtain an electricity load prediction model; and predicting the electricity load of the target region based on the electricity load prediction request and the electricity load prediction model when a request for electricity load prediction for the target region is received.

[0025] Please see Figure 2 , Figure 2 This is a flowchart illustrating a graph-based regional power load forecasting method according to an embodiment of this application. This embodiment primarily uses the application of this graph-based regional power load forecasting method to a server as an example. Specifically, the graph-based regional power load forecasting method provided in this embodiment may include the following steps: S101. Construct a multivariate feature dataset.

[0026] Among them, the multidimensional feature dataset refers to a structured time-series dataset formed by integrating multidimensional feature data related to electricity load for different types of areas (such as rural transformer substations or highway service areas) after preprocessing.

[0027] For example, based on the predicted scenario, feature types directly related to load changes can be filtered, including general basic features and scenario-specific features. General basic features include historical load features and meteorological features, while scenario-specific features include distribution network operation features and agricultural production features in rural transformer substation scenarios, and traffic flow features and charging pile operation features in highway service area scenarios.

[0028] Specifically, historical load data can be obtained from the power grid marketing system and the intelligent monitoring terminal of the distribution area; meteorological data can be obtained from the national meteorological platform and the API interface of local meteorological stations; traffic flow and charging pile data can be obtained from the traffic management department system and the service area charging pile operation platform in the highway scenario; and distribution network operation data and agricultural data can be obtained from the power grid operation and maintenance system and the statistical reports of township agricultural departments. After the data is collected, the raw data is cleaned and standardized, such as by using the 3σ principle or the IQR quartile method to filter out extreme values ​​in the load, traffic flow and other data. Then, the preprocessed multi-class features are integrated according to the dimension of sample and feature combination to form structured data that fits the basic prediction model, that is, to form a multi-feature dataset.

[0029] Optionally, in some embodiments of this application, the step "constructing a multivariate feature dataset" may specifically include: Obtain regional data from multiple reference areas; Divide multiple reference regions into regions based on the region type of each reference region; Based on the segmentation results and regional data from multiple reference regions, a multivariate feature dataset corresponding to different region types is constructed.

[0030] The reference area refers to a sample area with typical electricity load characteristics. Its electricity data and related characteristics can be used to build a general model framework to provide data support for load forecasting of the target area. The area type is a category divided according to the core function of the area and the driving factors of electricity load. The characteristic system and periodic patterns of different areas are significantly different (e.g., the load fluctuation logic of rural transformer substations and highway service areas are different).

[0031] For example, select 10-20 representative reference areas, including 5-8 rural power distribution areas and 7-12 highway service areas, whose historical electricity consumption data are continuous and complete, and cover different scales to ensure data diversity.

[0032] The system obtains real-time electricity load, cumulative load, and transformer capacity from the regional power dispatching system, intelligent monitoring terminals in the transformer substations, and charging pile operation platforms; it obtains distribution network line load rate, distributed photovoltaic access volume, and agricultural busy / slack seasons from the power grid operation and maintenance system and township agricultural departments; it obtains traffic flow, proportion of new energy vehicles, charging pile utilization rate, and emergency lane linkage information from the traffic management system and charging pile platform, and additionally collects road network status data around the service areas; it obtains meteorological data such as temperature, precipitation, and wind force from meteorological department APIs and regional sensor networks; and it automatically generates timestamps and peak / valley time period division information through system time.

[0033] Based on timestamps, all features of the same type of region are aligned with a preset step size. The preprocessed data of the same type of region are merged in chronological order to form two types of multi-feature datasets: rural service areas and highway service areas.

[0034] S102. Using each reference area as a node, construct the target map structure based on the corresponding geographic information of each reference area.

[0035] Each node refers to a selected reference region and is the basic unit of the target graph structure. Each node has a unique identifier and is associated with its own attributes, used to distinguish different reference regions in the graph structure. Geographic information includes not only the basic geographic location data of the reference regions but also geographic association information directly related to the spatial distribution of electricity load. The target graph structure is a combination of a node set and a normalized adjacency matrix, used to quantify the spatial association strength between reference regions. The node set records the attribute information of each reference region, and the normalized adjacency matrix uses numerical values ​​(range [0,1]) to represent the degree of association between nodes.

[0036] Specifically, each reference region is mapped to a node, and the basic information of the nodes is organized to form a node set. Then, based on the type of the reference region, core geographic information for quantifying spatial association is collected, and the initial association strength is calculated based on the geographic information using the nodes as row / column indices to form the original adjacency matrix. Next, the element values ​​of the original matrix are corrected by combining geographic association information (such as the actual capacity of cable interconnection and the traffic flow carrying capacity of emergency lanes). Finally, the node set and the normalized adjacency matrix are combined to form the target graph structure.

[0037] Optionally, in some embodiments of this application, the step "using each reference region as a node and constructing a target map structure based on the geographical information corresponding to each reference region" may specifically include: Based on the preset electricity load forecasting strategy, multiple reference areas are determined in the regional database; Based on the geographical information corresponding to each reference area, an original adjacency matrix is ​​constructed; The original adjacency matrix is ​​adjusted based on the region information corresponding to each reference region to obtain the target graph structure.

[0038] Electricity load forecasting strategies refer to the rules for selecting reference areas to achieve accurate forecasts. The core of these strategies is to clearly define the target forecast scenario and the typicality criteria of the reference areas. The region database stores basic information on all potential regions, including unique identifiers (such as IDs), core attributes, and basic data. Reference areas are typical sample areas selected from the region database according to the electricity load forecasting strategy. These areas must possess data integrity and scenario representativeness; their data serves as the training basis for building the model and provides a reference for forecasting the target area.

[0039] The original adjacency matrix is ​​an n×n matrix with reference areas as nodes, which is based on geographic information to initially quantify the spatial association strength between nodes. The element values ​​range from [0,1], and the larger the value, the stronger the association.

[0040] Specifically, the prediction object type is determined based on the strategy. Then, a list of reference regions is determined in the region library, and the geographical information of the reference regions is obtained from the region library to construct an n×n matrix. Next, the region information used to correct the correlation strength is obtained from the region library to correct the element values ​​of the original adjacency matrix. Finally, the attribute set of the reference regions and the normalized adjacency matrix are combined to form the target graph structure.

[0041] Optionally, in some embodiments of this application, the step "constructing an original adjacency matrix based on the geographic information corresponding to each reference area" may specifically include: Construct a basic matrix using each reference region as a node; Based on the geographic information corresponding to each reference area, calculate the geographic distance corresponding to each reference area; The base matrix is ​​adjusted based on the geographical distance to obtain the original adjacency matrix.

[0042] The foundation matrix is ​​an n×n matrix (where n is the number of reference regions) used to initially quantify the existence of basic associations between nodes, with each reference region as a node. It does not involve fine-grained quantification of association strength. Geographic distance refers to the straight-line distance calculated based on the latitude and longitude coordinates of the reference regions and is the core indicator for quantifying the spatial physical proximity between nodes.

[0043] For example, specifically, the selected n reference areas are sorted by area identifier, corresponding to the row and column indices (1 to n) of the matrix. Furthermore, the association criteria can be based on whether they belong to the same distribution network line; or whether they belong to the same high-speed main line.

[0044] For rural distribution area node pairs: if they belong to the same distribution network line, the basic matrix element is 1; otherwise, it is 0. For highway service area node pairs: if they belong to the same highway main line, the basic matrix element is 1; otherwise, it is 0.

[0045] Then, the latitude and longitude coordinates of each reference region are obtained from the region database, and the straight-line distance between any two reference regions can be calculated using any distance calculation method. The calculated distances are organized into an n×n distance matrix D. Based on the association existence framework of the base matrix, the association strength is quantified by combining geographical distance. The base matrix is ​​then adjusted according to different scenarios to obtain the original adjacency matrix.

[0046] Optionally, in some embodiments of this application, the step "adjusting the original adjacency matrix according to the region information corresponding to each reference region to obtain the target graph structure" may specifically include: The element values ​​of the original adjacency matrix are adjusted according to the region information corresponding to each reference region; The region information corresponding to the reference region is added to the adjusted original adjacency matrix, and the original adjacency matrix after adding the information is normalized to obtain the normalized adjacency matrix. The attribute information of the reference region is fused with the normalized adjacency matrix to obtain the target graph structure.

[0047] Element values ​​refer to the numerical values ​​at each position in the original adjacency matrix, used to quantify the spatial association strength between two reference regions (such as node i and node j). Values ​​range from 0 to 1, with larger values ​​indicating a stronger association and a value of 0 indicating no association. It should be noted that the initial values ​​are calculated based on geographical distance and need further correction based on regional information. Normalization is a mathematical preprocessing of the adjusted adjacency matrix, aiming to eliminate feature bias caused by differences in the number of associated nodes (e.g., nodes with more associated nodes may be overweighted), making matrix elements comparable on a uniform scale.

[0048] Specifically, for rural transformer substation scenarios, the differences in transformer capacity, the amount of distributed photovoltaic (PV) grid connection, and the status of monitoring equipment are extracted. For highway service area scenarios, the average daily traffic flow of emergency lanes (actual linkage capability, charging pile power level, and service area operation status) are extracted. Then, the element values ​​are adjusted based on different scenarios. For example, in rural transformer substations, if the transformer capacity ratio between node i and node j is ≥2, the element value between node i and node j is reduced; if the monitoring equipment of node i is faulty, the element values ​​of all its associated elements are set to 0 (excluding the influence of invalid nodes); if both node i and node j are connected to distributed PV and have similar capacities, the element value between them is increased. In highway service areas, if the average daily traffic flow of emergency lanes between node i and node j is <100 vehicles, the element value between node i and node j is reduced.

[0049] Based on this, the adjusted adjacency matrix is ​​output. Then, the region information corresponding to the reference region is added to the adjusted original adjacency matrix, and the original adjacency matrix after adding the information is normalized to obtain the normalized adjacency matrix. It should be noted that adding region information does not mean embedding text or structured information into the matrix, but rather performing matrix preprocessing based on region information, and then eliminating scale bias through normalization: preserving the node's own features (i.e., adding self-loops).

[0050] Furthermore, the attribute information of the reference region is the basic feature used to identify the nodes. After fusion, a complete structure is formed. The attribute information includes the identifier, core attributes, type, and basic functional description of the reference region. Specifically, the node attribute set is bound to the normalized adjacency matrix. Thus, the elements in the matrix can be traced back to the association representation of the specific reference region through the node attribute set, and the final target graph structure is obtained.

[0051] Through the above steps, the target graph structure not only corrects the accuracy of association strength through regional information, but also ensures the stability of model processing through normalization. Finally, it combines node attributes to form a complete spatial association model, providing reliable spatial feature input for subsequent training and prediction models by fusing multiple features.

[0052] S103. Based on the multivariate feature dataset and target graph structure, the preset basic prediction model is trained to obtain the electricity load prediction model.

[0053] The basic prediction model is a deep learning model that can integrate spatial correlation features and temporal fluctuation features. Its core architecture employs a fusion of GCN (Graph Convolutional Network) and BiLSTM (Bidirectional Long Short-Term Memory Network). Specifically: the GCN layer processes the target graph structure and extracts spatial correlation features between reference regions; the BiLSTM layer processes the multivariate feature dataset and captures the temporal patterns of load; the fusion layer weighted and integrates the spatial and temporal features, predicting electricity load through the output layer.

[0054] For example, the model structure is configured according to the prediction scenario, where the layer parameters are matched with the feature dimensions, as specifically configured as follows: The input to the GCN layer is the normalized adjacency matrix of the target graph structure and the static node features from the multivariate features. The output dimension of each hidden layer is 64 / 32 (stepwise feature compression), and the activation function is the ReLU function. Its output is the spatial aggregated feature of each reference region. The input to the BiLSTM layer is the temporal dynamic features from the multivariate feature dataset. Its hidden layer is a two-layer bidirectional LSTM with 128 hidden units per layer, returning sequence features; the output is the temporal dependency feature of each reference region. The fusion layer is used to weightedly fuse the spatial features output by GCN and the temporal features output by BiLSTM through an attention mechanism; the output layer is a fully connected layer with a linear activation function, used to output the predicted electricity load for future periods. In actual training, the temporal features of the training set and the target graph structure can be input into the basic prediction model, and the weight parameters of GCN, BiLSTM, and the fusion layer can be updated through gradient descent.

[0055] After each training round, the forward propagation process is repeated using the validation set, and the validation set loss is calculated. If the validation set loss increases for five consecutive rounds (overfitting signal), the current optimal parameters are saved. This yields the electricity load prediction model.

[0056] Optionally, in some embodiments of this application, the step "training a preset basic prediction model based on the multivariate feature dataset and target graph structure to obtain an electricity load prediction model" includes: The multivariate feature set is divided to obtain a training set, a validation set, and a test set. The preset basic prediction model is trained based on the training set and the target graph structure, and the trained prediction model is validated using the validation set. When the trained prediction model meets the preset conditions, the trained prediction model is tested based on the test set; When the trained prediction model meets the testing criteria, the prediction model that passes the test is determined as the electricity load prediction model.

[0057] S104. When a power load prediction request for the target area is received, the power load of the target area is predicted based on the power load prediction request and the power load prediction model.

[0058] Among them, the electricity load forecasting request is a load forecasting demand instruction initiated by the user or system for a specific target area. It includes: target area identifier, forecast time range, and additional scenario information. The additional scenario information can be the impact of the New Year's Day holiday or sudden rainfall, which is used to optimize the scenario adaptability of the forecast.

[0059] Extract and preprocess the time-series features required for prediction. For example, for rural transformer substation scenarios, historical load, temperature, precipitation, busy farming season indicators, and distribution network load rate of the target area can be extracted; for highway service area scenarios, historical charging load, traffic flow, proportion of new energy vehicles, meteorological data, and holiday indicators of the target area can be extracted.

[0060] Then, the extracted temporal features are normalized to the [0,1] interval. Next, the electricity load prediction model is called, and the preprocessed features and the target graph structure are input into the model. The GCN layer outputs spatial aggregation features; the BiLSTM layer outputs temporal dependency features; the fusion layer integrates the two types of features through an attention mechanism and outputs the normalized predicted load. Finally, the predicted load is output as structured text.

[0061] Optionally, in some embodiments of this application, the step "when a power load forecasting request for the target area is received, predicting the power load of the target area based on the power load forecasting request and the power load forecasting model" may specifically include: When a power load forecasting request for the target area is received, the forecasting time range and the target area identifier are obtained from the power load forecasting request. Based on the target area identifier, the target electricity load prediction model corresponding to the target area is determined in the electricity load prediction model corresponding to each area; Based on the predicted time range, real-time characteristic data of the target area, and the target electricity load prediction model, the electricity load of the target area is predicted.

[0062] To further understand the graph-based regional electricity load forecasting scheme of this application, the following descriptions will further illustrate the two scenarios of rural areas and highway service areas: For rural scenarios, such as Figure 3As shown, data is collected and organized based on an existing rural electricity load database. The model input data includes historical load data. Meteorological data (such as temperature) ,humidity Wind speed ) and charging characteristics (such as charging station utilization rate) Number of electric vehicles connected (etc.). To eliminate the influence of dimensions, all input features are normalized or standardized. The raw load data and meteorological features are aligned along the time dimension to ensure the consistency of the input sequence.

[0063] Since electricity load is influenced not only by its own historical and environmental factors but also by the status of spatially adjacent nodes (such as surrounding transformer substations or service areas), a Graph Convolutional Network (GCN) is used to extract spatial correlations. A node set V is constructed to represent each monitoring point or region, and an adjacency matrix A is used to represent geographical or power grid physical connections. The core operation of GCN is: in, It is the adjacency matrix after adding self-loops. for The degree matrix, These are the node features of the l-th layer. It is a learnable weight matrix. The activation function is ReLU. After passing through several GCN layers, a vector representation of the fused spatial features of each node is obtained.

[0064] To capture data dependencies between consecutive time steps within a time series, a Bidirectional Long Short-Term Memory (BiLSTM) network is introduced. BiLSTM performs forward and backward encoding on each time series input separately, outputting the hidden state. ,Right now: in The input features are processed by the GCN layer at time t. and These are the hidden states for the forward and backward directions, respectively. The final output concatenates the forward and backward information to more accurately capture long-term dependencies.

[0065] The GCN-BiLSTM joint model structure was built, with hyperparameters including the number of hidden layers L and the number of neurons. Learning rate Batch size B, etc. Mean Squared Error (MSE) is used as the loss function, defined as follows: in, This is the actual load. Here, n represents the model's predicted value, and n is the number of samples. The optimization algorithm is Adam, which automatically adjusts the learning rate and accelerates convergence. During training, the validation set loss is monitored, and an Early Stopping mechanism is used. If the validation loss does not improve after several rounds, training is stopped early to prevent overfitting.

[0066] 5) Model evaluation and parameter adjustment After training, the model performance is evaluated using a test set. Mainstream evaluation metrics include: If the evaluation indicators do not meet expectations, the model can be iteratively optimized by adjusting the neural network structure (such as the number of layers and nodes), optimizing hyperparameters, and expanding the training set until it meets the actual load forecasting application requirements.

[0067] For load forecasting of highway service areas, this paper analyzes and extracts the load characteristics of highway service areas with large-scale charging access based on Task 1. The data source is a database of electricity load characteristics of highway service areas obtained through multi-source information fusion from Task 1. This allows for better capture of the preceding and following information in the time series data.

[0068] Based on the rural electricity load database constructed in Task 1, a GCN-BiLSTM model is built. The inputs are historical load, meteorological, and charging characteristic data, and the output is the predicted future load value. Hyperparameters such as the number of hidden layers, the number of neurons, the learning rate, and the batch size are set.

[0069] The dataset was divided into training, validation, and test sets for model training, parameter tuning, and performance evaluation, respectively. During training, mean squared error (MSE) was used as the loss function, the Adam optimization algorithm was employed, and early stopping was combined to prevent overfitting.

[0070] Evaluate the model using a test set, with metrics including MAPE, RMSE, and MASE. If the prediction accuracy is insufficient, adjust the model structure, expand the training data, or optimize the parameters until the requirements are met.

[0071] The optimized model is used for short-term and long-term load forecasting of service areas along highways. Inputting current load, traffic flow, weather, and equipment status data, the model predicts future long-term loads recursively. Finally, the discrepancy between actual and predicted loads is continuously monitored, and the model is dynamically adjusted to improve accuracy and adaptability. The proposed model is validated using publicly available data.

[0072] The specific implementation is as follows: First, leveraging the electricity load characteristic database of highway service areas extracted and constructed in Task 1, multi-source heterogeneous data is aggregated. Input features include historical load data. Meteorological data (such as temperature) Precipitation ,humidity Charging behavior characteristics (real-time power of charging pile) Terminal access volume Traffic flow and status information of key equipment. To standardize the input format and improve training efficiency, data of different dimensions are standardized to ensure consistent feature scale and facilitate neural network learning.

[0073] To address the regional spatial correlation and temporal dependence characteristics of highway service areas, a GCN-BiLSTM model is employed to fuse these two types of information. A graph structure V is constructed with service areas as nodes, and an adjacency matrix A quantifies the spatial or traffic network associations between nodes. The GCN component extracts spatial association features using the following formula: in, , Let the node degree matrix be... For parameter matrices, This is the activation function. After obtaining the spatial embedding, it is input into a bidirectional long short-term memory (BiLSTM) network. BiLSTM extracts historical and future load evolution information in both forward and reverse directions along the time dimension, and its output at each time step is: in, and These are the hidden states of the forward and reverse LSTMs, respectively, which are then concatenated into a comprehensive feature for prediction.

[0074] The entire process divides the dataset into training, validation, and test sets. Model parameters (such as the number of GCN layers L and the number of neurons per layer) are configured in the training set. Learning rate Initialization and continuous optimization (e.g., batch size B). The loss function uses mean squared error (MSE). in, This is the actual load. Here, is the model's predicted value, and n is the total number of samples. Parameter optimization uses the Adam algorithm to dynamically adjust the learning rate. During training, early stopping is used to monitor the validation set loss. If the validation loss does not decrease after several rounds, training is terminated early to suppress overfitting.

[0075] After the model is trained, its real-world prediction performance is evaluated using a test set. Common metrics include: MAPE (Mean Absolute Percentage Error): RMSE (Root Mean Square Error): MASE (Mean Absolute Normalized Error): If the model performs poorly, further fine-tune the network structure (e.g., increase the number of GCN or LSTM layers), adjust hyperparameters, and expand the training samples until all metrics meet the actual requirements. In summary, the graph-based regional electricity load prediction method provided in this embodiment, after constructing a multivariate feature dataset, uses each reference region as a node and constructs a target graph structure based on the corresponding geographical information of each reference region. Then, a preset basic prediction model is trained based on the multivariate feature dataset and the target graph structure to obtain an electricity load prediction model. When a request for electricity load prediction for a target region is received, the electricity load of the target region is predicted based on the electricity load prediction request and the electricity load prediction model. In the graph-based regional electricity load prediction scheme provided in this application, constructing a target graph structure using reference regions as nodes and combining geographical information can capture the spatial dependence and coupling characteristics of multiple nodes in a region, making up for the shortcomings of traditional methods in lacking spatial correlation. The model training by integrating both methods takes into account spatiotemporal characteristics, improving the prediction accuracy and robustness in scenarios such as rural areas and highway service areas. Finally, the method responds to prediction requests and adapts to the target region, providing support for power supply stability, charging pile optimization, and energy dispatch.

[0076] It should be understood that, although Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0077] To facilitate better implementation of the graph-based regional electricity load forecasting method of this application, this application also provides a graph-based regional electricity load forecasting device based on the above-described graph-based regional electricity load forecasting method. The meanings of the terms used are the same as in the graph-based regional electricity load forecasting method described above, and specific implementation details can be found in the descriptions in the method embodiments.

[0078] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a graph-based regional electricity load prediction device provided in an embodiment of this application. Specifically, the graph-based regional electricity load prediction device may include a first construction module 201, a second construction module 202, a training module 203, and a prediction module 204, as follows: The first construction module 201 is used to construct a multivariate feature dataset; The second construction module 202 is used to construct the target map structure based on the geographic information corresponding to each reference area, with each reference area as a node. Training module 203 is used to train a preset basic prediction model based on the multivariate feature dataset and target graph structure to obtain an electricity load prediction model; The prediction module 204 is used to predict the electricity load of the target area based on the electricity load prediction request and the electricity load prediction model when a power load prediction request for the target area is received.

[0079] Specific limitations regarding the graph-based regional electricity load forecasting device can be found in the limitations of the graph-based regional electricity load forecasting method described above, and will not be repeated here. Each module in the aforementioned graph-based regional electricity load forecasting device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0080] The graph-based regional electricity load prediction device provided in this embodiment includes a first construction module 201 that constructs a multivariate feature dataset, and a second construction module 202 that constructs a target graph structure using each reference region as a node and based on the corresponding geographical information of each reference region. Then, a training module 203 trains a preset basic prediction model based on the multivariate feature dataset and the target graph structure to obtain an electricity load prediction model. When a prediction module 204 receives an electricity load prediction request for a target region, it predicts the electricity load of the target region based on the electricity load prediction request and the electricity load prediction model. In the graph-based regional electricity load prediction scheme provided in this application, constructing a target graph structure using reference regions as nodes and combining geographical information can capture the spatial dependence and coupling characteristics of multiple nodes in a region, making up for the shortcomings of traditional methods in terms of missing spatial correlation. The fusion of both methods in training the model takes into account spatiotemporal characteristics, improving the prediction accuracy and robustness in scenarios such as rural areas and highway service areas. Finally, the device responds to prediction requests and adapts to the target region, providing support for power supply stability, charging pile optimization, and energy dispatch.

[0081] Furthermore, embodiments of this application also provide an electronic device, such as... Figure 5 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically: The electronic device may include components such as a processor 301 with one or more processing cores, a memory 302 with one or more computer-readable storage media, a power supply 303, and an input unit 304. Those skilled in the art will understand that... Figure 5 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 301 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in the memory 302, and by calling data stored in the memory 302, thereby providing overall monitoring of the electronic device. Optionally, the processor 301 may include one or more processing cores; preferably, the processor 301 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 301.

[0082] The memory 302 can be used to store software programs and modules. The processor 301 executes various functional applications and graph-based regional power load forecasting methods by running the software programs and modules stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device, etc. In addition, the memory 302 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 302 may also include a memory controller to provide the processor 301 with access to the memory 302.

[0083] The electronic device also includes a power supply 303 that supplies power to various components. Preferably, the power supply 303 can be logically connected to the processor 301 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 303 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0084] The electronic device may also include an input unit 304, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0085] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 301 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 302 according to the following instructions, and the processor 301 runs the applications stored in the memory 302 to realize various functions, as follows: Construct a multivariate feature dataset; using each reference region as a node, construct a target map structure based on the corresponding geographical information of each reference region; train a preset basic prediction model based on the multivariate feature dataset and the target map structure to obtain an electricity load prediction model; when a request for electricity load prediction for a target region is received, predict the electricity load of the target region based on the electricity load prediction request and the electricity load prediction model.

[0086] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0087] In this embodiment, after constructing a multivariate feature dataset, a target graph structure is built using each reference region as a node and based on the corresponding geographical information of each reference region. Then, a preset basic prediction model is trained based on the multivariate feature dataset and the target graph structure to obtain an electricity load prediction model. When an electricity load prediction request for a target region is received, the electricity load of the target region is predicted based on the electricity load prediction request and the electricity load prediction model. In the graph-based regional electricity load prediction scheme provided in this application, using reference regions as nodes and combining geographical information to construct a target graph structure can capture the spatial dependence and coupling characteristics of multiple nodes in a region, making up for the shortcomings of traditional methods in lacking spatial correlation. The model training combines both methods, taking into account spatiotemporal characteristics, and improving the prediction accuracy and robustness in scenarios such as rural areas and highway service areas. Finally, the prediction request is adapted to the target region, providing support for power supply stability, charging pile optimization, and energy dispatch.

[0088] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0089] To this end, embodiments of this application provide a storage medium storing a plurality of instructions that can be loaded by a processor to execute steps in any of the graph-based regional electricity load forecasting methods provided in embodiments of this application. For example, the instructions can execute the following steps: Construct a multivariate feature dataset; using each reference region as a node, construct a target map structure based on the corresponding geographical information of each reference region; train a preset basic prediction model based on the multivariate feature dataset and the target map structure to obtain an electricity load prediction model; when a request for electricity load prediction for a target region is received, predict the electricity load of the target region based on the electricity load prediction request and the electricity load prediction model.

[0090] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0091] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0092] Since the instructions stored in the storage medium can execute the steps in any of the graph-based regional power load forecasting methods provided in the embodiments of this application, the beneficial effects that any of the graph-based regional power load forecasting methods provided in the embodiments of this application can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0093] The above provides a detailed description of a graph-based regional power load forecasting method and related equipment provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting regional electricity load based on graph structure, characterized in that, include: Construct a multi-feature dataset; Using each reference region as a node, a target map structure is constructed based on the corresponding geographic information of each reference region; The preset basic prediction model is trained based on the multivariate feature dataset and target graph structure to obtain the electricity load prediction model; When a power load forecasting request for a target area is received, the power load of the target area is forecasted based on the power load forecasting request and the power load forecasting model.

2. The regional electricity load forecasting method based on graph structure according to claim 1, characterized in that, The construction of the target map structure, using each reference region as a node and based on the corresponding geographical information of each reference region, includes: Based on the preset electricity load forecasting strategy, multiple reference areas are determined in the regional database; Based on the geographical information corresponding to each reference area, an original adjacency matrix is ​​constructed; The original adjacency matrix is ​​adjusted based on the region information corresponding to each reference region to obtain the target graph structure.

3. The regional electricity load forecasting method based on graph structure according to claim 2, characterized in that, The construction of the original adjacency matrix based on the geographical information corresponding to each reference region includes: Construct a basic matrix using each reference region as a node; Based on the geographic information corresponding to each reference area, calculate the geographic distance corresponding to each reference area; The base matrix is ​​adjusted based on the geographical distance to obtain the original adjacency matrix.

4. The regional electricity load forecasting method based on graph structure according to claim 2, characterized in that, The step of adjusting the original adjacency matrix based on the region information corresponding to each reference region to obtain the target graph structure includes: The element values ​​of the original adjacency matrix are adjusted according to the region information corresponding to each reference region; The region information corresponding to the reference region is added to the adjusted original adjacency matrix, and the original adjacency matrix after adding the information is normalized to obtain the normalized adjacency matrix. The attribute information of the reference region is fused with the normalized adjacency matrix to obtain the target graph structure.

5. The regional electricity load forecasting method based on graph structure according to claim 1, characterized in that, The step of training a preset basic prediction model based on the multivariate feature dataset and the target graph structure to obtain an electricity load prediction model may specifically include: The multivariate feature set is divided to obtain a training set, a validation set, and a test set. The preset basic prediction model is trained based on the training set and the target graph structure, and the trained prediction model is validated using the validation set. When the trained prediction model meets the preset conditions, the trained prediction model is tested based on the test set; When the trained prediction model meets the testing criteria, the prediction model that passes the test is determined as the electricity load prediction model.

6. The regional electricity load forecasting method based on graph structure according to claim 1, characterized in that, When a power load forecasting request for the target area is received, the step of forecasting the power load of the target area based on the power load forecasting request and the power load forecasting model includes: When a power load forecasting request for the target area is received, the forecasting time range and the target area identifier are obtained from the power load forecasting request. Based on the target area identifier, the target electricity load prediction model corresponding to the target area is determined in the electricity load prediction model corresponding to each area; Based on the predicted time range, real-time characteristic data of the target area, and the target electricity load prediction model, the electricity load of the target area is predicted.

7. The regional electricity load forecasting method based on graph structure according to claim 1, characterized in that, The construction of the multivariate feature dataset includes: Obtain regional data from multiple reference areas; Divide multiple reference regions into regions based on the region type of each reference region; Based on the segmentation results and regional data from multiple reference regions, a multivariate feature dataset corresponding to different region types is constructed.

8. A regional electricity load prediction device based on graph structure, characterized in that, include: The first building module is used to construct a multivariate feature dataset; The second construction module is used to construct the target map structure based on the geographic information corresponding to each reference area, with each reference area as a node. The training module is used to train the preset basic prediction model based on the multivariate feature dataset and the target graph structure to obtain the electricity load prediction model. The prediction module is used to predict the electricity load of the target area based on the electricity load prediction request and the electricity load prediction model when a power load prediction request for the target area is received.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the graph-based regional electricity load forecasting method as described in any one of claims 1-7.

10. A storage medium, characterized in that, The computer program is stored and can be loaded by a processor and executed as described in any one of claims 1-7, which is a graph-based regional electricity load forecasting method.

Images