Power transaction declaration data analysis method and device based on multi-time sequence diagram and graph convolutional encoder, equipment and medium

Abstract

This application discloses a method, apparatus, equipment, and medium for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders, relating to the field of data analysis technology. The method involves preprocessing segmented declaration data from multiple power trading entities to obtain standardized declaration data samples; constructing an Hermit-Laplacian matrix using a directed graph of the declaration data; performing eigenvalue decomposition on the Hermit-Laplacian matrix to obtain declaration spectrum feature vectors; constructing a multi-time-series power trading entity relationship graph; aggregating the declaration spectrum feature vectors of adjacent power trading entities using a graph convolutional encoder to obtain declaration graph embedding features; optimizing the graph convolutional encoder based on the declaration graph embedding features to determine the comprehensive declaration feature vectors of each power trading entity; and generating power transaction declaration data analysis results for each power trading entity based on the comprehensive declaration feature vectors, thereby improving the accuracy of power transaction declaration data feature analysis and entity pattern segmentation.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method, apparatus, equipment and medium for analyzing power transaction declaration data based on multi-time series graphs and graph convolution encoders. Background Technology

[0002] As the construction of the electricity market continues to advance, different types of market participants, including conventional generating units, new energy generating units, energy storage power stations, and electricity retailers, are gradually participating in real-time electricity trading. Market participants are typically required to submit segmented declaration data according to the prescribed trading periods. This segmented declaration data consists of multiple declared electricity volumes and prices, and is presented as a stepped curve reflecting the declaration characteristics of market participants within different electricity volume ranges. In the actual operation of electricity trading, different market participants differ in resource types, operating costs, regulation capabilities, forecasting accuracy, and trading strategies, resulting in significant heterogeneity in their declaration data.

[0003] Current power trading declaration data analysis typically employs manual statistical indicators, curve distance calculations, traditional clustering, or single-time-section analysis methods. Manual statistical indicators primarily describe the main characteristics using a limited set of metrics such as average bid price, highest bid price, lowest bid price, and bid volume, making it difficult to fully preserve the local morphology and step-by-step changes within segmented bid curves. Traditional curve distance methods usually convert bid data directly into fixed-length vectors, easily overlooking the connection order between bid segments and the joint changes in volume and price. Ordinary clustering methods mainly categorize main entities based on data from a single trading day or a single trading period, failing to reflect the continuity of the main entity's bid characteristics over time.

[0004] As can be seen from the above, how to solve the problems of difficulty in uniformly representing the segmented declaration curve structure, difficulty in depicting the dynamic relationship between different entities, and difficulty in extracting the time-series evolution characteristics of declaration data, and improve the accuracy of power transaction declaration data feature analysis and entity pattern classification are problems to be solved in this field. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders. This method can solve the problems of difficulty in uniformly representing the structure of segmented declaration curves, difficulty in depicting the dynamic relationships between different entities, and difficulty in extracting the temporal evolution characteristics of declaration data, thereby improving the accuracy of power transaction declaration data feature analysis and entity pattern classification. The specific solution is as follows: In a first aspect, this application discloses a method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders, including: The segmented declaration data from multiple power trading entities is preprocessed to obtain standardized declaration data samples; the segmented declaration data refers to data submitted over multiple consecutive trading days and multiple trading periods. Using each declaration segment in the standardized declaration data sample as a node, a directed graph of declaration data is constructed. The Hermit-Laplace matrix is ​​constructed using the directed graph of declaration data. The Hermit-Laplace matrix is ​​then subjected to eigenvalue decomposition to obtain the declaration spectrum feature vector of each power trading entity in each trading period. Using the aforementioned declaration spectrum feature vectors, a multi-time series power trading entity relationship diagram at different time scales is constructed; Based on each of the power trading entities, the declaration spectrum feature vectors, and the multi-time series power trading entity relationship graph, graph convolutional encoders at different time scales are constructed. The graph convolutional encoders are used to aggregate the declaration spectrum feature vectors of adjacent power trading entities to obtain declaration graph embedding features at different time scales. The graph convolutional encoder is optimized using the embedded features of the declared graph. The optimized graph convolutional encoder is used to determine the comprehensive declaration feature vector of each power trading entity. Based on the comprehensive declaration feature vector, the power trading declaration data analysis results of each power trading entity are generated.

[0006] Optionally, the preprocessing of segmented declaration data from multiple power trading entities to obtain standardized declaration data samples includes: The system acquires segmented declaration data submitted by multiple power trading entities over multiple consecutive trading days and trading periods; the segmented declaration data includes the declared electricity volume and declared price corresponding to each declaration segment. The segmented declaration data is processed for missing values ​​and outliers are corrected to obtain corrected segmented declaration data; According to the preset number of application segments, the corrected segmented application data is processed to unify the application segment structure, resulting in segmented application data with a unified application segment structure; the unified application segment structure processing includes interpolation, merging or completion processing; Normalize the segmented application data with a unified application segment structure to obtain standardized application data samples.

[0007] Optionally, a directed graph of declaration data is constructed using each declaration segment in the standardized declaration data sample as a node. A Hermit-Laplace matrix is ​​then constructed using the directed graph. Eigenvalue decomposition is performed on the Hermit-Laplace matrix to obtain the declaration spectrum feature vectors of each power trading entity in each trading period, including: Each segment in the standardized declaration data sample is used as a declaration data node; Based on the arrangement order between different declaration segments, the differences between declared electricity volumes, and the differences between declared prices, directed connections are constructed between the different declaration data nodes to obtain a directed graph of declaration data; Construct a Hermitian connectivity matrix and a degree matrix using the directed graph derived from the declared data; Based on the Hermitian connectivity matrix and the degree matrix, construct the Hermitian-Laplace matrix; The Hermitian-Laplace matrix is ​​subjected to eigenvalue decomposition to obtain eigenvectors; The feature vectors are normalized and clustered to obtain the declaration spectrum feature vectors of each power trading entity in each trading period.

[0008] Optionally, the Hermitian connectivity matrix is: ; in, i This refers to the serial number of the electricity trading entity. u , v Indicates the data node to be declared. Hermitian connection matrix data nodes in the middle and reporting u and reporting data nodes v The corresponding matrix elements, for conjugate, For reporting data nodes u With the data node of the declaration v Similarity of declaration features between them It is the root of unity of a complex number; The degree matrix is: ; in, For degree matrix, j Indicates the data node to be declared. For reporting data nodes u With the data node of the declaration j Similarity of declaration features between them; The Hermitian-Laplace matrix is: ; in, For Hermitian-Laplace matrix, It is an identity matrix.

[0009] Optionally, using the declared spectrum feature vectors, a multi-time series power trading entity relationship diagram at different time scales is constructed, including: Set multiple time series windows with different time spans; Calculate the Euclidean distance between the eigenvectors of the declaration spectrum of different power trading entities within each of the aforementioned time windows; Based on the Euclidean distance, the similarity of the declared features is determined; By utilizing the similarity of the declared features, the connection relationships and connection weights between different power trading entities within each of the time windows are determined; Construct a multi-time series power trading entity relationship graph that includes the connection relationships and connection weights at different time scales.

[0010] Optionally, based on each of the power trading entities, the declaration spectrum feature vectors, and the multi-time series power trading entity relationship graph, a graph convolutional encoder at different time scales is constructed. The graph convolutional encoder is then used to aggregate the declaration spectrum feature vectors of adjacent power trading entities to obtain declaration graph embedding features at different time scales, including: Each power trading entity is used as a graph node, the declaration spectrum feature vector is used as the node input feature, and the connection relationship and connection weight in the multi-time series power trading entity relationship graph are used as graph adjacency information to construct a graph convolutional encoder at different time scales. The attention coefficient of adjacent electricity trading entities is determined using an attention mechanism; Using the attention coefficients and the graph convolutional encoder, the declaration spectrum feature vectors of adjacent power trading entities are weighted and aggregated to obtain declaration graph embedding features at different time scales.

[0011] Optionally, the graph convolutional encoder is optimized using the declared graph embedding features. The optimized graph convolutional encoder is then used to determine the comprehensive declared feature vector for each power trading entity. Based on the comprehensive declared feature vector, power trading declaration data analysis results for each power trading entity are generated, including: A joint training objective is constructed using a multi-time series power trading entity relationship graph, declaration spectrum feature vector, and declaration graph embedding features. Based on the joint training objective, the graph convolutional encoder, time-scale fusion parameters, and class center parameters are optimized. Using the optimized graph convolutional encoder and time-scale fusion parameters, the comprehensive declaration feature vector of each power trading entity is determined. Based on the comprehensive declaration feature vector, the feature similarity between different power trading entities, the category membership degree of each power trading entity, and the degree of feature change between adjacent trading cycles are determined. The optimized category center parameters are used to determine the declaration characteristic categories of each power trading entity; Based on the declared feature categories, feature similarity, category membership, and feature variation, the power trading declaration data analysis results for each power trading entity are generated.

[0012] Secondly, this application discloses a power transaction declaration data analysis device based on a multi-time-series graph and a graph convolutional encoder, comprising: The preprocessing module is used to preprocess the segmented declaration data of multiple power trading entities to obtain standardized declaration data samples; the segmented declaration data refers to the data submitted within multiple consecutive trading days and multiple trading periods. The feature decomposition module is used to construct a directed graph of declaration data with each declaration segment in the standardized declaration data sample as a node, construct a Hermit-Laplace matrix using the directed graph of declaration data, and perform feature decomposition on the Hermit-Laplace matrix to obtain the declaration spectrum feature vector of each power trading entity in each trading period. The relationship graph construction module is used to construct a multi-time series power trading entity relationship graph at different time scales using the declared spectrum feature vector; The aggregation module is used to construct graph convolutional encoders at different time scales based on each of the power trading entities, the declaration spectrum feature vectors, and the multi-time series power trading entity relationship graph. The graph convolutional encoders are used to aggregate the declaration spectrum feature vectors of adjacent power trading entities to obtain declaration graph embedding features at different time scales. The analysis result generation module is used to optimize the graph convolutional encoder using the embedded features of the declared graph, determine the comprehensive declaration feature vector of each power trading entity using the optimized graph convolutional encoder, and generate the power trading declaration data analysis results of each power trading entity based on the comprehensive declaration feature vector.

[0013] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders.

[0014] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders.

[0015] Beneficial Effects: This application provides a method for analyzing power transaction declaration data based on multi-time series graphs and graph convolutional encoders. It includes preprocessing segmented declaration data from multiple power trading entities to obtain standardized declaration data samples. This improves the consistency of segmented declaration data representation across different trading days and periods for different power trading entities, reducing the impact of data structure and scale differences on subsequent analysis. Using each declaration segment in the standardized declaration data sample as a node, a directed graph of the declaration data is constructed. An Hermit-Laplace matrix is ​​then constructed using this directed graph. Feature decomposition of the Hermit-Laplace matrix yields the declaration spectrum feature vectors for each power trading entity in each trading period. This method preserves the sequential relationships and directional information between declaration segments while unifying the feature dimensions, reducing the loss of segmented declaration structure information by traditional statistical indicators. Finally, the declaration spectrum feature vectors are used to construct different time scales... A multi-time series power trading entity relationship graph enhances the ability to identify the reporting relationships between power trading entities at different time scales. Based on each power trading entity, the reporting spectrum feature vector, and the multi-time series power trading entity relationship graph, graph convolutional encoders at different time scales are constructed. The graph convolutional encoders are used to aggregate the reporting spectrum feature vectors of adjacent power trading entities to obtain reporting graph embedding features at different time scales, thus enabling the extraction of entity-specific reporting features and inter-entity relationship features. The graph convolutional encoders are optimized using the reporting graph embedding features. The optimized graph convolutional encoders determine the comprehensive reporting feature vectors of each power trading entity. Based on the comprehensive reporting feature vectors, the power trading reporting data analysis results of each power trading entity are generated, thus reflecting short-term reporting changes, medium-term behavioral evolution, and long-term stable features, improving the accuracy of power trading reporting data feature representation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This application discloses a flowchart of a power transaction declaration data analysis method based on multi-time series graphs and graph convolutional encoders. Figure 2 This is a schematic diagram illustrating the feature similarity between different power trading entities disclosed in this application; Figure 3 This is a schematic diagram illustrating the dimensionality reduction results between different power trading entities disclosed in this application; Figure 4This application discloses a specific flowchart for analyzing power transaction declaration data based on a multi-time-series graph and graph convolutional encoder. Figure 5 This is a schematic diagram of the structure of a power transaction declaration data analysis device based on a multi-time-series graph and graph convolutional encoder disclosed in this application; Figure 6 This is a schematic diagram of the structure of an electronic device disclosed in this application. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] With the continuous advancement of the power market construction, various types of market participants, including conventional generating units, new energy generating units, energy storage power stations, and electricity retailers, are gradually participating in real-time power trading. Market participants typically need to submit segmented declaration data according to the prescribed trading periods. These segmented declaration data consist of multiple declared electricity volumes and prices, and are presented as a stepped curve reflecting the declaration characteristics of market participants within different electricity volume ranges. In the actual operation of power trading, different market participants differ in resource types, operating costs, regulation capabilities, forecasting accuracy, and trading strategies, resulting in significant heterogeneity in their declaration data. Existing power trading declaration data analysis typically employs manual statistical indicators, curve distance calculations, traditional clustering, or single-time-section analysis methods. Manual statistical indicators mainly describe the characteristics of the participants through a few indicators such as average declared price, highest declared price, lowest declared price, and declared electricity volume, making it difficult to fully retain the local morphology and stepped change information in the segmented declaration curves. Traditional curve distance methods typically convert declaration data directly into fixed-length vectors, easily ignoring the connection order between declaration segments and the joint changes in electricity volume and price. Ordinary clustering methods mainly classify participants based on data from a single trading day or a single trading period, making it difficult to reflect the continuity of participant declaration characteristics over time. As can be seen from the above, how to solve the problems of difficulty in uniformly representing the segmented declaration curve structure, difficulty in depicting the dynamic relationship between different entities, and difficulty in extracting the time-series evolution characteristics of declaration data, and improve the accuracy of power transaction declaration data feature analysis and entity pattern classification are problems to be solved in this field.

[0020] See Figure 1 As shown, this embodiment of the invention discloses a method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders, which specifically includes: Step S11: Preprocess the segmented declaration data of multiple power trading entities to obtain standardized declaration data samples; the segmented declaration data refers to the data submitted within multiple consecutive trading days and multiple trading periods.

[0021] In this embodiment, segmented declaration data submitted by multiple power trading entities over multiple consecutive trading days and trading periods are acquired; the segmented declaration data is processed for missing values ​​and outliers to obtain corrected segmented declaration data; according to a preset number of declaration segments, the corrected segmented declaration data is processed to unify the declaration segment structure to obtain segmented declaration data with a unified declaration segment structure; the unified declaration segment structure processing includes interpolation, merging, or completion processing; the segmented declaration data with a unified declaration segment structure is normalized to obtain standardized declaration data samples; the segmented declaration data includes the declared electricity volume and declared price corresponding to each declaration segment.

[0022] Specifically, assuming the power generation entity The spot market The hourly ten-segment daytime price sequence is as follows: ; in, This is the daily price sequence. Main generator The spot market The price quoted in the first segment of the hourly declaration. Main generator The spot market The reported volume in the first reporting segment of the hour. Main generator The spot market The quote for the second segment of the hour. Main generator The spot market The reported volume in the second reporting segment of the hour. Main generator The spot market The price quoted in the 10th segment of the hour. Main generator The spot market The volume reported in the 10th reporting segment of the hour.

[0023] To address the issue of inconsistent numbers of declaration segments among different power trading entities, segmented declaration data is interpolated, merged, or supplemented according to a unified number of declaration segments. To address the inconsistency in the declared capacity and price scales among different entities, declared electricity and prices are normalized separately to obtain standardized declared electricity and standardized declared prices. Through these processes, segmented declaration data is transformed into standardized declaration data samples with consistent dimensions and unified scales, providing a unified input for subsequent structure coding and graph learning. The sample format is as follows: ; ; ; in, Main generator No. The continuous curve after the celestial expansion is standardized. A stepped curve formed by piecing together elements. and They are the main power generation entities exist The first hour Standardized pricing and quantity reports for each declaration segment. and They are the main power generation entities exist The first hour Standardized pricing and quantity reports for each declaration segment. for At any point in the expansion curve, for The number of extended repetitions, The integer is a relatively large integer; in this embodiment, it is 1000. This is the floor function.

[0024] Furthermore, when the number of bidding segments differs among different power trading entities, the hourly bidding prices and bidding volumes of each power trading entity are normalized with a target of a pre-set uniform number of 10 bidding segments: ; ; in, and They are the main power generation entities Daily minimum and maximum prices Main generator Maximum reported electricity consumption per hour.

[0025] By handling missing values ​​and correcting outliers, the segmented declaration data that enters the unified structure processing remains usable; by interpolation, merging, or completion processing, the number of declaration segments for different power trading entities is made consistent with the preset number of declaration segments; by normalizing the segmented declaration data with unified declaration segment structure, the declared electricity volume and declared price of different entities are on a unified scale, resulting in standardized declaration data samples.

[0026] Step S12: Construct a directed graph of declaration data using each declaration segment in the standardized declaration data sample as a node; construct a Hermit-Laplace matrix using the directed graph of declaration data, and perform eigenvalue decomposition on the Hermit-Laplace matrix to obtain the declaration spectrum feature vector of each power trading entity in each trading period.

[0027] In this embodiment, each declaration segment in the standardized declaration data sample is used as a declaration data node; based on the arrangement order between different declaration segments, the differences between declared electricity quantities, and the differences between declared prices, directed connections are constructed between different declaration data nodes to obtain a directed graph of declaration data; Hermitian connection matrix and degree matrix are constructed using the directed graph of declaration data; Hermitian-Laplace matrix is ​​constructed based on the Hermitian connection matrix and the degree matrix; eigenvalue decomposition is performed on the Hermitian-Laplace matrix to obtain eigenvectors; normalized clustering encoding is performed on the eigenvectors to obtain the declaration spectrum feature vectors of each power trading entity in each trading period.

[0028] In this embodiment, the Hermitian-Laplacian matrix is ​​constructed as follows: Definition of the first Tianhe Di The similarity measure of the evolution of the daily declaration curve is: ; in, Market entities In the Tianhe Di The similarity of the evolution of the Tianbao declaration curves was calculated based on Euclidean distance. This represents the theoretical maximum curve distance, used for normalization.

[0029] Furthermore, the Hermitian-Laplacian matrix is: ; in, i This refers to the serial number of the electricity trading entity. u , v Indicates the data node to be declared. Hermitian connection matrix data nodes in the middle and reporting u and reporting data nodes v The corresponding matrix elements, for conjugate, For reporting data nodes u With the data node of the declaration v Similarity of declaration features between them It is the root of unity of a complex number; The degree matrix is: ; in, For degree matrix, j The data nodes are represented in diagonal matrix form. For reporting data nodes u With the data node of the declaration j Similarity of declaration features between them; The Hermitian-Laplace matrix is: ; in, For Hermitian-Laplace matrix, It is an identity matrix.

[0030] Furthermore, the eigenvalue decomposition of the Hermitian-Laplacian matrix and the method for normalized clustering encoding are as follows: ; ; ; in, It is a length of T The vector, if the first u Heaven belongs to If the vector is clustered, then the first... u The value of each element is Conversely, it is 0. T The total number of days in the observation period. For vectors The result after normalization For market entities i The normalized optimal clustering coding result.

[0031] In the above process, each declaration segment in the standardized declaration data sample is used as a declaration data node. Directed connections are established based on the arrangement order, differences in declared electricity volume, and differences in declared price between different declaration segments, constructing a directed graph of declaration data that reflects the internal structure of the segmented declaration data. A Hermitian-Laplacian matrix is ​​constructed using the directed graph of declaration data, and eigenvalue decomposition and normalized clustering encoding are performed on the Hermitian-Laplacian matrix to obtain the declaration spectrum feature vectors of each power trading entity in each trading period.

[0032] Step S13: Using the declared spectrum feature vector, construct a multi-time series power trading entity relationship diagram at different time scales.

[0033] In this embodiment, multiple time-series windows with different time spans are set; the Euclidean distance between the declaration spectrum feature vectors of different power trading entities within each time-series window is calculated; based on the Euclidean distance, the declaration feature similarity is determined; using the declaration feature similarity, the connection relationship and connection weight between different power trading entities within each time-series window are determined; and a multi-time-series power trading entity relationship graph with different time scales including the connection relationship and the connection weight is constructed.

[0034] Specifically, construct separate systems for different historical window lengths. Each attribute graph consists of a view composed of the aforementioned extracted feature codes. The edge weights reflect the similarity of the reporting behavior between entities at the corresponding time scale, specifically: ; in, For a set of nodes, , for The edge set of time periods, for Time period node attribute feature matrix.

[0035] The adjacency matrix is ​​constructed by comparing the Euclidean distance and the cutoff distance between the price codes of market participants: ; ; in, for main time period and main body Euclidean distance of feature encoding, for Threshold for time period.

[0036] In this step, based on the temporal continuity of the power transaction declaration data, multiple time-series windows with different time spans are set up. Within each time-series window, the declaration spectrum feature vectors of each power transaction entity are collected, the declaration feature similarity between different power transaction entities is calculated, and the connection relationship and connection weight between entities are determined based on the declaration feature similarity, thereby constructing a multi-time-series entity relationship graph corresponding to different time scales.

[0037] Step S14: Based on each of the power trading entities, the declaration spectrum feature vectors, and the multi-time series power trading entity relationship graph, construct graph convolutional encoders at different time scales, and use the graph convolutional encoders to aggregate the declaration spectrum feature vectors of adjacent power trading entities to obtain declaration graph embedding features at different time scales.

[0038] In this embodiment, each power trading entity is used as a graph node, the declaration spectrum feature vector is used as the node input feature, and the connection relationship and connection weight in the multi-time series power trading entity relationship graph are used as graph adjacency information to construct graph convolutional encoders at different time scales. Attention coefficients of adjacent power trading entities are determined using an attention mechanism. Using the attention coefficients and the graph convolutional encoder, the declaration spectrum feature vectors of adjacent power trading entities are weighted and aggregated to obtain declaration graph embedding features at different time scales.

[0039] Specifically, the graph convolution process considering the attention mechanism includes: ; ; in, for l Layers can be trained to learn weights. This is the output of the first layer. This is the result after double encoding. for l Layer attention coefficient matrix.

[0040] The specific process is as follows: ; ; in, for i All neighbors are determined by the set of edges in the attribute graph. These are trainable row vectors.

[0041] In this step, each power trading entity is used as a graph node, the declaration spectrum feature vector is used as the node input feature, and the connection relationships and connection weights in the multi-time series power trading entity relationship graph are used as graph adjacency information. Graph convolutional encoders corresponding to different time scales are constructed respectively. The graph convolutional encoder uses an attention mechanism to determine the attention coefficients of adjacent power trading entities, and performs weighted aggregation of the declaration spectrum features of adjacent power trading entities at each time scale based on the attention coefficients. The declaration graph embedding features of each power trading entity at different time scales are then obtained.

[0042] Step S15: Optimize the graph convolutional encoder using the declared graph embedding features, determine the comprehensive declaration feature vector of each power trading entity using the optimized graph convolutional encoder, and generate the power trading declaration data analysis results of each power trading entity based on the comprehensive declaration feature vector.

[0043] In this embodiment, a joint training objective is constructed using a multi-time-series power trading entity relationship graph, declaration spectrum feature vectors, and declaration graph embedding features. Based on the joint training objective, the graph convolutional encoder, time-scale fusion parameters, and category center parameters are optimized. Using the optimized graph convolutional encoder and time-scale fusion parameters, the comprehensive declaration feature vector of each power trading entity is determined. Based on the comprehensive declaration feature vector, the feature similarity between different power trading entities, the category membership of each power trading entity, and the degree of feature change between adjacent trading periods are determined. The optimized category center parameters are used to determine the declaration feature category of each power trading entity. Based on the declaration feature category, the feature similarity, the category membership, and the degree of feature change, the power trading declaration data analysis results of each power trading entity are generated.

[0044] The joint training objectives include clustering constraint loss, feature reconstruction loss, and temporal consistency loss: ; in, and These are weighting coefficients used to balance the influence of different loss terms during training. For the purpose of joint training, For clustering constraint loss, For feature reconstruction loss, This represents a loss due to timing consistency issues.

[0045] In this step, the optimization process of the graph convolutional encoder is as follows: Based on the multi-time series power trading entity relationship graph, the declaration spectrum feature vector, and the declaration graph embedding features, a joint training objective is constructed, including clustering constraint loss, feature reconstruction loss, and time series consistency loss; by minimizing the joint training objective, the parameters, time scale fusion parameters, and class center parameters of the graph convolutional encoder at different time scales are jointly optimized to obtain the optimized graph convolutional encoder, and the comprehensive declaration feature vector of each power trading entity is determined.

[0046] Specifically, for each time period of the test set t The Euclidean distance is calculated based on the quotation attribute vector, and an adjacency matrix is ​​constructed through threshold sparsification. and node feature matrix Then, using the parameter set Θ obtained from the training phase optimization, a frozen graph attention forward propagation is performed on each view. The resulting multi-time embedding representation is as follows: ; By embedding and concatenating all time periods, we can obtain the overall embedding matrix of the test set: ; Determining the optimal number of clusters Then, the standard FCM (Fuzzy C-Means) algorithm is executed in the embedding space, and its iterative calculation is as follows: ; .

[0047] Iterate until satisfied The final soft membership matrix of the test set is The final clustering result is as follows: .

[0048] Furthermore, the specific construction methods for the clustering constraint loss, feature reconstruction loss, and temporal consistency loss in the joint training objective are as follows: The clustering loss is determined by summing the clustering losses of each time period and the weights of multiple time-series views, as shown below: ; ; ; in, for t Clustering constraint loss for time-period views, Weights for multiple time-series views.

[0049] Clustering loss for each time period Calculated using fuzzy membership degrees: ; ; ; in, for t Membership matrix for time period for t main time period i For cluster k membership degree m The ambiguity coefficient is set to 2 here. for t Inter-class separability coefficient for time periods.

[0050] Furthermore, other losses consist of feature reconstruction loss and temporal consistency loss to enhance the ability of the embedded representation to preserve the original declaration information and its stability across time scales. Feature reconstruction loss The method used to constrain the embedded representation to reconstruct the original node features after decoding is as follows: ; in, and This represents the feature reconstruction result obtained by decoding the embedding representation.

[0051] Meanwhile, considering the inherent continuity of reporting behavior across different time scales, a temporal consistency loss is introduced. Constrain the embedded representation under adjacent time views: .

[0052] Furthermore, based on the embedding matrix An adaptive selection of the number of clusters is performed using a clustering effectiveness index. The FCM algorithm is run in the embedding space and the XB (Xie-Beni clustering effectiveness) index is calculated.

[0053] .

[0054] The joint training objectives include clustering constraint loss, feature reconstruction loss, and temporal consistency loss. By minimizing the joint training objectives, the graph convolutional encoder, temporal scale fusion parameters, and class center parameters at different time scales are optimized. The multi-time-series power trading entity relationship graph and the declaration spectrum feature vector are input into the optimized graph convolutional encoder, and the optimized temporal scale fusion parameters are used to fuse the declaration graph embedding features at different time scales to obtain the comprehensive declaration feature vector for each power trading entity.

[0055] Specifically, the characteristic mixing index for declaration is: .

[0056] When an entity's membership degree to a certain cluster is significantly higher than that to other clusters, A smaller value indicates a clearer risk appetite in the application process; when an entity has a certain degree of affiliation to multiple clusters... A larger value indicates that the application behavior is in a transitional state between different strategy modes, and the behavior is more uncertain.

[0057] Based on the comprehensive declaration feature vector, analysis can be conducted from three aspects: relationships between entities, entity category affiliation, and changes between adjacent trading cycles. Feature similarity among different power trading entities is used to characterize the declaration correlation between entities; the category membership degree of each power trading entity and the optimized category centrality parameter are used together to determine the declaration feature category; and the degree of feature change between adjacent trading cycles is used to characterize the temporal changes in declaration features. The generated power trading declaration data analysis results include the aforementioned declaration feature categories, feature similarity, category membership, and degree of feature change.

[0058] In this embodiment, the graph convolutional neural network uses the connection relationships and connection weights of the multi-time series power trading entity relationship graph as graph adjacency information, and the declaration spectrum feature vector as node input features. It aggregates features through graph convolutional encoders at different time scales, and optimizes the graph convolutional encoder, time scale fusion parameters, and category center parameters through joint training objectives.

[0059] A diagram illustrating the similarity of features among different electricity trading entities is shown below. Figure 2 As shown in the diagram, the dimensionality reduction results between different power trading entities are as follows: Figure 3 As shown, the declaration feature analysis results are used to represent the feature similarity relationships between entities and the distribution of different declaration feature categories. The power trading declaration data analysis results include the declaration feature categories of each power trading entity, the feature similarity between different power trading entities, the category membership degree of each power trading entity, and the degree of feature change between adjacent trading periods. Declaration feature category indicates the category to which an entity's declaration feature belongs; feature similarity indicates the degree of correlation between declarations from different entities; category membership degree indicates the degree to which an entity belongs to a corresponding category; and the degree of feature change indicates how the entity's declaration features change between adjacent trading periods.

[0060] In this embodiment, the specific process of power transaction declaration data analysis based on multi-time-series graphs and graph convolutional encoders is as follows: Figure 4As shown, the specific steps include: data standardization processing, characterization of entity declaration trends based on spectral graphs, market cluster convergence analysis based on GCN (Graph Convolutional Networks), and spot market entity declaration detection. Compared with the prior art, the beneficial effects of this invention are: (1) In view of the segmented structure and quantity-price coupling characteristics of power transaction declaration data, each declaration segment is constructed as a directed graph node, and Hermitian-Laplacian spectral decomposition is used to extract the structural features of the declaration data. This can preserve the sequential relationship and directional information between declaration segments while unifying the feature dimensions, and reduce the loss of segmented declaration structure information by traditional statistical indicators.

[0061] (2) Construct a multi-time series power trading entity relationship graph based on the declaration spectrum characteristics within different time spans, so that the connection relationship between entities can be dynamically updated as the declaration data changes, and aggregate the associated entity features through a graph convolution encoder, which can simultaneously extract the entity's own declaration features and the inter-entity relationship features.

[0062] (3) By training with multi-time series convolutional graphs, the graph convolutional encoder, time scale fusion parameters and category center parameters at different time scales are jointly optimized so that the comprehensive declaration features can simultaneously reflect short-term declaration changes, medium-term behavioral evolution and long-term stable features, thereby improving the accuracy of power transaction declaration data feature representation.

[0063] As can be seen, this application provides a method for analyzing power transaction declaration data based on multi-time series graphs and graph convolutional encoders. This includes preprocessing segmented declaration data from multiple power trading entities to obtain standardized declaration data samples. This improves the consistency of segmented declaration data representation across different trading days and periods for different power trading entities, reducing the impact of data structure and scale differences on subsequent analysis. Using each declaration segment in the standardized declaration data sample as a node, a directed graph of the declaration data is constructed. An Hermit-Laplace matrix is ​​then constructed using this directed graph, and eigenvalue decomposition is performed on the Hermit-Laplace matrix to obtain the declaration spectrum feature vectors for each power trading entity in each trading period. This method preserves the sequential relationships and directional information between declaration segments while unifying the feature dimensions, reducing the loss of segmented declaration structure information by traditional statistical indicators. Finally, using the declaration spectrum feature vectors, a multi-time series graph with different time scales is constructed. The power trading entity relationship graph enhances the ability to identify the reporting relationships between power trading entities at different time scales. Based on the power trading entities, reporting spectrum feature vectors, and multi-time series power trading entity relationship graphs, graph convolutional encoders at different time scales are constructed. These encoders aggregate the reporting spectrum feature vectors of adjacent power trading entities to obtain reporting graph embedding features at different time scales, thus enabling the extraction of entity-specific reporting features and inter-entity relationship features. The graph convolutional encoders are then optimized using these reporting graph embedding features. The optimized encoders determine the comprehensive reporting feature vectors for each power trading entity, and the power trading reporting data analysis results for each entity are generated based on these comprehensive feature vectors. This approach reflects short-term reporting changes, medium-term behavioral evolution, and long-term stable characteristics, improving the accuracy of power trading reporting data feature representation.

[0064] See Figure 5 As shown, this embodiment of the invention discloses a power transaction declaration data analysis device based on a multi-time-series graph and a graph convolutional encoder, which may specifically include: Preprocessing module 11 is used to preprocess the segmented declaration data of multiple power trading entities to obtain standardized declaration data samples; the segmented declaration data is data submitted within multiple consecutive trading days and multiple trading periods; The feature decomposition module 12 is used to construct a directed graph of declaration data with each declaration segment in the standardized declaration data sample as a node, construct a Hermit-Laplace matrix using the directed graph of declaration data, and perform feature decomposition on the Hermit-Laplace matrix to obtain the declaration spectrum feature vector of each power trading entity in each trading period. The relationship graph construction module 13 is used to construct a multi-time series power trading entity relationship graph at different time scales using the declared spectrum feature vector; The aggregation module 14 is used to construct graph convolutional encoders at different time scales based on each of the power trading entities, the declaration spectrum feature vectors, and the multi-time series power trading entity relationship graph. The graph convolutional encoders are used to aggregate the declaration spectrum feature vectors of adjacent power trading entities to obtain declaration graph embedding features at different time scales. The analysis result generation module 15 is used to optimize the graph convolution encoder by using the embedded features of the declared graph, determine the comprehensive declaration feature vector of each power trading entity by using the optimized graph convolution encoder, and generate the power trading declaration data analysis results of each power trading entity based on the comprehensive declaration feature vector.

[0065] In some specific embodiments, the preprocessing module 11 may specifically include: The segmented declaration data acquisition module is used to acquire segmented declaration data submitted by multiple power trading entities over multiple consecutive trading days and multiple trading periods; the segmented declaration data includes the declared electricity volume and declared price corresponding to each declaration segment; The correction module is used to process missing values ​​and correct outliers in the segmented declaration data to obtain corrected segmented declaration data. The unified application segment structure module is used to process the corrected segmented application data according to the preset number of application segments, so as to obtain segmented application data with unified application segment structure; the unified application segment structure processing includes interpolation, merging or completion processing; The normalization module is used to normalize segmented application data with a unified application segment structure to obtain standardized application data samples.

[0066] In some specific embodiments, the feature decomposition module 12 may specifically include: The declaration data node determination module is used to identify each declaration segment in the standardized declaration data sample as a declaration data node. The directed connection construction module is used to construct directed connections between different declared data nodes based on the arrangement order between different declaration segments, the differences between declared electricity volumes, and the differences between declared prices, thereby obtaining a directed graph of declaration data; The matrix construction module is used to construct a Hermitian connectivity matrix and a degree matrix using the directed graph of the declared data; The Hermit-Laplace matrix construction module is used to construct the Hermit-Laplace matrix based on the Hermitian connection matrix and the degree matrix. The eigenvalue decomposition submodule is used to perform eigenvalue decomposition on the Hermit-Laplace matrix to obtain eigenvectors; The normalized clustering coding module is used to perform normalized clustering coding on the feature vector to obtain the declaration spectrum feature vector of each power trading entity in each trading period.

[0067] In some specific embodiments, the Hermitian connectivity matrix is: ; in, i This refers to the serial number of the electricity trading entity. u , v Indicates the data node to be declared. Hermitian connection matrix data nodes in the middle and reporting u and reporting data nodes v The corresponding matrix elements, for conjugate, For reporting data nodes u With the data node of the declaration v Similarity of declaration features between them It is the root of unity of a complex number; The degree matrix is: ; in, For degree matrix, j Indicates the data node to be declared. For reporting data nodes u With the data node of the declaration j Similarity of declaration features between them; The Hermitian-Laplace matrix is: ; in, For Hermitian-Laplace matrix, It is an identity matrix.

[0068] In some specific embodiments, the relationship graph construction module 13 may specifically include: The timing window setting module is used to set multiple timing windows with different time spans; The Euclidean distance calculation module is used to calculate the Euclidean distance between the declaration spectrum feature vectors of different power trading entities within each of the aforementioned time windows; The declaration feature similarity determination module is used to determine the declaration feature similarity based on the Euclidean distance; The connection information determination module is used to determine the connection relationship and connection weight between different power trading entities within each time window by utilizing the similarity of the declared features; A multi-time series relationship graph generation module is used to construct a multi-time series power trading entity relationship graph with different time scales, including the connection relationships and the connection weights.

[0069] In some specific embodiments, the aggregation module 14 may specifically include: The encoder construction module is used to construct graph convolutional encoders at different time scales by taking each power trading entity as a graph node, the declaration spectrum feature vector as the node input feature, and the connection relationship and connection weight in the multi-time series power trading entity relationship graph as graph adjacency information. The attention coefficient determination module is used to determine the attention coefficient of adjacent power trading entities using an attention mechanism. The weighted aggregation module is used to perform weighted aggregation of the declaration spectrum feature vectors of adjacent power trading entities using the attention coefficients and the graph convolution encoder to obtain declaration graph embedding features at different time scales.

[0070] In some specific embodiments, the analysis result generation module 15 may specifically include: A joint training objective construction module is used to construct joint training objectives using multi-time series power trading entity relationship graphs, declaration spectrum feature vectors, and declaration graph embedding features. The comprehensive declaration feature vector determination module is used to optimize the graph convolutional encoder, time scale fusion parameters and category center parameters based on the joint training objective, and to determine the comprehensive declaration feature vector of each power trading entity using the optimized graph convolutional encoder and time scale fusion parameters. The feature change degree determination module is used to determine the feature similarity between different power trading entities, the category membership degree of each power trading entity, and the feature change degree between adjacent trading cycles based on the comprehensive declaration feature vector. The module for determining the declaration feature category is used to determine the declaration feature category of each power trading entity using the optimized category center parameters; The power transaction declaration data analysis result generation module is used to generate power transaction declaration data analysis results for each power transaction entity based on the declaration feature category, the feature similarity, the category membership degree, and the feature change degree.

[0071] Figure 6 This is a schematic diagram of an electronic device provided in an embodiment of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the power transaction declaration data analysis method based on multi-time-series graphs and graph convolutional encoders disclosed in any of the foregoing embodiments.

[0072] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0073] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0074] The operating system 221 manages and controls the various hardware devices on the electronic device 20 and the computer program 222 to enable the processor 21 to perform operations and processing on the data 223 in the memory 22. It can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the power transaction declaration data analysis method based on multi-time-series graphs and graph convolutional encoders executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the power transaction declaration data analysis device based on multi-time-series graphs and graph convolutional encoders from external devices, and may also include data collected by its own input / output interface 25.

[0075] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0076] Furthermore, this application also discloses a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by a processor, it implements the steps of the power transaction declaration data analysis method based on multi-time-series graphs and graph convolutional encoders disclosed in any of the foregoing embodiments.

[0077] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such 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 said element.

[0078] The foregoing has provided a detailed description of the power transaction declaration data analysis method, apparatus, equipment, and medium based on multi-time-series graphs and graph convolutional encoders provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. 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 the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders, characterized in that, include: The segmented declaration data from multiple power trading entities are preprocessed to obtain standardized declaration data samples; The segmented reporting data refers to data submitted over multiple consecutive trading days and trading periods. Using each declaration segment in the standardized declaration data sample as a node, a directed graph of declaration data is constructed. The Hermit-Laplace matrix is ​​constructed using the directed graph of declaration data. The Hermit-Laplace matrix is ​​then subjected to eigenvalue decomposition to obtain the declaration spectrum feature vector of each power trading entity in each trading period. Using the aforementioned declaration spectrum feature vectors, a multi-time series power trading entity relationship diagram at different time scales is constructed; Based on each of the power trading entities, the declaration spectrum feature vectors, and the multi-time series power trading entity relationship graph, graph convolutional encoders at different time scales are constructed. The graph convolutional encoders are used to aggregate the declaration spectrum feature vectors of adjacent power trading entities to obtain declaration graph embedding features at different time scales. The graph convolutional encoder is optimized using the embedded features of the declared graph. The optimized graph convolutional encoder is used to determine the comprehensive declaration feature vector of each power trading entity. Based on the comprehensive declaration feature vector, the power trading declaration data analysis results of each power trading entity are generated.

2. The method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders according to claim 1, characterized in that, The preprocessing of segmented declaration data from multiple power trading entities to obtain standardized declaration data samples includes: The system acquires segmented declaration data submitted by multiple power trading entities over multiple consecutive trading days and trading periods; the segmented declaration data includes the declared electricity volume and declared price corresponding to each declaration segment. The segmented declaration data is processed for missing values ​​and outliers are corrected to obtain corrected segmented declaration data; According to the preset number of application segments, the corrected segmented application data is processed to unify the application segment structure, resulting in segmented application data with a unified application segment structure; the unified application segment structure processing includes interpolation, merging or completion processing; Normalize the segmented application data with a unified application segment structure to obtain standardized application data samples.

3. The method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders according to claim 2, characterized in that, Using each declaration segment in the standardized declaration data sample as a node, a directed graph of declaration data is constructed. A Hermit-Laplace matrix is ​​then constructed using this directed graph. Eigenvalue decomposition is performed on the Hermit-Laplace matrix to obtain the declaration spectrum feature vectors of each power trading entity in each trading period, including: Each segment in the standardized declaration data sample is used as a declaration data node; Based on the arrangement order between different declaration segments, the differences between declared electricity volumes, and the differences between declared prices, directed connections are constructed between the different declaration data nodes to obtain a directed graph of declaration data; Construct a Hermitian connectivity matrix and a degree matrix using the directed graph derived from the declared data; Based on the Hermitian connectivity matrix and the degree matrix, construct the Hermitian-Laplace matrix; The Hermitian-Laplace matrix is ​​subjected to eigenvalue decomposition to obtain eigenvectors; The feature vectors are normalized and clustered to obtain the declaration spectrum feature vectors of each power trading entity in each trading period.

4. The method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders according to claim 3, characterized in that, The Hermitian connection matrix is: ； in, i This refers to the serial number of the electricity trading entity. u , v Indicates the data node to be declared. Hermitian connection matrix data nodes in the middle and reporting u and reporting data nodes v The corresponding matrix elements, for conjugate, For reporting data nodes u With the data node of the declaration v Similarity of declaration features between them It is the root of unity of a complex number; The degree matrix is: ； in, For degree matrix, j Indicates the data node to be declared. For reporting data nodes u With the data node of the declaration j Similarity of declaration features between them; The Hermitian-Laplace matrix is: ； in, For Hermitian-Laplace matrix, It is an identity matrix.

5. The method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders according to claim 1, characterized in that, Using the aforementioned declaration spectrum feature vectors, a multi-time series power trading entity relationship graph is constructed at different time scales, including: Set multiple time series windows with different time spans; Calculate the Euclidean distance between the eigenvectors of the declaration spectrum of different power trading entities within each of the aforementioned time windows; Based on the Euclidean distance, the similarity of the declared features is determined; By utilizing the similarity of the declared features, the connection relationships and connection weights between different power trading entities within each of the time windows are determined; Construct a multi-time series power trading entity relationship graph that includes the connection relationships and connection weights at different time scales.

6. The method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders according to claim 5, characterized in that, Based on each of the power trading entities, the declaration spectrum feature vectors, and the multi-time series power trading entity relationship graph, a graph convolutional encoder at different time scales is constructed. This graph convolutional encoder is then used to aggregate the declaration spectrum feature vectors of adjacent power trading entities to obtain declaration graph embedding features at different time scales, including: Each power trading entity is used as a graph node, the declaration spectrum feature vector is used as the node input feature, and the connection relationship and connection weight in the multi-time series power trading entity relationship graph are used as graph adjacency information to construct a graph convolutional encoder at different time scales. The attention coefficient of adjacent electricity trading entities is determined using an attention mechanism; Using the attention coefficients and the graph convolutional encoder, the declaration spectrum feature vectors of adjacent power trading entities are weighted and aggregated to obtain declaration graph embedding features at different time scales.

7. The method for analyzing power transaction declaration data based on multi-time-series graphs and graph convolutional encoders according to any one of claims 1 to 6, characterized in that, The graph convolutional encoder is optimized using the embedded features of the declared graph. The optimized graph convolutional encoder is then used to determine the comprehensive declared feature vector for each power trading entity. Based on the comprehensive declared feature vector, the power trading declaration data analysis results for each power trading entity are generated, including: A joint training objective is constructed using a multi-time series power trading entity relationship graph, declaration spectrum feature vector, and declaration graph embedding features. Based on the joint training objective, the graph convolutional encoder, time-scale fusion parameters, and class center parameters are optimized. Using the optimized graph convolutional encoder and time-scale fusion parameters, the comprehensive declaration feature vector of each power trading entity is determined. Based on the comprehensive declaration feature vector, the feature similarity between different power trading entities, the category membership degree of each power trading entity, and the degree of feature change between adjacent trading cycles are determined. The optimized category center parameters are used to determine the declaration characteristic categories of each power trading entity; Based on the declared feature categories, feature similarity, category membership, and feature variation, the power trading declaration data analysis results for each power trading entity are generated.

8. A power transaction declaration data analysis device based on multi-time-series graphs and graph convolutional encoders, characterized in that, include: The preprocessing module is used to preprocess the segmented declaration data of multiple power trading entities to obtain standardized declaration data samples; The segmented reporting data refers to data submitted over multiple consecutive trading days and trading periods. The feature decomposition module is used to construct a directed graph of declaration data with each declaration segment in the standardized declaration data sample as a node, construct a Hermit-Laplace matrix using the directed graph of declaration data, and perform feature decomposition on the Hermit-Laplace matrix to obtain the declaration spectrum feature vector of each power trading entity in each trading period. The relationship graph construction module is used to construct a multi-time series power trading entity relationship graph at different time scales using the declared spectrum feature vector; The aggregation module is used to construct graph convolutional encoders at different time scales based on each of the power trading entities, the declaration spectrum feature vectors, and the multi-time series power trading entity relationship graph. The graph convolutional encoders are used to aggregate the declaration spectrum feature vectors of adjacent power trading entities to obtain declaration graph embedding features at different time scales. The analysis result generation module is used to optimize the graph convolutional encoder using the embedded features of the declared graph, determine the comprehensive declaration feature vector of each power trading entity using the optimized graph convolutional encoder, and generate the power trading declaration data analysis results of each power trading entity based on the comprehensive declaration feature vector.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the power transaction declaration data analysis method based on a multi-time-series graph and graph convolutional encoder as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein, when the computer programs are executed by a processor, they implement the power transaction declaration data analysis method based on a multi-time-series graph and graph convolutional encoder as described in any one of claims 1 to 7.

Images