Adjustable resource multi-dimension data center based on metrology center

Abstract

The application relates to the technical field of electric power Internet of Things and data processing, and particularly discloses a multi-dimensional data center construction system of adjustable resources based on a measurement center, which comprises the following steps: collecting electrical measurement data of adjustable resources in high precision by an edge collection terminal; generating candidate edges by fusing the correlation and power transfer entropy of the measurement data; obtaining a weighted space-time correlation graph by combining a graph attention network; performing Laplace interpolation on missing data by using the graph and outputting a confidence degree; identifying a low-confidence degree area according to the spatial distribution of the confidence degree, and dynamically optimizing a collection frequency and a cleaning strategy; generating a virtual resource cluster with electrical close coupling by adopting hierarchical clustering based on the edge weight of the graph; and constructing an adjustable feasible region and optimizing the distribution of regulation and control instructions by taking the edge weight of the graph as a constraint in the cluster, so that the problems of topology dynamic tracking, data quality enhancement, resource aggregation rationality and regulation and control collaboration are solved.

Description

Technical Field

[0001] This invention relates to the field of power Internet of Things and data processing technology, specifically to a system for constructing a multi-dimensional data platform based on adjustable resources from a measurement center. Background Technology

[0002] With the large-scale integration of adjustable resources into new power systems, building a multi-dimensional data platform based on measurement centers has become a key infrastructure for achieving observable, measurable, and controllable resources. However, the following technical challenges exist in practical applications: Topology dynamics are difficult to track: Frequent adjustments to the distribution network topology mean that static archives cannot reflect real-time electrical connections, leading to significant errors in resource aggregation and control decisions; Inconsistent data quality and isolated processing: Measurement data is incomplete or anomaly-prone, and traditional methods process data from each node in isolation, failing to utilize the electrical coupling characteristics between nodes for collaborative repair, resulting in poor interpolation performance; Lack of electrical basis for resource aggregation: Existing aggregations are mostly based on geographical location or resource type, without considering the electrical coupling strength between nodes, making it easy for aggregate adjustments to trigger internal limit exceedances; Local optimization of control strategies: Traditional control is based on individual resources or fixed areas, ignoring electrical coupling constraints between nodes, making it difficult to achieve global coordination; Disconnect between data governance and business applications: Data cleaning strategies are fixed and fail to be dynamically adjusted according to data quality; Model training does not consider sample reliability, resulting in limited prediction accuracy.

[0003] To this end, the present invention provides a system for constructing an adjustable resource multi-dimensional data platform based on a measurement center. Summary of the Invention

[0004] The purpose of this invention is to provide a system for constructing an adjustable resource multi-dimensional data platform based on a measurement center. This system addresses the problems mentioned above, such as existing aggregations being based primarily on geographical location or resource type without considering the electrical coupling strength between nodes, leading to internal limit violations during aggregation adjustment; local optimization of control strategies: traditional control is based on a single resource or fixed area, ignoring electrical coupling constraints between nodes, making global collaboration difficult; disconnect between data governance and business applications: data cleaning strategies are fixed and fail to be dynamically adjusted according to data quality; and model training does not consider sample reliability, resulting in limited prediction accuracy.

[0005] The objective of this invention can be achieved through the following technical solution: a system for constructing a multi-dimensional data platform based on adjustable resources of a measurement center, comprising the following modules: Graph construction module: Based on measurement data, correlation analysis is performed on each pair of adjustable resource nodes, power transfer entropy is calculated, initial edge weights are obtained by fusion and candidate edges are screened, and graph attention network training is used to update the edge weights to obtain a weighted spatiotemporal correlation graph. Data correction module: When missing measurement data is detected, the module searches for neighboring nodes associated with the missing data node in the weighted spatiotemporal correlation graph, performs missing data interpolation using graph signal processing algorithms, and outputs the interpolation confidence score. Acquisition optimization module: Divides the target area into multiple grid areas, and identifies low-confidence areas by statistically analyzing the distribution of interpolation confidence in each grid area, and optimizes the acquisition priority of low-confidence areas; Cluster aggregation module: Based on the weights of each edge of the weighted spatiotemporal correlation graph, hierarchical clustering is used to aggregate nodes into virtual resource clusters and generate the aggregation attributes of the virtual resource clusters; Regulation and optimization module: Within the virtual resource cluster, the weighted spatiotemporal correlation graph edge weights are used as constraints to construct a feasible domain for cluster power regulation. When scheduling instructions are issued, the regulation amount of each node is optimized and allocated under the constraints of the weighted spatiotemporal correlation graph.

[0006] Furthermore, the process of obtaining initial edge weights and filtering candidate edges through fusion is as follows: Take the voltage amplitude time series of all adjustable resource nodes within the current time window, and calculate the Pearson correlation coefficient of voltage amplitude between any two adjustable resource nodes; Take the active power time series of all adjustable resource nodes within the current time window, and calculate the power transfer entropy between any two adjustable resource nodes; The initial edge weights of each adjustable resource node pair are obtained by summing the normalized Pearson correlation coefficient and power transfer entropy. Adjustable resource node pairs whose initial edge weights meet the requirements are selected as candidate edges.

[0007] Furthermore, the method for updating the weights of each edge using graph attention network training is as follows: Construct an initial graph using all adjustable resource nodes as vertices and candidate edges as edges; For each adjustable resource node, a feature vector is constructed, which includes resource type, rated capacity, historical average power, historical voltage fluctuation rate, and geographical location coordinates, forming a node feature matrix; A graph attention network is used for training. After training, all nodes are input into the graph attention network to obtain the connection probabilities of all node pairs. Edges whose connection probabilities meet the requirements are taken as the final topological edges, and the edge weights are updated to the attention coefficients output by the graph attention network to form a weighted spatiotemporal correlation graph.

[0008] Furthermore, the method of using graph signal processing algorithms for missing data interpolation is as follows: For any node with missing measurement data at time t, find the top k neighbor nodes with the highest weight that have an edge with the node with missing measurement data in the weighted spatiotemporal correlation graph. Constructing the Laplacian Matrix in Graph Signal Processing , where A is the adjacency matrix of the weighted spatiotemporal correlation graph and D is the degree matrix; The interpolated value of the node with missing measurement data at time t is calculated by weighted average of the measurement values ​​of the first k neighboring nodes at time t, with the weights being the graph edge weights.

[0009] Furthermore, the output method of the interpolation confidence is as follows: Confidence level is estimated based on the correlation strength and difference error of neighboring nodes: , where Q j A represents the quality index of the neighbor node data itself, set to 1 if there are no anomalies, where K is the number of neighbor nodes, and A is the quality index of the neighbor node data itself. ij Let N(i) be the edge weights between the missing data node i on both sides and each neighbor node j, and let N(i) be the set of neighbor nodes.

[0010] Furthermore, the method for identifying the low-confidence region is as follows: The target area is divided into unit grid areas. The interpolation confidence distribution of all interpolation records in each unit grid area is statistically analyzed at fixed time windows, and the average confidence is calculated. Interpolation records that meet the interpolation confidence requirements are marked as low-confidence records. The proportion of low-confidence records to the total number of interpolation records in the grid area is calculated to obtain the proportion of low-confidence records. Grid regions that meet both the requirements for average confidence and the proportion of low confidence records are marked as normal regions; conversely, grid regions that do not meet the requirements for either average confidence or the proportion of low confidence records are marked as low confidence regions.

[0011] Furthermore, the optimization process for prioritizing the collection of low-confidence regions is as follows: For grid regions marked as low-confidence areas, perform at least one of the following optimization measures: Increase the sampling frequency of adjustable resource access point terminals in low-confidence areas, shorten the data upload cycle, relax the anomaly detection threshold, adopt a conservative interpolation strategy, use only high-confidence nodes among neighbors for interpolation, and increase the number of neighbor nodes used during interpolation.

[0012] Furthermore, the aggregation method of the virtual resource cluster is as follows: All adjustable resource nodes are integrated into a node set, and the edge weights between each node pair in the weighted spatiotemporal correlation graph are constructed into a weighted adjacency matrix; The edge weights are converted into the distances between nodes using a linear transformation: distance between nodes = 1 - edge weights, thus linearly mapping the similarity to the [0,1] interval. Each node is initialized as an independent cluster. A hierarchical clustering algorithm based on the full-link criterion is used. In each iteration, the two clusters with the smallest distance are found and merged. The distance between the new cluster and other clusters is defined as the maximum distance between a node in the new cluster and all nodes in the other cluster. Repeat the merging process until all nodes are merged into a single cluster, construct a tree diagram and record the merging information; The elbow rule is used to determine the optimal number of clusters. Based on the determined number of clusters, the tree diagram is divided to obtain k virtual resource clusters, each containing multiple adjustable resource nodes.

[0013] Furthermore, the feasible region for cluster power adjustment is constructed as follows: For any virtual resource cluster, the electrical distance between any pair of nodes within the cluster is defined as the reciprocal of the graph edge weights; A linear DistFlow model is used to describe the power flow relationship within the cluster. Taking the total regulation demand of the cluster as a variable, the feasible region of cluster power regulation that satisfies node power balance, node regulation range, voltage constraints and branch capacity constraints is solved, and the feasible region interval and the corresponding regulation cost curve are obtained.

[0014] Furthermore, the optimized allocation method for the adjustment amounts of each node is as follows: Once the scheduling center issues the cluster adjustment target, an optimization model is established with the goal of minimizing the weighted sum of adjustment cost and credibility penalty. The constraints of the optimization model include: the sum of the power regulation of each node is equal to the cluster regulation target, the power regulation of each node is within its adjustable range, and branch power constraints and node voltage constraints. Among them, the branch power constraint includes: the power change of each branch is limited by its thermal stability limit, that is, the absolute value of the sum of the original power of the branch and the adjusted power must be less than or equal to the thermal stability limit of the branch. Node voltage constraints include: the voltage amplitude of each node must be kept within a safe range; The optimal adjustment amount for each node is obtained by solving the optimization model using linear programming.

[0015] The beneficial effects of this invention are as follows: By using correlation analysis and graph attention networks, electrical topology is constructed and updated in real time, eliminating the reliance on static files, greatly improving the accuracy of topology identification, and achieving precise tracking of dynamic topology. Based on the map, collaborative interpolation and correction improve the accuracy of missing data recovery and the accuracy of anomaly detection; the interpolation confidence drives the dynamic adjustment of the cleaning strategy, optimizes resource utilization, and enhances the quality of measurement data. Virtual clusters are generated based on hierarchical clustering of electrical coupling strength to ensure tight coupling within the clusters, reduce the risk of exceeding limits, lay the foundation for coordinated regulation, and make the aggregation of resources electrically reasonable. By constructing a feasible region for regulation under graph constraints and optimizing instruction allocation, the regulation cost is reduced, the instruction execution success rate is improved, and global collaborative optimization of regulation is achieved. Imputation confidence-weighted training improves model prediction accuracy, and model evaluation feedback guides data governance, forming a continuous optimization mechanism and achieving a closed loop of data and business collaboration. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a functional module diagram of the adjustable resource multi-dimensional data platform construction system based on the measurement center of this invention; Figure 2 This is a flowchart for identifying low-confidence regions in this invention. Detailed Implementation

[0018] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0019] Example: Please refer to Figure 1 As shown, the adjustable resource multi-dimensional data platform construction system based on a measurement center described in this invention addresses the construction of such a platform by using edge acquisition terminals to synchronously and with high precision collect electrical measurement data of adjustable resources. Candidate edges are generated by fusing correlation and power transfer entropy of the measurement data, and a weighted spatiotemporal correlation graph is obtained through graph attention network training. The graph is used to perform Laplace interpolation on missing data and output confidence levels. Low-confidence regions are identified based on the spatial distribution of confidence levels, and the acquisition frequency and cleaning strategy are dynamically optimized. Hierarchical clustering is used based on the graph edge weights to generate tightly coupled virtual resource clusters. Within the cluster, an adjustable feasible region is constructed using the graph edge weights as constraints, and the allocation of control commands is optimized. This solves multiple coupling problems such as dynamic topology tracking, data quality enhancement, rationality of resource aggregation, and synergistic control. Specifically, it includes the following modules: Graph construction module: Based on measurement data, correlation analysis is performed on each pair of adjustable resource nodes, power transfer entropy is calculated, initial edge weights are obtained by fusion and candidate edges are screened, and graph attention network training is used to update the edge weights to obtain a weighted spatiotemporal correlation graph. The process of acquiring the measurement data includes: An edge data acquisition terminal is deployed at each adjustable resource access point. The data collected by the terminal includes: Data types collected: instantaneous three-phase voltage, instantaneous three-phase current, active power, reactive power, frequency, and equipment operating status codes (e.g., normal / fault / offline). The sampling frequency is 1 time per second (1Hz) under normal operating conditions. When a sudden change in local electrical quantity is detected (such as a voltage change rate exceeding 5% / s), the frequency is automatically increased to 100Hz and continues until the event ends. Time synchronization: All gateways are equipped with GPS / BeiDou dual-mode timing modules, with synchronization accuracy better than 1 millisecond, ensuring the consistency of time profiles of data across nodes; The process of performing correlation analysis on each pair of adjustable resource nodes includes: Take the voltage amplitude time series and active power time series of all adjustable resource nodes within the current time window, with a time resolution of one second; For any pair of adjustable resource nodes, calculate the Pearson correlation coefficient of the voltage amplitude. The closer the Pearson correlation coefficient is to 1, the more synchronized the voltage fluctuations of the two nodes are and the tighter the electrical connection. The calculation process of the power transfer entropy between the adjustable resource nodes includes: The formula for the transition entropy between node i and node j is: , where P i P j t represents the power of node i and node j, respectively, and t is the timestamp; The power transfer entropy is obtained by calculating the active power transfer entropy between any two nodes according to the transfer entropy calculation formula. It is understandable that the physical meaning of power transfer entropy is: high transfer entropy indicates that the power change of one node can effectively predict the future change of another node, indicating the electrical upstream and downstream relationship, and is used to determine the causal direction of the regulation effect; For any pair of adjustable resource nodes: The Pearson correlation coefficient and power transfer entropy are normalized and summed to obtain the initial edge weights. Edges with initial edge weights greater than the threshold are selected as candidate edges. The generation process of the spatiotemporal correlation map includes: A spatiotemporal relational graph is constructed using all adjustable resource nodes as vertices and candidate edges as edges, with the weight of each edge being the fusion weight of the candidate edges. The feature vector of each adjustable resource node includes: resource type, rated capacity, historical average power, historical voltage fluctuation rate, and geographical location coordinates, resulting in a node feature matrix; Obtain known topology labels from the power grid dispatching system, such as the set of nodes downstream of substation outgoing switches and known feeder connection relationships; The training process of the graph attention network includes: It employs two attention layers, each containing eight attention heads, with a hidden layer dimension of 64. The input node feature matrix is ​​processed by the attention mechanism to calculate the influence coefficient of neighbor nodes on the central node, dynamically aggregates neighbor information, and outputs the updated adjustable resource node embedding. In the last layer, the adjustable resource nodes are embedded into a binary classifier to predict whether there is an electrical connection edge between each pair of adjustable resource nodes. The loss function is weighted cross-entropy. The loss is calculated for edges with known labels, and the GAT parameters are updated through backpropagation. After training, all adjustable resource nodes are input into GAT to obtain the connection probabilities of all adjustable resource node pairs. Edges with probabilities greater than the preset probability are taken as the final topology edges, and the weights of the edges are updated to the attention coefficients output by GAT to form a weighted topology graph. Understandably, GAT can adaptively learn the influence weights of different neighboring nodes on the central node. It not only utilizes the correlation of measurement data but also integrates the static attributes of adjustable resource nodes to infer the implicit electrical connection relationships. It is especially suitable for scenarios where the topology changes dynamically and is partially unknown, and can obtain high-precision global topology with a small number of labels. It should be noted that the purpose of constructing the spatiotemporal correlation graph is to lay the data foundation for the entire system, so that subsequent data quality enhancement, resource aggregation, and regulation optimization can be carried out under the guidance of real electrical relationships, avoiding errors caused by relying on static archives; Data correction module: When missing measurement data is detected, the module searches for neighboring nodes associated with the missing data node in the weighted spatiotemporal correlation graph, performs missing data interpolation using graph signal processing algorithms, and outputs the interpolation confidence score. The missing data imputation process includes: For any node with missing measurement data at time t, find the top k neighbor nodes (k=5) in the spatiotemporal correlation graph that have an edge with the node with missing measurement data. Constructing the Laplacian Matrix in Graph Signal Processing , where A is the adjacency matrix (edge ​​weights) and D is the degree matrix; The value of a node with missing measurement data is calculated by weighting the first k neighboring nodes, with the weights being the graph edge weights. If the data of node i is outside the normal range, the expected value is calculated using the weighted average of the neighbors. If the deviation between the actual value and the expected value exceeds three times the standard deviation, the expected value is used to replace it and it is marked as map correction data. The calculation process for the interpolation confidence level includes: Confidence level is estimated based on the correlation strength and difference error of neighboring nodes: , where Q j A represents the quality index of the neighbor node data itself, set to 1 if there are no anomalies, where K is the number of neighbor nodes, and A is the quality index of the neighbor node data itself. ij Let N(i) be the edge weight between the missing data node i on both sides and each neighbor node j, and let N(i) be the set of neighbor nodes with a confidence level of 1, reflecting the reliability of the imputed value. It should be noted that the purpose of correcting missing or abnormal data is to transform isolated node data problems into collaborative repair problems on the graph, significantly improve data integrity, and provide key parameters (confidence) for adaptive management of data quality. Acquisition optimization module: Divides the target area into multiple grid areas, and identifies low-confidence areas by statistically analyzing the distribution of interpolation confidence in each grid area, and optimizes the acquisition priority of low-confidence areas; Please see Figure 2 As shown, the process for identifying the low-confidence region includes: The target area is divided into unit grid areas, and the nodes in each unit grid area are counted. For any given unit grid region: Using a fixed time window (e.g., 24 hours) as the period, statistically analyze the interpolation confidence distribution of all interpolation records within each unit grid area, and calculate the average confidence level. The interpolation confidence of each node within a unit grid area is compared with the preset confidence. Interpolation records with an interpolation confidence less than or equal to the preset confidence are marked as low confidence records. The proportion of low confidence records to the total number of interpolation records in the historical period is calculated to obtain the proportion of low confidence records. Unit grid areas with an average confidence level greater than the confidence threshold and a low confidence level record ratio less than the preset ratio are marked as normal areas; conversely, unit grid areas with an average confidence level less than or equal to the confidence threshold and a low confidence level record ratio greater than or equal to the preset ratio are marked as low confidence areas. The process of optimizing the collection priority of low-confidence regions includes: Increase the sampling frequency of terminals in low-confidence areas (e.g., from 1 minute to 30 seconds), or shorten the data upload cycle to obtain more raw data and reduce reliance on interpolation; The anomaly detection threshold has been relaxed from a strict 3σ to 4σ to avoid a large amount of normal data being mistakenly judged as anomalies and discarded due to poor data quality; Conservative interpolation can be temporarily adopted (such as using only nodes with high confidence among the neighbors to participate in interpolation), or the number of neighbors during interpolation can be increased to smooth the results; Understandably, through confidence feedback, the system can adaptively allocate resources to areas with poor data quality while avoiding over-cleaning, forming a closed-loop optimization and gradually improving the overall data quality. It should be noted that the purpose of optimizing the collection priority of low-confidence areas based on interpolation confidence is to enable the data platform to have adaptive capabilities, tilt limited resources toward areas with poor data quality, and gradually improve the overall data quality through dynamic adjustment, avoiding the drawbacks of a one-size-fits-all strategy. Cluster aggregation module: Based on the weights of each edge of the weighted spatiotemporal correlation graph, hierarchical clustering is used to aggregate nodes into virtual resource clusters and generate the aggregation attributes of the virtual resource clusters; The aggregation process of the virtual resource cluster includes: All adjustable resource nodes are integrated into a node set, and the edge weights between each pair of nodes are used to construct a weighted adjacency matrix; Hierarchical clustering is usually based on distance dissimilarity, which requires converting edge weights into distances between nodes. A linear transformation is used, where distance between nodes = 1 - edge weight, which linearly maps similarity to the interval [0,1]. When the edge weight is 1, the distance is 0, and when the weight is 0, the distance is 1. Initialize each node as an independent cluster, and simultaneously initialize the cluster index list. Calculate the distance between all cluster pairs to obtain the initial inter-cluster distance matrix. Find the two matrices with the minimum distance in the initial cluster distance matrix, merge the two matrices to form a new cluster, delete the rows and columns related to the two clusters, and add the distance of the new cluster to all other remaining clusters, that is, the maximum distance between the new cluster and all node pairs in the other cluster; Record the merging information, namely the merged self-group and the merging distance, for use in constructing a tree diagram; Repeated distance calculation and cluster merging: specify that all nodes be merged into one cluster, or reach a preset number of clusters; The elbow rule is used to determine the optimal number of clusters. Based on the determined number of clusters k, the tree diagram is divided to obtain k clusters, each containing multiple resource nodes. For each cluster, compute the aggregate attributes, including: Total adjustable capacity: The sum of the rated capacities of all resources within the cluster; Average response time: a zero-weighted average of the response times of all resources, with the weights being capacity or reliability; Overall adjustment cost: a weighted average of the bids for each resource; Cluster boundary nodes: Identify key nodes that are connected to the external power grid (such as nodes that have edges with the outside). Internal electrical coupling strength of the cluster: can calculate the average edge weight within the cluster and measure the cluster's compactness; It should be noted that the purpose of generating electrically tightly coupled virtual resource clusters is to organize dispersed and massive adjustable resources into physically meaningful virtual clusters, providing reasonable basic units for subsequent adjustable capability assessment and coordinated control, and avoiding electrical mismatch problems caused by administrative or type-based aggregation. Regulation and optimization module: Within the virtual resource cluster, the weighted spatiotemporal correlation graph edge weights are used as constraints to construct a feasible domain for cluster power regulation. When scheduling instructions are issued, the regulation amount of each node is optimized and allocated under the constraints of the weighted spatiotemporal correlation graph. The process of constructing the feasible region for cluster power adjustment includes: For any pair of nodes within the cluster, the electrical distance is defined as the reciprocal of the graph edge weight. The electrical distance reflects the strength of electrical coupling; the smaller the distance, the stronger the coupling. Linearized branch power flow equations are used to approximate the power flow relationships within the cluster: Input: The set of nodes within the cluster, the electrical distance matrix between nodes, the active and reactive power adjustable range of each node, and the nodes that couple the cluster to the external power grid; Feasible region modeling: A linearized DistFlow model is used to approximate the power flow within the cluster, and the existence of branches and impedance range can be derived using the edges of the graph; Constructing the feasible region: Taking the total cluster regulation demand as a variable, we search for all possible total cluster regulation demands such that there exists a node regulation allocation scheme that satisfies node power balance, node regulation range, voltage constraints, and branch capacity constraints. The feasible region and the corresponding adjustment cost curve are obtained by vertex enumeration or sampling method; The process of optimizing the allocation of adjustment values ​​for each node under the constraints of a weighted spatiotemporal correlation graph includes: Once the scheduling center issues the cluster adjustment target, an optimization model is established with the goal of minimizing the weighted sum of adjustment cost and credibility penalty. The constraints of the optimization model include: the sum of the power regulation of each node is equal to the cluster regulation target, the power regulation of each node is within its adjustable range, and branch power constraints and node voltage constraints. Specifically, the branch power constraints include: Using the linearized DistFlow power flow model, the power variation of each branch is limited by its thermal stability limit, that is, the absolute value of the sum of the original power and the adjusted power of the branch must be less than or equal to the branch's thermal stability limit. It should be noted that the linearized DistFlow power flow model is a set of AC power flow equations specifically designed for radial distribution networks (i.e., tree structure, loop-free). It describes the relationship between branch power, node voltage, and injected power. While retaining key physical characteristics, it simplifies calculations through reasonable approximations and is widely used for problems such as distribution network optimization, voltage control, and adjustable resource scheduling. For example, a simple feeder: Node 1 (root node) - Node 2 - Node 3, with known parameters. If the load on Node 2 and Node 3 increases, the linearized DistFlow can quickly calculate the changes in power and voltage of each branch, thereby determining whether the limits are exceeded. In cluster optimization, after the scheduling center issues the overall adjustment target, the optimizer uses these linear constraints to allocate the adjustment amount of each node to ensure safety. Node voltage constraints include: the voltage amplitude of each node must be kept within a safe range; The optimal adjustment amount for each node is obtained by solving the optimization model using linear programming. It should be noted that the purpose of constructing the cluster regulation feasible region under the graph constraint and optimizing the allocation of regulation commands is to directly embed electrical coupling constraints into the regulation decision, upgrade from local optimization to global coordination, improve the success rate of command execution, reduce regulation costs, and ensure the safe and stable operation of the power grid.

[0020] The technical solution and advantages of this application are as follows: Correlation analysis is performed on each pair of adjustable resource nodes based on measurement data, and power transfer entropy is calculated. Initial edge weights are obtained through fusion, and candidate edges are screened. Graph attention network training is used to update the edge weights, resulting in a weighted spatiotemporal correlation graph. When missing measurement data is detected, neighboring nodes associated with the missing data node are searched in the weighted spatiotemporal correlation graph. Graph signal processing algorithms are used to imput the missing data, and the imputation confidence is output. The target area is divided into multiple grid areas, and the distribution of imputation confidence in each grid area is statistically analyzed to identify low-confidence areas, optimizing the collection priority of low-confidence areas. Based on the edge weights of the weighted spatiotemporal correlation graph, hierarchical clustering is used to aggregate nodes into virtual resource clusters, and the aggregation attributes of the virtual resource clusters are generated. Within the virtual resource clusters, the edge weights of the weighted spatiotemporal correlation graph are used as constraints to construct a feasible domain for cluster power adjustment. When scheduling instructions are issued, the adjustment amount of each node is optimized under the constraints of the weighted spatiotemporal correlation graph. This invention employs an edge acquisition terminal to synchronously and with high precision acquire electrical measurement data of adjustable resources. Candidate edges are generated by fusing the correlation and power transfer entropy of the measurement data, and a weighted spatiotemporal correlation graph is obtained through graph attention network training. The graph is used to perform Laplace interpolation on missing data and output confidence scores. Low-confidence regions are identified based on the spatial distribution of confidence scores, and the acquisition frequency and cleaning strategy are dynamically optimized. Hierarchical clustering based on graph edge weights generates a tightly coupled virtual resource cluster. Within the cluster, an adjustable feasible region is constructed using graph edge weights as constraints, and the allocation of control commands is optimized. This invention solves multiple coupling problems, including dynamic topology tracking, data quality enhancement, rationality of resource aggregation, and synergistic control.

[0021] The embodiments of the present invention have been described in detail above, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A system for constructing a multi-dimensional data platform based on adjustable resources of a measurement center, characterized by: Includes the following modules: Graph construction module: Based on measurement data, correlation analysis is performed on each pair of adjustable resource nodes, power transfer entropy is calculated, initial edge weights are obtained by fusion and candidate edges are screened, and graph attention network training is used to update the edge weights to obtain a weighted spatiotemporal correlation graph. The method of updating the weights of each edge using graph attention network training is as follows: Construct an initial graph using all adjustable resource nodes as vertices and candidate edges as edges; For each adjustable resource node, a feature vector is constructed, which includes resource type, rated capacity, historical average power, historical voltage fluctuation rate, and geographical location coordinates, forming a node feature matrix; A graph attention network is used for training. After training, all nodes are input into the graph attention network to obtain the connection probabilities of all node pairs. Edges whose connection probabilities meet the requirements are taken as the final topological edges, and the edge weights are updated to the attention coefficients output by the graph attention network to form a weighted spatiotemporal correlation graph. Data correction module: When missing measurement data is detected, the module searches for neighboring nodes associated with the missing data node in the weighted spatiotemporal correlation graph, performs missing data interpolation using graph signal processing algorithms, and outputs the interpolation confidence score. Acquisition optimization module: Divides the target area into multiple grid areas, and identifies low-confidence areas by statistically analyzing the distribution of interpolation confidence in each grid area, and optimizes the acquisition priority of low-confidence areas; Cluster aggregation module: Based on the weights of each edge of the weighted spatiotemporal correlation graph, hierarchical clustering is used to aggregate nodes into virtual resource clusters and generate the aggregation attributes of the virtual resource clusters; Regulation and optimization module: Within the virtual resource cluster, the weighted spatiotemporal correlation graph edge weights are used as constraints to construct a feasible domain for cluster power regulation. When scheduling instructions are issued, the regulation amount of each node is optimized and allocated under the constraints of the weighted spatiotemporal correlation graph. The feasible region for cluster power adjustment is constructed as follows: For any virtual resource cluster, the electrical distance between any pair of nodes within the cluster is defined as the reciprocal of the graph edge weight. The electrical distance reflects the strength of electrical coupling; the smaller the distance, the stronger the coupling. A linear DistFlow model is used to describe the power flow relationship within the cluster. Taking the total regulation demand of the cluster as a variable, the feasible region of cluster power regulation that satisfies node power balance, node regulation range, voltage constraints and branch capacity constraints is solved to obtain the feasible region interval and the corresponding regulation cost curve. The optimized allocation method for the adjustment amount of each node is as follows: Once the scheduling center issues the cluster adjustment target, an optimization model is established with the goal of minimizing the weighted sum of adjustment cost and credibility penalty. The constraints of the optimization model include: the sum of the power regulation of each node is equal to the cluster regulation target, the power regulation of each node is within its adjustable range, and branch power constraints and node voltage constraints. Among them, the branch power constraint includes: the power change of each branch is limited by its thermal stability limit, that is, the absolute value of the sum of the original power of the branch and the adjusted power must be less than or equal to the thermal stability limit of the branch. Node voltage constraints include: the voltage amplitude of each node must be kept within a safe range; The optimal adjustment amount for each node is obtained by solving the optimization model using linear programming.

2. The adjustable resource multi-dimensional data platform construction system based on a measurement center as described in claim 1, characterized in that: The process of obtaining initial edge weights through fusion and filtering candidate edges is as follows: Take the voltage amplitude time series of all adjustable resource nodes within the current time window, and calculate the Pearson correlation coefficient of voltage amplitude between any two adjustable resource nodes; Take the active power time series of all adjustable resource nodes within the current time window and calculate the power transfer entropy between any two adjustable resource nodes; The initial edge weights of each adjustable resource node pair are obtained by summing the normalized Pearson correlation coefficient and power transfer entropy. Adjustable resource node pairs whose initial edge weights meet the requirements are selected as candidate edges.

3. The adjustable resource multi-dimensional data platform construction system based on a measurement center according to claim 1, characterized in that: The method of using graph signal processing algorithms for missing data interpolation is as follows: For any node with missing measurement data at time t, find the top k neighbor nodes with the highest weight that have an edge with the node with missing measurement data in the weighted spatiotemporal correlation graph. Construct the Laplacian matrix L = D − A in graph signal processing, where A is the adjacency matrix of the weighted spatiotemporal correlation graph and D is the degree matrix; The interpolated value of the node with missing measurement data at time t is calculated by weighted average of the measurement values ​​of the first k neighboring nodes at time t, with the weights being the graph edge weights.

4. The adjustable resource multi-dimensional data platform construction system based on a measurement center as described in claim 3, characterized in that: The interpolation confidence level is output in the following manner: Confidence level is estimated based on the correlation strength and difference error of neighboring nodes: , where Q j A represents the quality index of the neighbor node data itself, set to 1 if there are no anomalies, where K is the number of neighbor nodes, and A is the quality index of the neighbor node data itself. ij Let N(i) be the edge weights between the missing node i and its neighboring nodes j, and let N(i) be the set of neighboring nodes.

5. The adjustable resource multi-dimensional data platform construction system based on a measurement center according to claim 1, characterized in that: The method for identifying the low-confidence region is as follows: The target area is divided into unit grid areas. The interpolation confidence distribution of all interpolation records in each unit grid area is statistically analyzed at fixed time windows, and the average confidence is calculated. Interpolation records that meet the interpolation confidence requirements are marked as low-confidence records. The proportion of low-confidence records to the total number of interpolation records in the grid area is calculated to obtain the proportion of low-confidence records. Grid regions that meet both the requirements for average confidence and the proportion of low confidence records are marked as normal regions; conversely, grid regions that do not meet the requirements for either average confidence or the proportion of low confidence records are marked as low confidence regions.

6. The adjustable resource multi-dimensional data platform construction system based on a measurement center as described in claim 5, characterized in that: The optimization process for prioritizing the collection of low-confidence regions is as follows: For grid regions marked as low-confidence areas, perform at least one of the following optimization measures: Increase the sampling frequency of adjustable resource access point terminals in low-confidence areas, shorten the data upload cycle, relax the anomaly detection threshold, adopt a conservative interpolation strategy, use only high-confidence nodes among neighbors for interpolation, and increase the number of neighbor nodes used during interpolation.

7. The adjustable resource multi-dimensional data platform construction system based on a measurement center according to claim 1, characterized in that: The aggregation method of the virtual resource cluster is as follows: All adjustable resource nodes are integrated into a node set, and the edge weights between each node pair in the weighted spatiotemporal correlation graph are constructed into a weighted adjacency matrix; The edge weights are converted into the distances between nodes using a linear transformation: distance between nodes = 1 - edge weights, thus linearly mapping the similarity to the [0,1] interval. Each node is initialized as an independent cluster. A hierarchical clustering algorithm based on the full-link criterion is used. In each iteration, the two clusters with the smallest distance are found and merged. The distance between the new cluster and other clusters is defined as the maximum distance between a node in the new cluster and all nodes in the other cluster. Repeat the merging process until all nodes are merged into a single cluster, construct a tree diagram and record the merging information; The elbow rule is used to determine the optimal number of clusters. Based on the determined number of clusters, the tree diagram is divided to obtain k virtual resource clusters, each containing multiple adjustable resource nodes.

Images