Multimodal data aggregation access and feature extraction method and system for power grid operation risk

Abstract

The present invention discloses a method and system for aggregating, accessing and extracting multimodal data of power grid operation risks, and relates to the technical field of power grid operation risk detection. The method comprises the following steps: obtaining a number of power grid data source nodes, labeling and constructing an interface set, and performing multimodal data aggregation; constructing a power grid operation risk prediction model based on a graph neural network according to historical risk characteristics and multi-source risk characteristics, and generating an operation risk value; dividing and cutting the global power grid operation process based on a topological structure to obtain a set of local processes; reconstructing the operation risk values ​​of the local processes based on a process-risk graph to obtain a reconstructed operation risk value, and performing risk maintenance on the local processes in sequence according to the size of the reconstructed operation risk value; the present application solves the problem that the prior art relies on a single data source and has no clear standard for the priority of risk processing.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid operation risk detection, and more specifically, to a method and system for aggregating, accessing and extracting multimodal data of power grid operation risks. Background Art

[0002] With the increasing complexity of power grid systems and the variability of the power system operating environment, traditional power grid risk management methods face problems such as low data processing efficiency and untimely risk warning.

[0003] For example, the invention patent announcement with announcement number: CN117674119A discloses a method, device, computer equipment and storage medium for assessing the risk of power grid operation, including obtaining power grid operation status data and meteorological monitoring data to generate a power grid section data set; in the offline stage of the power grid, constructing a generative adversarial network model based on the power grid section data set; obtaining at least one type of power grid operation risk data through the generative adversarial network model to generate a set of power grid operation risk scenarios; using a time series operation simulation method to deduce and calculate at least one power grid operation risk scenario to generate at least one power grid forward-looking scenario sequence; in the online operation stage of the power grid, obtaining the power grid operation risk assessment result based on the preset risk assessment index and the current power grid operation scenario information, and matching to obtain the corresponding power grid forward-looking operation scenario sequence. The use of this method can enhance the accuracy and timeliness of the power grid operation risk assessment.

[0004] For example, the invention patent announcement with announcement number: CN112531707A discloses a method for dynamically assessing the operating risks of a distribution network including a wind-storage combined system.

[0005] The above disclosed technical solutions have at least the following technical problems:

[0006] The existing technology relies on a single data source, which limits the accuracy and timeliness of risk identification. Especially in terms of multimodal data processing and feature extraction of power grid operation, the existing methods are difficult to meet the needs of safe and stable power grid operation, and there is no clear standard for risk processing priority. In view of the above problems, the present invention proposes a solution. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method and system for multimodal data aggregation, access and feature extraction of power grid operation risks. Through multimodal data aggregation, access and feature extraction, the problem that the prior art relies on a single data source and has no clear standard for risk processing priority is solved.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A method and system for aggregating, accessing and extracting features of multimodal data of power grid operation risks, comprising the following steps: obtaining a number of power grid data source nodes, labeling and constructing an interface set, and aggregating multimodal data; obtaining historical power grid operation multimodal data, and extracting features based on a self-learning feature extraction network to obtain historical risk features; interactively analyzing historical power grid operation multimodal data based on a graph neural network to obtain multi-source risk features; constructing a power grid operation risk prediction model based on a graph neural network based on historical risk features and multi-source risk features to generate an operation risk value; predicting global process risks based on the power grid operation risk prediction model, and when there is a global risk, dividing and cutting the global power grid operation process based on a topological structure to obtain a local process set; setting corresponding interface constraints based on the local process set, predicting local process risks based on a power grid operation risk prediction model combined with the interface constraints, and constructing a three-dimensional process-risk map of the generated operation risk values ​​and the corresponding local processes; reconstructing the operation risk values ​​of the local processes based on the process-risk map to obtain reconstructed operation risk values, and performing risk maintenance on the local processes in sequence according to the size of the reconstructed operation risk values.

[0010] In a preferred embodiment, the multimodal data aggregation is specifically: real-time data from each power grid data source node is collected and transmitted through a standardized interface protocol, and the data is aggregated and stored using a message queue; the power grid data source nodes include substation monitoring nodes, transmission line monitoring nodes, distribution network monitoring nodes, smart meter nodes, environmental monitoring nodes, and equipment status monitoring nodes.

[0011] In a preferred embodiment, the feature extraction based on the self-learning feature extraction network to obtain the historical risk characteristics is specifically as follows: a self-learning feature extraction network is constructed based on an autoencoder, and the input historical power grid operation multimodal data is first encoded to compress the high-dimensional multimodal data into a low-dimensional potential multimodal feature space; key features and implicit patterns in the historical data in the potential multimodal feature space are automatically captured through unsupervised learning; noise and redundant information are filtered; then, the low-dimensional potential multimodal feature space is restored to high-dimensional multimodal data using a decoder; and the historical risk characteristics are obtained by repeatedly converting high-dimensional to low-dimensional and then to high-dimensional data through comparison errors with the original high-dimensional multimodal data until a preset error range is met.

[0012] In a preferred embodiment, the multi-modal data of historical power grid operation is interactively analyzed based on the graph neural network to obtain multi-source risk characteristics, specifically: the multi-modal data nodes of historical power grid operation are represented as a graph structure, and each feature representing a different data source and power grid operation status is regarded as a graph node; a graph node set and an edge set are obtained according to the graph structure; the multi-source data corresponding to each graph node is calculated and updated based on the characteristics of neighbor nodes and attention weights based on a preset multi-source data relationship propagation formula; and a high-order feature representation of each node is generated based on the propagation and aggregation of the multi-layer graph neural network according to the updated graph structure to obtain multi-source risk characteristics.

[0013] In a preferred embodiment, the power grid operation risk prediction model is constructed based on the graph neural network according to the historical risk characteristics and multi-source risk characteristics to generate an operation risk value, specifically: the historical risk characteristics and multi-source risk characteristics are combined to form a model input feature matrix, and the model input feature matrix is ​​input into a preset multimodal graph convolution layer to obtain a power grid operation node feature matrix; the obtained power grid operation node feature matrix is ​​calculated with the pre-acquired historical risk feature matrix. When the output result is greater than the preset constraint threshold, the power grid operation node feature matrix is ​​input into the preset multimodal graph convolution layer again; the power grid operation node feature matrix is ​​input into the preset power grid operation risk fully connected layer to output the operation risk value.

[0014] In a preferred embodiment, the global power grid operation process is divided and cut based on the operation process topology structure to obtain a local process set, specifically: the power grid topology map is traversed according to the depth-first search, starting from any node in the power grid topology map, it is marked as visited and added to the traversal queue; all adjacent nodes of the arbitrarily selected node are checked, and the connection relationship between the power grid nodes is represented by the adjacency matrix and the adjacency table; each unvisited adjacent node is marked as visited and added to the traversal queue; during the traversal process, the connection information between each node and its adjacent nodes is recorded to form a node connection list; after completing the traversal of the entire power grid operation process topology structure, according to the node connection list, several groups of nodes with direct connection relationships are identified, and the cutting criteria are set according to the transmission delay between the power grid nodes, and the global power grid operation process is divided into several independent local process sets.

[0015] In a preferred embodiment, the specific method for obtaining the interface constraint condition is as follows: the interface constraint condition includes a first constraint condition, a second constraint condition and a third constraint condition; based on all the nodes covering the interface set in the local process, a new interface set is taken, and the new interface set is taken as the first constraint condition; the interface data is filtered through a preset multi-source data threshold, and redundant interfaces are eliminated, and the eliminated interfaces are taken as the second constraint condition; the remaining interfaces are clustered using a clustering algorithm to identify interface groups with similar data flows; the number of similar interfaces is reduced by selecting the central node of each group of interfaces, and the reduced interfaces are taken as the third constraint condition.

[0016] In a preferred embodiment, the local process risk prediction is performed based on the power grid operation risk prediction model in combination with the interface constraint conditions, and a three-dimensional process-risk diagram is constructed by combining the generated operation risk values ​​with the corresponding local processes. Specifically, the interface set is constrained according to the constraint conditions to obtain several groups of different interface sets corresponding to the local operation processes; the local process risk prediction is performed according to the interface set corresponding to the local operation process to generate several groups of operation risk values ​​corresponding to the local operation process; the topological structure of the local operation process is used as the X-axis and Y-axis; the operation risk value corresponding to each local operation process is used as the Z-axis; and the data points corresponding to the X, Y, and Z axes are connected to draw a three-dimensional process-risk diagram.

[0017] In a preferred embodiment, the operation risk value of the local process is reconstructed based on the process-risk map for priority, specifically: N times the Euclidean distance of the process nodes adjacent to the local operation process node to be reconstructed on their respective process-risk maps is used as the reconstruction weight; the product of the operation risk values ​​of all adjacent process nodes of the local operation process node to be reconstructed and the corresponding reconstruction weight is added, and combined with the operation risk value of the local operation process node to be reconstructed itself, to obtain the reconstructed operation risk value; the reconstructed operation risk values ​​of all local processes are sorted in descending order, and risk maintenance is performed in sequence from large to small.

[0018] Technical effects and advantages of the power grid operation risk multimodal data aggregation access and feature extraction method and system of the present invention:

[0019] 1. The present invention significantly improves the accuracy and efficiency of power grid operation risk prediction through the convergence and feature extraction of multiple power grid data source nodes. The method first integrates time series and non-time series data from multiple monitoring nodes such as substations, transmission lines, and smart meters to ensure the comprehensiveness of the information, and then uses self-learning feature extraction networks and graph neural networks to deeply analyze historical data and multi-source data features to automatically capture complex risk patterns. This dynamic adaptability and efficient data processing capabilities not only improve the accuracy of the model, but also respond to power grid changes in real time and identify potential safety hazards in a timely manner. In addition, through the interactive analysis of graph neural networks, the capture of the relationship between nodes is further strengthened, and more accurate risk assessment and prediction are achieved, which ultimately provides decision makers with intuitive risk value display and corresponding decision support to ensure the safe and stable operation of the power grid.

[0020] 2. The present invention conducts an in-depth analysis of the topological structure of the power grid, and the method can accurately identify global risks, thereby realizing the division and cutting of the power grid operation process and generating a set of local processes. When the global operation risk value exceeds the set threshold, the system uses depth-first search to traverse the power grid topology map, record the connection information between nodes, and effectively divide multiple independent local processes. This method makes the understanding of the connection relationship between each node in the power grid (such as power plants, substations, transmission lines, etc.) clearer, thereby laying a good foundation for subsequent risk assessment; in the local process risk prediction, this method combines interface constraints, through multi-source data threshold filtering, cluster analysis and other means, effectively reducing redundant interfaces and improving the accuracy of risk prediction. Finally, the generated three-dimensional process-risk map clearly shows the operation risk value of each local process, allowing power grid managers to intuitively understand the risk situation of different local processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a structural schematic diagram of the multimodal data aggregation access and feature extraction method of power grid operation risk of the present invention.

[0022] Figure 2 This is a schematic diagram of the structure of the multimodal data aggregation access and feature extraction system for power grid operation risks of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Embodiment 1, Figure 1The present invention provides a method for aggregating, accessing and extracting multimodal data of power grid operation risks, comprising the following steps:

[0025] S1, obtain several power grid data source nodes, label and construct interface sets, and perform multimodal data aggregation.

[0026] In this implementation, the power grid data source nodes include but are not limited to substation monitoring nodes, transmission line monitoring nodes, distribution network monitoring nodes, smart meter nodes, environmental monitoring nodes, and equipment status monitoring nodes.

[0027] In this implementation, the multimodal data aggregation is specifically:

[0028] The real-time data of each power grid data source node is collected and transmitted through a standardized interface protocol, and the message queue is used to aggregate and store the data.

[0029] It should be noted that the data types collected by various data source nodes in the power grid are different, including both time series data (such as voltage, current and other operating status data) and non-time series data (such as equipment status, environmental conditions, etc.); multimodal data aggregation is to collect these different types of data through a unified interface to ensure that multiple dimensions of information can be taken into account simultaneously in subsequent feature extraction and analysis, thereby providing data support for the global and local process prediction of power grid operation risks.

[0030] S2, obtain historical multimodal data of power grid operation, and extract features based on the self-learning feature extraction network to obtain historical risk characteristics; interactively analyze the historical multimodal data of power grid operation based on the graph neural network to obtain multi-source risk characteristics.

[0031] In this embodiment, the feature extraction is performed based on the self-learning feature extraction network to obtain historical risk features, specifically:

[0032] A self-learning feature extraction network is constructed based on an autoencoder. First, the input historical power grid operation multimodal data is encoded to compress the high-dimensional multimodal data into a low-dimensional potential multimodal feature space.

[0033] Automatically capture key features and hidden patterns in historical data in the latent multimodal feature space through unsupervised learning;

[0034] Filter out noise and redundant information; then, use the decoder to restore the low-dimensional latent multimodal feature space to high-dimensional multimodal data;

[0035] The historical risk characteristics are obtained by repeatedly converting high-dimensional to low-dimensional and then to high-dimensional data through comparison errors with the original high-dimensional multimodal data until the preset error range is met.

[0036] It should be noted that historical risk features are based on multimodal historical data of power grid operation. They are key information extracted by the self-learning feature extraction network, reflecting the operation status and potential risks of the power grid in different time periods. These features include important factors related to the safety of power grid operation, such as equipment failure, load fluctuation, and external environment changes, and can provide accurate data support for subsequent risk prediction and assessment.

[0037] Analyzing historical risk characteristics through self-learning feature extraction network has the following advantages for predicting power grid operation risks:

[0038] Automatic learning capability: The self-learning feature extraction network can automatically extract key features from large-scale multimodal power grid data, reducing manual intervention and deviation, and ensuring the comprehensiveness and accuracy of feature extraction;

[0039] Capturing complex patterns: The network can identify complex associations and potential patterns in grid operation data, especially nonlinear changes in historical risk characteristics, which helps to better predict potential risks;

[0040] Dynamic adaptability: The self-learning feature extraction network can continuously optimize and adjust its extraction mechanism according to the changes in historical data, ensuring the dynamic and real-time nature of power grid risk characteristics, making risk prediction more timely;

[0041] Improve prediction accuracy: The features extracted by the self-learning network have higher discrimination ability, which can enhance the sensitivity of the power grid risk prediction model to future risks, thereby improving the accuracy and reliability of the prediction;

[0042] Cross-modal data processing: It can integrate multi-modal information such as historical operation data of the power grid, external environment, and equipment status into a unified feature space to improve the assessment effect of the overall risk of the power grid;

[0043] This makes the grid operation risk prediction model more intelligent and efficient, and can provide more reliable data support for global and local risk assessment.

[0044] In this embodiment, the multi-source risk characteristics are obtained by interactively analyzing the historical multi-modal data of power grid operation based on the graph neural network, specifically:

[0045] The multimodal data nodes of historical power grid operation are represented as a graph structure, and each feature representing a different data source and power grid operation status is regarded as a graph node; the graph node set and edge set are obtained according to the graph structure;

[0046] The multi-source data corresponding to each graph node is calculated and updated based on the characteristics and attention weights of neighboring nodes based on the preset multi-source data relationship propagation formula;

[0047] According to the propagation and aggregation of the updated graph structure based on the multi-layer graph neural network, the high-order feature representation of each node is generated to obtain the multi-source risk characteristics.

[0048] It should be noted that there are mature technologies for generating high-order feature representations of each node through propagation and aggregation of multi-layer graph neural networks, which will not be described in detail in this application.

[0049] Multi-source risk characteristics are comprehensive characteristics obtained through interactive analysis of multimodal data of historical power grid operation through graph neural networks; these characteristics reflect the correlation and mutual influence between different data sources, and can capture potential risk patterns and abnormal behaviors; by dynamically adjusting the attention weights of different nodes, multi-source risk characteristics not only reveal the changes in various power grid operation indicators, but also identify potential risk factors and their interactions, providing an important basis for the accurate prediction and management of power grid operation risks; this integration of multi-source characteristics enhances the model's understanding of complex power grid systems, thereby improving the accuracy and reliability of risk prediction.

[0050] Analyzing multi-source risk characteristics through graph neural networks has the following advantages for predicting power grid operation risks:

[0051] Analyzing multi-source risk characteristics through graph neural networks has the following advantages for predicting power grid operation risks: First, graph neural networks can effectively capture the complex relationships between nodes in the power grid, and by dynamically adjusting the attention weights between nodes, it enhances the focus on important features, thereby improving the accuracy of feature extraction; second, the network can process heterogeneous data and integrate multimodal information from different sources to ensure that the risk prediction model takes into account more relevant factors, thereby improving the comprehensiveness and accuracy of the model; in addition, graph neural networks have good scalability and can adapt to the ever-changing power grid structure and dynamic operating environment, so that risk prediction can respond to emerging risk situations in real time to ensure the safe and stable operation of the power grid; finally, the analysis results based on graph neural networks can provide more accurate data support for power grid operation decisions and promote the formulation of effective risk management and control strategies.

[0052] S3, builds a power grid operation risk prediction model based on graph neural network according to historical risk characteristics and multi-source risk characteristics, and generates operation risk value.

[0053] In this embodiment, the construction of the power grid operation risk prediction model is specifically as follows:

[0054] The historical risk characteristics and multi-source risk characteristics are combined to form a model input feature matrix, and the model input feature matrix is ​​input into a preset multimodal graph convolution layer to obtain a grid operation node feature matrix;

[0055] The obtained grid operation node feature matrix is ​​used to calculate the mean square error with the previously acquired historical risk feature matrix. When the output result is greater than the preset constraint threshold, the grid operation node feature matrix is ​​input into the preset multimodal graph convolution layer again.

[0056] The grid operation node feature matrix is ​​input into the preset grid operation risk fully connected layer, and the operation risk value is output.

[0057] The operation risk value is a quantitative indicator obtained through deep learning processing of the grid operation node feature matrix, which aims to reflect the risk level of the grid under a specific operating state. It comprehensively considers historical risk characteristics and multi-source risk characteristics, and indicates the potential risks that the grid system may face under specific conditions through the output of the fully connected layer of the model. A higher operation risk value may indicate that the system is facing failure, overload or other safety hazards, and corresponding preventive measures or intervention measures need to be taken in time to ensure the stable and safe operation of the grid.

[0058] The calculation formula of the multimodal graph convolution layer is as follows:

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] The calculation formula of the fully connected layer of power grid operation risk is as follows:

[0065]

[0066] In the formula, is the introduced coefficient of the input matrix, is the sigmoid activation function, is the weight matrix of the multimodal graph convolutional layer, is the risk value of the t-1 operation, is the third weighted data, is the retention coefficient of the memory cell state of the previous iteration, is the weight matrix of the forget gate, is the candidate memory cell state, is the tanh activation function, is the weight matrix for selecting the memory cell state, is the memory cell state at t iterations, The state of the memory cell at the t-1 iteration is the state of the memory cell at the previous iteration. is the predicted influence coefficient of the memory cell state on the predicted operation risk value, is the weight matrix of the output gate, is the running risk value of t iterations.

[0067] The above-mentioned multi-modal data aggregation access and feature extraction risk prediction method for power grid operation risk has the following advantages:

[0068] Comprehensive data utilization: By gathering power grid data from multiple sources, the comprehensiveness of information is ensured, which can fully reflect the power grid operation status and potential risks, and enhance the accuracy of prediction;

[0069] Deep feature extraction: Based on the self-learning feature extraction network and graph neural network method, it can effectively mine the complex features in the data, thereby improving the ability to identify risk features and helping to build a more accurate prediction model;

[0070] Dynamic adaptability: Through the self-learning mechanism, the model can automatically adjust the feature extraction method according to the newly acquired data, which enhances the model's adaptability to power grid changes and enables risk prediction to be updated at any time;

[0071] Efficient risk assessment: Using graph neural networks for interactive analysis can accurately capture the relationship and impact between nodes in the power grid, thereby achieving more efficient risk assessment and improving the overall risk management level;

[0072] Accurate risk prediction: By combining historical risk characteristics and multi-source risk characteristics, the grid operation risk prediction model can provide more accurate operation risk values, which helps to timely identify and prevent potential safety hazards;

[0073] Visualization and decision support: The generated operational risk values ​​can be displayed graphically, allowing decision makers to quickly understand the current risk status of the power grid and formulate corresponding response strategies.

[0074] This embodiment significantly improves the accuracy and efficiency of grid operation risk prediction by aggregating and extracting features from multiple grid data source nodes. The method first integrates time series and non-time series data from multiple monitoring nodes such as substations, transmission lines, and smart meters to ensure the comprehensiveness of the information, and then uses self-learning feature extraction networks and graph neural networks to deeply analyze historical data and multi-source data features to automatically capture complex risk patterns. This dynamic adaptability and efficient data processing capability not only improves the accuracy of the model, but also responds to grid changes in real time and identifies potential safety hazards in a timely manner. In addition, through the interactive analysis of graph neural networks, the capture of the relationship between nodes is further strengthened, achieving more accurate risk assessment and prediction, and ultimately providing decision makers with intuitive risk value display and corresponding decision support to ensure the safety and stable operation of the power grid.

[0075] Embodiment 2, S4, performs global process risk prediction based on the power grid operation risk prediction model. When there is a global risk, the global power grid operation process is divided and cut based on the topological structure to obtain a local process set.

[0076] As described above, the global process risk prediction is performed according to the power grid operation risk prediction model. When there is a global risk, specifically:

[0077] The operation risk value generated by the global process risk prediction of the power grid operation risk prediction model is compared with the preset global risk threshold. When the operation risk value is greater than the global risk threshold, there is a global risk.

[0078] The global power grid operation process is divided and cut based on the operation process topology structure to obtain a local process set, which is specifically:

[0079] Traverse the power grid topology map according to the depth-first search, starting from any node in the power grid topology map, mark it as visited and add it to the traversal queue;

[0080] All adjacent nodes of any selected node are checked, and the connection relationship between power grid nodes is represented by adjacency matrix and adjacency table; each unvisited adjacent node is marked as visited and added to the traversal queue;

[0081] During the traversal process, the connection information between each node and its adjacent nodes is recorded to form a node connection list;

[0082] After completing the traversal of the topological structure of the entire power grid operation process, several groups of nodes with direct connection relationships are identified according to the node connection list. The cutting criteria are set according to the transmission delay between the power grid nodes, and the global power grid operation process is divided into several independent local process sets.

[0083] It should be noted that the operation process topology refers to the overall layout and network organization of each node (such as power plants, substations, transmission lines, etc.) and their interconnected relationships in the operation of the power grid; through the topology, the connection mode and interaction between the various components in the power grid can be clearly displayed, thereby reflecting the logical relationship and information flow path of the power grid operation; when predicting the global process risk, analyzing the topology of the operation process helps to identify potential risk sources and key nodes, and then divide and cut the overall process of the power grid into multiple local processes, so as to more accurately assess and manage risks. This structured approach not only improves the efficiency of risk identification, but also provides a scientific basis for subsequent local process optimization and risk control.

[0084] S5, set the corresponding interface constraints based on the local process set, perform local process risk prediction based on the power grid operation risk prediction model combined with the interface constraints, and construct a three-dimensional process-risk map with the generated operation risk value and the corresponding local process.

[0085] The specific method for obtaining the interface constraint conditions is as follows:

[0086] The interface constraint conditions include a first constraint condition, a second constraint condition, and a third constraint condition.

[0087] All nodes covering the interface set in the local process are used as a new interface set, and the new interface set is used as the first constraint condition;

[0088] The interface data is filtered through a preset multi-source data threshold, redundant interfaces are eliminated, and the eliminated interfaces are used as the second constraint condition;

[0089] A clustering algorithm is used to cluster the remaining interfaces to identify interface groups with similar data flows; the number of similar interfaces is reduced by selecting the central node of each group of interfaces, and the reduced interfaces are used as the third constraint condition.

[0090] It should be noted that the specific method for obtaining the third constraint condition is as follows:

[0091] Quantify the data flow characteristics of each interface to form a feature vector; the feature vector may include information such as the transmission frequency, data size, data type, and delay of the interface;

[0092] Clustering feature vectors based on hierarchical clustering;

[0093] Randomly select the same number of initial center points as the process;

[0094] Assign the feature vector of each interface to the nearest center point to form several clusters;

[0095] Update the center point of each cluster to the mean of all interface feature vectors in the cluster;

[0096] Repeat the assignment and updating process until the center point no longer changes significantly;

[0097] After clustering is completed, the interfaces contained in each cluster are identified and compared with the original interfaces to obtain the reduced interfaces as the third constraint.

[0098] The specific calculation formula of the hierarchical clustering is as follows:

[0099]

[0100]

[0101]

[0102]

[0103]

[0104] In the formula, is the silhouette coefficient, represents the average distance between a sample and other samples in the same cluster. Represents the average distance between a sample and all samples in the nearest other clusters; is the feature vector of sample j; is the objective function (intra-cluster sum of squares), k is the number of clusters, is the i-th cluster, represents the center of the i-th cluster, is the cluster, ||*|| is the Euclidean distance, Represents the total number of samples in cluster i.

[0105] The local process risk prediction is performed based on the power grid operation risk prediction model combined with the interface constraint conditions, and the generated operation risk value and the corresponding local process are used to construct a three-dimensional process-risk diagram, specifically:

[0106] Constrain the interface set according to the constraint conditions to obtain several groups of different interface sets corresponding to the local running processes;

[0107] Perform local process risk prediction based on the interface set corresponding to the local operation process, and generate several groups of operation risk values ​​corresponding to the local operation process;

[0108] The topology of the local running process is used as the X-axis and Y-axis;

[0109] The operation risk value corresponding to each local operation process is used as the Z axis;

[0110] A three-dimensional process-risk diagram is drawn by connecting the data points corresponding to the X, Y, and Z axes.

[0111] S6, reconstructing the operation risk value of the local process based on the process-risk map to obtain the reconstructed operation risk value, and performing risk maintenance on the local process in sequence according to the size of the reconstructed operation risk value.

[0112] The operation risk value of the local process is reconstructed based on the process-risk map, specifically:

[0113] The reconstruction weight is N times the Euclidean distance between the process nodes adjacent to the local running process node to be reconstructed and the local running process node to be reconstructed on their respective process-risk graphs;

[0114] The product of the operation risk values ​​of all adjacent process nodes of the local operation process node to be reconstructed and the corresponding reconstruction weight is added, and combined with the operation risk value of the local operation process node to be reconstructed itself, to obtain the reconstructed operation risk value;

[0115] Sort the reconstructed operational risk values ​​of all local processes in descending order, and perform risk maintenance from large to small.

[0116] This embodiment uses an in-depth analysis of the power grid topology structure, and the method can accurately identify global risks, thereby realizing the division and cutting of the power grid operation process and generating a set of local processes. When the global operation risk value exceeds the set threshold, the system uses depth-first search to traverse the power grid topology map, record the connection information between nodes, and effectively divide multiple independent local processes. This method makes the understanding of the connection relationship between each node in the power grid (such as power plants, substations, transmission lines, etc.) clearer, thereby laying a good foundation for subsequent risk assessment; in the local process risk prediction, this method combines interface constraints, through multi-source data threshold filtering, cluster analysis and other means, effectively reducing redundant interfaces and improving the accuracy of risk prediction. Finally, the generated three-dimensional process-risk map clearly shows the operation risk value of each local process, allowing power grid managers to intuitively understand the risk situation of different local processes.

[0117] In summary, this embodiment significantly improves the efficiency of grid risk identification, assessment and management through structured risk management strategies, and provides strong support for achieving safe and stable grid operation.

[0118] Embodiment 3, Figure 2 The invention provides a multimodal data aggregation access and feature extraction system for power grid operation risk, including a multimodal data aggregation module, a feature extraction module, a risk prediction and analysis module, a local process division module, a local risk prediction module and a maintenance module, and there are connections between the modules;

[0119] The multimodal data aggregation module is used to obtain several power grid data source nodes, label and construct interface sets, and perform multimodal data aggregation;

[0120] The feature extraction module is used to obtain historical multimodal data of power grid operation and extract features based on the self-learning feature extraction network to obtain historical risk features; interactively analyze the historical multimodal data of power grid operation based on the graph neural network to obtain multi-source risk features;

[0121] The risk prediction and analysis module is used to build a power grid operation risk prediction model based on the graph neural network according to historical risk characteristics and multi-source risk characteristics, and generate an operation risk value;

[0122] The local process division module is used to predict the global process risk according to the power grid operation risk prediction model. When there is a global risk, the global power grid operation process is divided and cut based on the topological structure to obtain a local process set.

[0123] The local risk prediction module is used to set the corresponding interface constraints based on the local process set, perform local process risk prediction based on the power grid operation risk prediction model combined with the interface constraints, and construct a three-dimensional process-risk map with the generated operation risk value and the corresponding local process;

[0124] The maintenance module is used to reconstruct the operation risk value of the local process based on the process-risk map to obtain the reconstructed operation risk value, and perform risk maintenance on the local process in sequence according to the size of the reconstructed operation risk value.

[0125] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0126] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0127] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0128] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0129] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0130] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for aggregating, accessing and extracting multimodal data of power grid operation risks, characterized in that: The steps include: Obtain several power grid data source nodes, label and construct interface sets, and perform multi-modal data aggregation; Obtain historical multimodal data of power grid operation, and extract features based on a self-learning feature extraction network to obtain historical risk features; Based on graph neural network, interactive analysis of historical power grid operation multimodal data is performed to obtain multi-source risk characteristics; According to historical risk characteristics and multi-source risk characteristics, a grid operation risk prediction model is built based on graph neural network to generate operation risk value; The global process risk prediction is carried out according to the power grid operation risk prediction model. When there is a global risk, the global power grid operation process is divided and cut based on the topological structure to obtain a local process set. Based on the local process set, the corresponding interface constraints are set, and the local process risk prediction is performed based on the power grid operation risk prediction model combined with the interface constraints. The generated operation risk value and the corresponding local process are used to construct a three-dimensional process-risk diagram; Reconstruct the operation risk value of the local process based on the process-risk map to obtain the reconstructed operation risk value, and perform risk maintenance on the local process in sequence according to the size of the reconstructed operation risk value; The local process risk prediction is performed based on the power grid operation risk prediction model combined with the interface constraint conditions, and the generated operation risk value and the corresponding local process are used to construct a three-dimensional process-risk diagram, specifically: Constrain the interface set according to the constraint conditions to obtain several groups of different interface sets corresponding to the local running processes; Perform local process risk prediction based on the interface set corresponding to the local operation process, and generate several groups of operation risk values ​​corresponding to the local operation process; The topology of the local running process is used as the X-axis and Y-axis; The operation risk value corresponding to each local operation process is used as the Z axis; Draw a three-dimensional process-risk diagram by connecting the data points corresponding to the X, Y, and Z axes; The operation risk value of the local process is reconstructed based on the process-risk map, specifically: The reconstruction weight is N times the Euclidean distance between the process nodes adjacent to the local running process node to be reconstructed and the local running process node to be reconstructed on their respective process-risk graphs; The product of the operation risk values ​​of all adjacent process nodes of the local operation process node to be reconstructed and the corresponding reconstruction weight is added, and combined with the operation risk value of the local operation process node to be reconstructed itself, to obtain the reconstructed operation risk value; Sort the reconstructed operational risk values ​​of all local processes in descending order, and perform risk maintenance from large to small.

2. The method for aggregating, accessing and extracting multimodal data of power grid operation risks according to claim 1 is characterized in that: The multimodal data aggregation is specifically as follows: Collect real-time data from each power grid data source node and transmit it through a standardized interface protocol, and use a message queue to aggregate and store the data; The power grid data source nodes include substation monitoring nodes, transmission line monitoring nodes, distribution network monitoring nodes, smart meter nodes, environmental monitoring nodes, and equipment status monitoring nodes.

3. The method for aggregating, accessing and extracting multimodal data of power grid operation risks according to claim 2 is characterized in that: The feature extraction based on the self-learning feature extraction network is performed to obtain historical risk features, which are specifically: A self-learning feature extraction network is constructed based on an autoencoder. First, the input historical power grid operation multimodal data is encoded to compress the high-dimensional multimodal data into a low-dimensional potential multimodal feature space. Automatically capture key features and hidden patterns in historical data in the latent multimodal feature space through unsupervised learning; Filter out noise and redundant information; then, use the decoder to restore the low-dimensional latent multimodal feature space to high-dimensional multimodal data; The historical risk characteristics are obtained by repeatedly converting high-dimensional to low-dimensional and then to high-dimensional data through comparison errors with the original high-dimensional multimodal data until the preset error range is met.

4. The method for gathering, accessing and extracting multimodal data of power grid operation risk according to claim 3 is characterized in that: The multi-source risk characteristics are obtained by interactively analyzing the multi-modal data of historical power grid operation based on the graph neural network, which are specifically: The multimodal data nodes of the historical power grid operation are represented as a graph structure, and each feature representing a different data source and power grid operation status is regarded as a graph node; the graph node set and edge set are obtained according to the graph structure; The multi-source data corresponding to each graph node is calculated and updated based on the characteristics and attention weights of neighboring nodes based on the preset multi-source data relationship propagation formula; According to the propagation and aggregation of the updated graph structure based on the multi-layer graph neural network, the high-order feature representation of each node is generated to obtain the multi-source risk characteristics.

5. The method for aggregating, accessing and extracting multimodal data of power grid operation risks according to claim 4 is characterized in that: The grid operation risk prediction model is constructed based on the graph neural network according to the historical risk characteristics and multi-source risk characteristics to generate the operation risk value, specifically: The historical risk characteristics and multi-source risk characteristics are combined to form a model input feature matrix, and the model input feature matrix is ​​input into a preset multimodal graph convolution layer to obtain a grid operation node feature matrix; The obtained grid operation node feature matrix is ​​used to calculate the mean square error with the previously acquired historical risk feature matrix. When the output result is greater than the preset constraint threshold, the grid operation node feature matrix is ​​input into the preset multimodal graph convolution layer again. The grid operation node feature matrix is ​​input into the preset grid operation risk fully connected layer, and the operation risk value is output.

6. The method for aggregating, accessing and extracting multimodal data of power grid operation risks according to claim 5 is characterized in that: The global power grid operation process is divided and cut based on the topological structure to obtain a local process set, which is specifically: Traverse the power grid topology map according to the depth-first search, starting from any node in the power grid topology map, mark it as visited and add it to the traversal queue; All adjacent nodes of any selected node are checked, and the connection relationship between power grid nodes is represented by adjacency matrix and adjacency table; each unvisited adjacent node is marked as visited and added to the traversal queue; During the traversal process, the connection information between each node and its adjacent nodes is recorded to form a node connection list; After completing the traversal of the topological structure of the entire power grid operation process, several groups of nodes with direct connection relationships are identified according to the node connection list. The cutting criteria are set according to the transmission delay between the power grid nodes, and the global power grid operation process is divided into several independent local process sets.

7. The method for gathering, accessing and extracting multimodal data of power grid operation risks according to claim 6 is characterized in that: The specific method for obtaining the interface constraint conditions is as follows: The interface constraint conditions include a first constraint condition, a second constraint condition, and a third constraint condition; All nodes covering the interface set in the local process are used as a new interface set, and the new interface set is used as the first constraint condition; The interface data is filtered through a preset multi-source data threshold, redundant interfaces are eliminated, and the eliminated interfaces are used as the second constraint condition; Clustering algorithms are used to cluster the remaining interfaces to identify groups of interfaces with similar data flows; By selecting the central node of each group of interfaces, the number of similar interfaces is reduced, and the reduced interfaces are used as the third constraint condition.

8. A system using the method for aggregating, accessing and extracting multimodal data of power grid operation risks as claimed in any one of claims 1 to 7, characterized in that: It includes a multimodal data aggregation module, a feature extraction module, a risk prediction and analysis module, a local process division module, a local risk prediction module, and a maintenance module, and there are connections between the modules; The multimodal data aggregation module is used to obtain several power grid data source nodes, label and construct interface sets, and perform multimodal data aggregation; The feature extraction module is used to obtain historical multimodal data of power grid operation and extract features based on a self-learning feature extraction network to obtain historical risk features; Based on graph neural network, interactive analysis of historical power grid operation multimodal data is performed to obtain multi-source risk characteristics; The risk prediction and analysis module is used to build a power grid operation risk prediction model based on the graph neural network according to historical risk characteristics and multi-source risk characteristics, and generate an operation risk value; The local process division module is used to predict the global process risk according to the power grid operation risk prediction model. When there is a global risk, the global power grid operation process is divided and cut based on the topological structure to obtain a local process set. The local risk prediction module is used to set the corresponding interface constraints based on the local process set, perform local process risk prediction based on the power grid operation risk prediction model combined with the interface constraints, and construct a three-dimensional process-risk map with the generated operation risk value and the corresponding local process; The maintenance module is used to reconstruct the operation risk value of the local process based on the process-risk map to obtain the reconstructed operation risk value, and perform risk maintenance on the local process in sequence according to the size of the reconstructed operation risk value.

Images