Power grid analysis system and analysis method based on topological graph

Abstract

The invention discloses a power grid analysis system and analysis method based on a topological graph, and the method comprises the steps: collecting power grid equipment information and topological connection relation data, and constructing an equipment state library and a power grid topological model; obtaining a plurality of node importance degrees and a plurality of edge connection closeness degrees according to the power grid topology model, and partitioning the power grid by using a multi-scale strongly connected region finite division method to obtain a plurality of power grid partitions; obtaining a comprehensive risk index of the plurality of power grid partitions according to the plurality of node importance degrees and the plurality of edge connection closeness degrees, and screening out the power grid partitions exceeding a preset risk threshold to obtain risk partitions; and collecting operation data of the risk partition, and inputting the operation data into a fault prediction model constructed based on deep learning to obtain fault time and fault equipment predicted by the risk partition. The invention relates to the technical field of power grid risk analysis, and solves the technical problems of low analysis efficiency and insufficient pertinence in an existing power grid analysis method.

Description

Technical Field

[0001] The invention belongs to the field of power grid analysis and relates to risk assessment technology, in particular to a power grid analysis system and analysis method based on a topology diagram. Background Art

[0002] As the energy transmission and distribution system of modern society, the power grid is responsible for safely and efficiently delivering the electricity produced by power plants to industrial, commercial and residential users. Its stable and reliable operation is the basis for ensuring the smooth progress of social and economic activities and maintaining the order of people's daily lives. Once a power grid fails, it may not only cause large-scale power outages, but also have a direct impact on industrial production and commercial operations, and even affect the basic living needs of residents, thereby bringing non-negligible economic losses and social impacts.

[0003] As an important means to improve the efficiency of power grid operation and prevent potential risks, power grid analysis aims to provide scientific basis and decision support for power grid planning and layout, daily operation and maintenance, fault handling and optimization and upgrading through in-depth research on the operation characteristics, performance and possible safety hazards of the power grid. Power grid topology data is one of the core elements of power grid analysis. It accurately describes the physical connection relationship between various devices in the power grid in the form of nodes and edges, and constructs a clear and intuitive power grid network model. In the process of risk assessment, through in-depth analysis of the power grid topology, key nodes and key lines that have an important impact on the overall stability of the power grid can be accurately identified. At the same time, combined with electrical quantity information, the operating status of the power grid can be further judged, the fault point can be quickly located, and a strong guarantee can be provided for fault handling and safe operation of the power grid.

[0004] However, current power grid analysis methods often focus on the overall analysis of power grid operation data. The invention patent with publication number CN117895661B discloses "a distribution network control method and system combined with risk analysis". The system collects distribution network operation data in real time, performs preprocessing and risk assessment (including equipment aging, power grid topology, and external environment analysis), generates control instructions based on risk warnings, sets operation and management cycles for decision analysis, evaluates operation efficiency and timeliness of operation and maintenance, and finally displays risk warnings and operation and management signals through a visual warning terminal; the invention patent with publication number CN117094475A discloses "a power distribution network fault analysis system". The system obtains designated area information and distribution network history and real-time data, calculates future fault frequency and probability weight, analyzes the impact of meteorological factors on faults, evaluates current and future operation safety factors, and finally comprehensively obtains the probability of occurrence of various fault types in the future.

[0005] Among the above existing methods, one mainly relies on information such as historical fault frequency, meteorological data, and overall operation safety factor to evaluate the probability of failure, and the other focuses on risk assessment and control based on overall operation data. Both methods require real-time collection of a large amount of power grid operation data for analysis and judgment. The data sources are extensive and complex, and zoning management is not considered to optimize the data processing process, resulting in complex calculations and low efficiency when facing massive data during the analysis process. At the same time, it is difficult to conduct detailed fault prediction and risk prevention and control based on the characteristics of different regions, which affects the accuracy and pertinence of the analysis results and cannot meet the needs of efficient operation and precise management of modern power grids. Summary of the invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a power grid analysis system and analysis method based on a topology diagram, which are used to solve the technical problems of low analysis efficiency and insufficient pertinence in the existing power grid analysis methods.

[0007] To achieve the above object, a first aspect of the present invention provides a power grid analysis system based on a topology diagram, comprising:

[0008] Model building module: used to collect power grid equipment information and topological connection relationship data, build equipment status library and power grid topology model;

[0009] Power grid partitioning module: used to obtain the importance of several nodes and the density of several edge connections according to the power grid topology model, and use the multi-scale strongly connected area finite partitioning method to partition the power grid according to the importance of several nodes and the density of several edge connections to obtain several power grid partitions;

[0010] Risk assessment module: used to obtain the comprehensive risk index of several power grid partitions according to the importance of several nodes and the connection density of several edges, and screen out the power grid partitions whose comprehensive risk index exceeds the preset risk threshold to obtain risk partitions;

[0011] Fault prediction module: used to collect the operating data of the risk partition, input the operating data into the fault prediction model, and obtain the fault time and faulty equipment predicted by the risk partition; wherein, the fault prediction model is built based on a deep learning algorithm.

[0012] Based on the above technical modules, the topology-based power grid analysis system covers a series of complete and critical links from the collection of power grid equipment information and topological relationships to power grid zoning, risk assessment, and fault prediction. It can comprehensively analyze and control the operation status of the power grid, and provide a systematic solution to ensure the safe and stable operation of the power grid.

[0013] Furthermore, the building of the device status library includes:

[0014] A1, based on the D5000CIME model file, extract several devices and status information in the power grid to obtain a basic information library of device status; wherein the status information includes device identification, type, parameters, opening and closing status, voltage, current and power;

[0015] A2, receives the device status change data pushed by the D5000 system in real time, obtains the device status auxiliary information through the remote signal, verifies and analyzes the device status change data and the device status auxiliary information, and obtains the analyzed real-time device status data;

[0016] A3, compare and update the real-time device status data with the existing data in the device status basic information library to obtain a real-time updated device status library.

[0017] The D5000 system plays an important role in power grid operation monitoring. By using the D5000CIME model file and pushed data, it can efficiently and accurately obtain and update equipment status information, meeting the requirements of power grid analysis for real-time and accuracy of equipment status.

[0018] Furthermore, the grid topology model is obtained by:

[0019] B1, obtain the topological connection relationship data of the power grid from the D5000 system, parse and transform the topological connection relationship data, and obtain the transformed topological data structure; wherein the nodes of the topological data structure are the power grid equipment, and the edges represent the connection lines;

[0020] B2, associate the device status library with the topology data structure, verify and update the node devices and electrical parameters of the topology data structure, and obtain the power grid topology model.

[0021] Associating the device status library with the topological data structure can not only verify and update the node devices and electrical parameters of the topological data structure, but also make the power grid topology model no longer just a simple connection relationship model, but a comprehensive model that integrates device status information. It can more truly reflect the actual operation architecture of the power grid and provide a more accurate basis for various analyses based on the topological model.

[0022] Furthermore, the obtaining of several node importances and several edge connection densities according to the power grid topology model includes:

[0023] C1, the edge connection density Wij of edge eij between node i and node j is calculated according to the formula Wij = (Iij)^2 / [(Uij)^2×(Rij)^2+(Xij)^2]; where Iij represents the current of edge eij, Uij represents the voltage difference between the two ends of edge eij, Rij and Xij represent the resistance and reactance of the connection line corresponding to edge eij respectively;

[0024] C2, calculate the average weight of the edge connecting node vi according to the formula AW(vi) = (1 / |N(vi)|) × ∑Wij; where |N(vi)| represents the degree of the node, that is, the number of neighboring nodes of node vi, and the summed Wij represents the closeness of the edge connection between node vi and its neighbor node vj;

[0025] C3, calculate the power consumption weight of node vi according to the formula r(vi)=PG(vi) / (PG+PL); where PG represents the total power generation of all power generation equipment in the power grid, PL represents the total load capacity consumed by all power consumption equipment in the power grid, and PG(vi) represents the power consumption of node vi;

[0026] C4, calculate the importance of node vi according to the formula Ip(vi)=a1×|N(vi)|+a2×AW(vi)+a3×r(vi); where a1, a2, and a3 represent the weight coefficients of each item, and a1+a2+a3=1.

[0027] Current is the key parameter that reflects the flow of charge in the line, voltage difference reflects the potential difference between the two ends of the line, and resistance and reactance describe the line's resistance to current. These parameters together determine the electrical connection characteristics of the edge in the power grid, which can fully consider the physical state of the edge in the power grid operation, thereby accurately measuring its connection tightness;

[0028] At the same time, by linking the node power consumption with the total power generation power and total load capacity of the power grid, the importance of the node can be evaluated in the context of the overall supply and demand relationship of the power grid, which can reflect the role of the node in the energy flow and distribution process of the power grid and accurately quantify the importance of the node in electricity consumption.

[0029] Furthermore, the multi-scale strongly connected region finite partitioning method comprises:

[0030] D1, sort several nodes according to their importance Ip(vi), and select m nodes as several initial partition centers C = {c1, c2, ..., cm};

[0031] D2, using Dijkstra algorithm to calculate the shortest path p(vi,cj) from several nodes to several initial partition centers, and calculate the distance from several nodes vi to several initial partition centers cj according to the shortest path, node importance and edge connection density, and obtain several distances d(vi,cj);

[0032] D3, for each node vi, according to the logical formula vi∈Pk if d(vi,ck)(vi,ck)=min{d(vi,cj),j=1,2,3,...,m}, assign vi to the partition to which the partition center with the smallest distance belongs, and obtain the preliminary partition result; where Pk represents the partition, k=1,2,3,...,m;

[0033] D4, adjust the preliminary partitioning results based on strong connectivity to obtain several power grid partitions.

[0034] Based on the multi-scale strongly connected region finite partitioning method, the power grid can be partitioned scientifically and rationally. In this method, the layout of each power grid partition is reasonable based on factors such as node importance and distance. At the same time, with the help of strong connectivity adjustment, the strong connectivity within the partition is ensured, so that the characteristics of each partition can be analyzed and processed more conveniently.

[0035] Furthermore, the calculation formulas of the several distances d(vi,cj) are: d(vi,cj)=λ∑wkl+(1-λ)|Ip(vi)-Ip(cj)|; wherein wkl represents the connection density of edge ekl, ekl∈p(vi,cj), and p(vi,cj) represents the shortest distance from vi to cj, and λ represents a preset weight.

[0036] By comprehensively considering the differences in edge connection density and node importance, the obtained distance can more reasonably measure the distance from the node to the partition center. Compared with the distance calculation based solely on the topological shortest path, it is more in line with the electrical characteristics and actual needs of power grid operation, making the partition results more scientific and reasonable, and better reflecting the internal structure and operation characteristics of the power grid.

[0037] Furthermore, the adjusting of the preliminary partitioning result based on strong connectivity includes:

[0038] D4-1, starting from any node vi∈Vk in the current partition, use the depth-first search algorithm to determine whether vi can access all other nodes in the current partition; if yes, mark the current partition as strongly connected, and node vi does not need to be partitioned; if no, jump to D4-2; where Vk represents the set of nodes in the current partition, and the center of the current partition is ck;

[0039] D4-2, backtrack the depth-first search process, record the nodes where branches appear in the search path and cannot continue to visit other unvisited nodes, and obtain several key nodes va;

[0040] D4-3, calculate the distance d(va,cj) from the key node to the center of other partitions, assign the key node to the partition to which the partition center with the smallest distance belongs, and obtain the adjusted partition; where cj∈C and cj≠ck;

[0041] D4-4, repeat D4-1 to D4-3 until each partition is marked as strongly connected.

[0042] The preliminary partitioning results are adjusted based on strong connectivity. The depth-first search algorithm is used to determine whether a node can access all other nodes in the partition. The key nodes of the partitions that do not meet the strong connectivity are reallocated, ensuring that each partition is strongly connected in terms of electrical connection and avoiding the occurrence of electrical islands within the partition. This makes the partitioning results not only reasonable in division, but also more stable and reliable in terms of electrical connection.

[0043] Furthermore, the method for obtaining the comprehensive risk index of the plurality of power grid partitions includes:

[0044] E1, calculate the average AI(Pk) and standard deviation σI(Pk) of the importance of several nodes in partition Pk;

[0045] E2, calculate the average value AW(Pk) and standard deviation σW(Pk) of the connection density of several edges in partition Pk;

[0046] E3. The comprehensive risk index Rk of partition Pk is calculated according to the formula: Rk = b1×(σI(Pk) / AI(Pk))+b2×(σW(Pk) / AW(Pk)); wherein b1 and b2 represent risk weights, and b1+b2=1.

[0047] Furthermore, the fault prediction model is constructed based on a deep learning algorithm, including:

[0048] F1, collect historical operation data from the power grid operation history database to obtain a historical operation data set containing equipment operation parameters, equipment status change data, and environmental monitoring data within a certain period of time; wherein the equipment operation parameters include voltage, current, and power, the equipment status change data includes the start time and stop time of the equipment, and the environmental monitoring data includes temperature, humidity, and wind speed;

[0049] F2, after preprocessing the historical operation data set, annotate it according to the faulty equipment and fault time to obtain the preprocessed data set and annotated set;

[0050] F3, divide the preprocessed data set annotation set according to a preset ratio to obtain a training set and a validation set;

[0051] F4, build a time series model based on the deep learning algorithm, input the training set and the validation set into the time series model for iterative training and optimization, and obtain the time series model with the best validation accuracy;

[0052] F5, save the time series model with the best verification accuracy as a fault prediction model; the input of the fault prediction model is the equipment operation parameters, equipment status change data, and environmental monitoring data within a period of time, and the output is the predicted faulty equipment and predicted fault time.

[0053] Deep learning algorithms require a large amount of high-quality data for training. Through the above-mentioned data processing and model building steps, we can give full play to the advantages of deep learning in processing complex data relationships, improve the accuracy and reliability of fault prediction, and meet the requirements of power grid operation and maintenance for fault prediction.

[0054] A second aspect of the present invention provides a power grid analysis method based on a topology diagram, comprising:

[0055] S1, collect power grid equipment information and topological connection relationship data, and build a device status library and power grid topology model;

[0056] S2, obtaining the importance of several nodes and the density of several edge connections according to the power grid topology model, and using the multi-scale strongly connected region finite partitioning method to partition the power grid according to the importance of several nodes and the density of several edge connections, to obtain several power grid partitions;

[0057] S3, obtaining comprehensive risk indexes of several power grid partitions according to the importance of several nodes and the connection density of several edges, and screening out power grid partitions exceeding a preset risk threshold to obtain risk partitions;

[0058] S4, collecting the operating data of the risk partition, inputting the operating data into the fault prediction model, and obtaining the fault time and faulty equipment predicted by the risk partition; wherein the fault prediction model is constructed based on a deep learning algorithm.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The importance of nodes and the density of edge connections are calculated through the power grid topology model, and the power grid is partitioned using the multi-scale strongly connected region finite partitioning method. The power grid is divided into several relatively independent and internally closely connected regions, so that subsequent analysis can be carried out separately for each partition, greatly reducing the scale of data processing;

[0061] By calculating the relevant indicators in the partition to obtain a comprehensive risk index, the risk partition is accurately located to improve the pertinence of risk prevention and control. When predicting faults, data is collected based on risk partitions, and deep learning models are used to mine the operating characteristics of the partitioned power grid, accurately predict fault time and equipment, and provide effective early warning for operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 A schematic diagram of a process flow of a power grid analysis system based on a topology diagram provided by the present invention;

[0064] Figure 2 A schematic diagram of a process for adjusting the preliminary partitioning result based on strong connectivity provided by the present invention;

[0065] Figure 3 This is a schematic diagram of the operation of the risk assessment module provided by the present invention. DETAILED DESCRIPTION

[0066] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. 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.

[0067] See also Figure 1-Figure 3 The first embodiment of the present invention provides a power grid analysis system based on a topology diagram, comprising:

[0068] Model building module: used to collect power grid equipment information and topological connection relationship data, build equipment status library and power grid topology model;

[0069] Power grid partitioning module: used to obtain the importance of several nodes and the density of several edge connections according to the power grid topology model, and use the multi-scale strongly connected area finite partitioning method to partition the power grid according to the importance of several nodes and the density of several edge connections to obtain several power grid partitions;

[0070] Risk assessment module: used to obtain the comprehensive risk index of several power grid partitions according to the importance of several nodes and the connection density of several edges, and screen out the power grid partitions exceeding the preset risk threshold to obtain risk partitions;

[0071] Fault prediction module: used to collect the operating data of risk partitions, input the operating data into the fault prediction model, and obtain the fault time and faulty equipment predicted by the risk partitions; among them, the fault prediction model is built based on a deep learning algorithm.

[0072] It should be noted that the model building module, power grid partitioning module, risk assessment module and fault prediction module in the system are communicatively connected.

[0073] Accurate and comprehensive basic data is the premise for effective power grid analysis. Only by clearly understanding the equipment status and power grid connection status can we further analyze the operation characteristics and potential risks of the power grid. In this invention, the model building module is used to collect power grid equipment information and topological connection relationship data, and to build a device status library and a power grid topology model, so as to lay a reliable foundation for subsequent analysis work:

[0074] A1, based on the D5000CIME model file, extract various equipment in the power grid and its various status information, such as equipment identifier, equipment type parameters, equipment operating parameters, equipment opening and closing status, as well as voltage, current and power data that play a key role in the power grid operation status, and build a basic equipment status information database;

[0075] A2, receive the equipment status change data pushed by the D5000 system in real time, obtain the auxiliary information of the equipment status through the remote signal, conduct strict data verification and analysis on the above data, and obtain the real-time equipment status data after analysis;

[0076] A3, compare and update the real-time equipment status data with the existing data in the equipment status basic information library to ensure that the equipment status library can reflect the actual status of the equipment in real time and accurately, becoming a reliable basis for power grid equipment status monitoring.

[0077] After obtaining the grid topology connection relationship data from the D5000 system, it is parsed and converted to form a topology data structure with grid equipment as nodes and connection lines as edges;

[0078] Specifically, the topological connection relationship data in the D5000 system has a specific format and encoding, which needs to be interpreted by a special parsing program. When parsing, it is necessary to identify the node identification, attributes (such as type, rated parameters) and connection line (such as connection node pairs, line impedance, type) information of the power grid equipment, and then convert the parsed information into a topological data structure according to the internal data structure specification. For example, the node information is organized into a node list (including identification, attributes and edge pointers), and the edge information is composed of an edge list (including connection node identification, attributes and operation function pointers).

[0079] Next, traverse the device records in the device status library, obtain the identification and real-time status information (such as opening and closing, operating parameters), find the corresponding node in the topological data structure and update its attributes. For example, if the circuit breaker is closed, mark the node as "conducting" in the topological data structure, and update the temporary variables related to the electrical parameters of the connected edges;

[0080] Then, the validity and connection relationship of the edges in the topological data structure are dynamically adjusted according to the changes in the device status. For example, if the disconnector is opened, the relevant edges in the topological data structure are disconnected to reflect the actual electrical connection status of the power grid. A reverse index mechanism can also be established to facilitate quick access to device status information from the topological data structure;

[0081] Afterwards, a comprehensive check is conducted to see if the devices and nodes in the device status library and the topology data structure correspond one to one, the identification is compared, any mismatches are recorded, and the power grid topology is updated in a timely manner. If a new device is connected but the topology is not updated, the node and connection relationship and initial status information are added; if the device is retired or removed, the relevant nodes and edges are deleted and marked in the device status library; the electrical parameters of the device in the updated device status library are compared with the electrical parameters of the corresponding nodes or connected edges in the topology data structure to ensure the consistency of the electrical parameters and the accuracy of subsequent power grid analysis and calculations;

[0082] Finally, verify the rationality of the topology structure, check the connection between the busbar and the outgoing equipment, eliminate unreasonable ring networks and island nodes, and make corrections based on power grid rules. At the same time, establish a dynamic topology update mechanism to monitor equipment status changes in real time to trigger the update process, and automatically repair or alarm when problems are found.

[0083] Through the above steps, a complete and dynamically associated power grid topology model is constructed. This model not only reflects the physical connection relationship between power grid equipment, but also reflects the operating status of the equipment and its impact on the power grid topology in real time, so that the operating behavior of the power grid can be simulated and analyzed more accurately, providing strong support for the stable operation and effective management of the power system.

[0084] Then, in the power grid partitioning module, several node importances and several edge connection densities are obtained according to the power grid topology model, including:

[0085] C1, edge connection density calculation: The edge connection density Wij of edge eij between node i and node j is calculated according to the formula Wij = (Iij)^2 / [(Uij)^2×(Rij)^2+(Xij)^2]; where Iij represents the current of edge eij, Uij represents the voltage difference between the two ends of edge eij, Rij and Xij represent the resistance and reactance of the connection line corresponding to edge eij respectively;

[0086] C2, calculation of average weight of node connection edge: the average weight of node vi connection edge is calculated according to the formula AW(vi) = (1 / |N(vi)|) × ∑Wij; where |N(vi)| represents the degree of the node, that is, the number of neighbor nodes of node vi, reflecting the connection breadth of the node in the local network, and the summed Wij represents the closeness of the edge connection between node vi and its neighbor node vj;

[0087] C3, calculation of node power consumption weight: The power consumption weight of node vi is calculated according to the formula r(vi)=PG(vi) / (PG+PL); wherein PG represents the total power generation power of all power generation equipment in the power grid, representing the power generation capacity of the power grid; PL represents the total load capacity consumed by all power consumption equipment in the power grid, reflecting the overall demand for electric energy by power consumption equipment in the power grid; PG(vi) represents the power consumption of node vi, reflecting the power consumption of the power consumption equipment connected to the node;

[0088] C4, node importance calculation: The importance of node vi is calculated according to the formula Ip(vi)=a1×|N(vi)|+a2×AW(vi)+a3×r(vi); where a1, a2, and a3 represent weight coefficients of each item, obtained through historical experience, and a1+a2+a3=1;

[0089] It should be noted that the electrical-related data in the above-mentioned calculation of node importance and edge connection density are obtained in real time through the device status library. The device status library is closely related to the power grid topology model and is updated in real time. It can timely reflect the changes in device status and dynamic adjustment of electrical parameters during the operation of the power grid, effectively improving the tracking and analysis capabilities of the entire power grid analysis system for the real-time operation status of the power grid, providing a reliable data basis for subsequent operations such as power grid zoning, risk assessment and fault prediction based on these parameters, and ensuring the accuracy and timeliness of the system analysis results.

[0090] By setting the weight coefficient reasonably based on historical experience, the contribution of different factors in the node importance assessment can be adjusted according to the actual operation characteristics and analysis focus of the power grid. For example, for some power grids that take power supply reliability as the primary goal, the value of the power consumption weight coefficient a3 may be appropriately increased; while for situations that focus on network structure stability analysis, the weight coefficients a1 or a2 related to node connections can be appropriately increased. The node importance calculated in this way can fully and accurately reflect the actual importance of each node in the power grid, and provide key quantitative indicators for subsequent operations such as power grid partitioning.

[0091] Next, the multi-scale strongly connected region finite partitioning method is used to partition the power grid according to the importance of several nodes and the density of several edge connections, and several power grid partitions are obtained;

[0092] Specifically, D1, select the initial partition center:

[0093] Several nodes are sorted according to their importance, and m nodes are selected as several initial partition centers; the more important the node is, the more critical its position in the power grid is, and the greater its influence on the surrounding nodes is. Selecting the first m high-importance nodes as the core reference points for the subsequent partitioning process can make the partitioning results better reflect the internal structure of the power grid and the distribution of important nodes.

[0094] D2, calculate the distance from the node to the center of the initial partition:

[0095] The Dijkstra algorithm is used to calculate the shortest paths from several nodes to several initial partition centers, and the distances from several nodes to several initial partition centers are obtained according to the distance calculation formula d(vi,cj)=λ∑wkl+(1-λ)|Ip(vi)-Ip(cj)|, and several distances d(vi,cj) are obtained; wherein wkl represents the connection density of edge ekl, ekl∈p(vi,cj), and p(vi,cj) represents the shortest distance from vi to cj, and λ represents the preset weight, which can be based on historical experience or according to the actual situation and analysis focus of the power grid; for example, when the electrical connection stability of the power grid has a greater impact on the partition, the value of λ can be appropriately increased, and more attention can be paid to the role of edge connection density in distance calculation; conversely, when it is hoped that the partition is divided more based on the importance of the node, the value of (1-λ) can be increased;

[0096] The Dijkstra algorithm is a classic shortest path algorithm that can efficiently and accurately find the shortest path from a node to the initial partition center by gradually searching and updating the distance information from the node to the target node. When calculating the distance, it not only considers the edge connection density on the shortest path, but also combines the importance of the node and the initial partition center, making the distance calculation more consistent with the actual operating characteristics of the power grid, and can also more accurately measure the degree of association between the node and the initial partition center.

[0097] D3, preliminary partitioning:

[0098] For each node, according to the logical formula vi∈Pk if d(vi,ck)(vi,ck)=min{d(vi,cj), j=1,2,3,...,m}, vi is assigned to the partition to which the partition center with the smallest distance belongs, and the preliminary partition result is obtained. Among them, Pk represents the partition, k=1,2,3,...,m; by comparing the distance between the node and each initial partition center, it is assigned to the partition corresponding to the partition center with the closest distance, which reflects the distribution of the power grid based on the node importance and distance relationship to a certain extent, and provides an initial framework for subsequent adjustments;

[0099] D4, adjust partitions based on strong connectivity:

[0100] Furthermore, the preliminary partitioning results are adjusted based on strong connectivity to obtain several grid partitions, including:

[0101] First, starting from any node vi∈Vk in the current partition Pk, use the depth-first search algorithm to determine whether vi can access all other nodes in Pk;

[0102] The depth-first search algorithm is a classic algorithm for traversing graph structures. It starts from a given starting node and explores as deeply as possible along a path until it cannot continue or reaches the target condition. If a node can access all other nodes in the partition, it means that the partition is strongly connected under the current structure, that is, there is a path between any two nodes in the partition. In this case, the partition is marked as strongly connected and the node does not need to be partitioned; on the contrary, if a node cannot access all nodes in the partition, it means that the partition has a connection problem and needs further adjustment.

[0103] Then, the depth-first search process is backtracked to record the nodes where branches appear in the search path and other unvisited nodes cannot be accessed, and several key nodes va are obtained. In the depth-first search backtracking process, key nodes va are the key factors that cause the partition to fail to meet strong connectivity. They may be located in the weakest link in the partition, or isolated nodes or node groups caused by unreasonable partition division;

[0104] Then, according to the distance calculation formula in D2 above, the distance d(va,cj) from the key node va to other partition centers is calculated, and the key node is assigned to the partition to which the partition center with the smallest distance belongs, to obtain the adjusted partition; where cj∈C and cj≠ck, that is, the distance between the other partition centers except the current partition center and the key node is calculated, and then it is reallocated to the partition with the closest distance to improve the strong connectivity of the partition;

[0105] Finally, the entire adjustment process needs to be repeated continuously to check and optimize each partition to ensure that the partition results of the entire power grid meet the requirements of strong connectivity.

[0106] In this embodiment, it is assumed that the relevant data of a simple power grid is:

[0107] Grid topology: There are 10 nodes (labeled as v1-v10) and 12 edges; the connection relationship is: v1 is connected to v2 and v3; v2 is connected to v4 and v5; v3 is connected to v6 and v7; v4 is connected to v8; v5 is connected to v9; v6 is connected to v10, forming a small network structure;

[0108] Electrical parameters: Edge parameters are: for example, the current I12 = 5A, the voltage difference U12 = 10V, the resistance R12 = 2Ω, and the reactance X12 = 3Ω of edge e12 (connecting v1 and v2); other edges also have their own parameters; power generation and power consumption are: assuming that the total power generation of all power generation equipment in the power grid is PG = 100kW, and the total load capacity consumed by all power consumption equipment is PL = 80kW, where the power consumption of node v1 is PG(v1) = 10kW, and the power consumption of v2 is PG(v2) = 8kW, etc.; the weight coefficients are obtained based on historical experience: a1 = 0.3, a2 = 0.3, a3 = 0.4 for calculating node importance; the preset weight λ = 0.5 is used to calculate the distance;

[0109] C1, calculate the edge connection density:

[0110] Taking edge e12 as an example, according to the formula W12 = (I12)^2 / [(U12)^2×(R12)^2+(X12)^2], we can get: W12 = 25 / (100×4+9) = 25 / 409≈0.061; the connection density Wij of other edges is calculated in the same way;

[0111] C2, calculate the average weight of the edge connecting the node:

[0112] For node v1, its degree |N(v1)|=2 (connected to v2 and v3), W12=0.061, W13=0.05 (assuming it has been calculated), then according to the formula AW(v1)=(1 / |N(v1)|)×∑Wij, we can get AW(v1)=(1 / 2)×(0.061+0.05)=0.0555; calculate the average weight AW(vi) of the connecting edges of other nodes in turn;

[0113] C3, calculates the node power consumption weight:

[0114] For node v1, according to the formula r(v1)=PG(v1) / (PG+PL)=10 / (100+80)≈0.056; similarly calculate the power consumption weights r(v_i) of other nodes;

[0115] C4, calculates the importance of nodes:

[0116] Taking node v1 as an example, according to the formula Ip(v1=a1×|N(v1)|+a2×AW(v1)+a3×r(v1)=0.3×2+0.3×0.0555+0.4×0.056≈0.625; calculate the importance Ip(v_i) of all nodes in turn;

[0117] D1, select the initial partition center:

[0118] Sort the 10 nodes by importance. Assume that the top three nodes are selected as the initial partition centers. For example, c1 = v3 (importance is 0.7), c2 = v5 (importance is 0.68), c3 = v7 (importance is 0.65), then C = {c1, c2, c3};

[0119] D2, calculate the distance from the node to the center of the initial partition:

[0120] Taking node v1 as an example, use Dijkstra algorithm to calculate its shortest path to c1, c2, and c3; assume that the shortest path p(v1, c1) to c1 passes through edge e13 (the connection density W13 has been calculated), and then calculate d(v1, c1) according to the distance formula; similarly calculate the distance to c2 and c3;

[0121] D3, preliminary partitioning: compare the distances from node v1 to c1, c2, and c3. Assuming d(v1, c1) is the smallest, v1 is assigned to partition P1 centered on c1. All nodes are assigned in this way to obtain preliminary partitioning results.

[0122] D4, finally adjust the partitions based on strong connectivity until each partition is marked as strongly connected, and finally obtain the adjusted power grid partitions.

[0123] As a relatively independent and interrelated part of the power grid, the internal electrical characteristics and operating status of each partition have a vital impact on the stability and security of the entire power grid. Therefore, after obtaining the power grid partition, the risk assessment module is used to conduct risk assessment on each partition:

[0124] First, calculate the average AI(Pk) and standard deviation σI(Pk) of the importance of several nodes in partition Pk; and calculate the average AW(Pk) and standard deviation σW(Pk) of the connection density of several edges in partition Pk; find out whether there are nodes with large differences in importance or areas with abnormal connection density in the partition; these nodes may be risk focus points in the partition, for example, nodes with extremely high or extremely low importance may have a greater impact on the operation of the partition; similarly, too high connection density will also increase the risk of fault propagation, while too low connection density will affect power supply reliability; therefore, by calculating the average and standard deviation, the characteristics of potential risk areas are locked in advance;

[0125] Then, the comprehensive risk index Rk of partition Pk is calculated according to the formula: Rk = b1×(σI(Pk) / AI(Pk))+b2×(σW(Pk) / AW(Pk)); b1 and b2 represent risk weights, which are determined based on historical experience and expert evaluation. The formula measures the fluctuation of node importance and edge connection density within the partition by calculating the ratio of standard deviation to the mean value, which is used as the risk assessment standard to obtain a reasonable and accurate risk assessment method.

[0126] Finally, the calculated comprehensive risk index is rigorously compared with the preset risk threshold to screen out the grid partitions that exceed the preset risk threshold and obtain the risk partitions that need special attention and timely processing.

[0127] Next, the fault prediction module receives the risk zoning results of the risk assessment module, and then collects the operating data of the risk zoning, including equipment operating parameters, equipment status change data, and environmental monitoring data; the operating data is input into the fault prediction model to obtain the fault time and faulty equipment predicted by the risk zoning, and provide reliable early warning information for power grid operation and maintenance, helping operation and maintenance personnel to take necessary preventive measures or make emergency repair preparations before the fault occurs, thereby effectively reducing the impact of the fault on the power grid system and ensuring the stability and reliability of power supply; wherein, the equipment operating parameters include but are not limited to voltage, current and power, the equipment status change data include but are not limited to the start time and stop time of the equipment, and the environmental monitoring data include but are not limited to temperature, humidity and wind speed, etc.;

[0128] In the fault prediction module, the fault prediction model construction process is as follows:

[0129] F1, collect historical operation data from the power grid operation history database to obtain a historical operation data set containing equipment operation parameters, equipment status change data, and environmental monitoring data within a certain period of time;

[0130] F2, after preprocessing the historical operation data set, annotate it according to the faulty equipment and fault time to obtain the preprocessed data set and annotation set; preprocessing can remove interference factors such as noise and outliers in the data and improve data quality. The annotation operation makes the data have clear fault orientation, which is convenient for the model to learn fault characteristics and improve the accuracy and efficiency of model training;

[0131] F3, divide the preprocessed data set annotation set according to a preset ratio to obtain a training set and a validation set;

[0132] F4, build a time series model based on the deep learning algorithm, input the training set and the validation set into the time series model for iterative training and optimization, and obtain the time series model with the best validation accuracy;

[0133] In this embodiment, a model architecture based on LSTM-Attention can be used to construct a time series model; LSTM is used to process time-series power grid operation data and can effectively learn long-term dependencies in the data; the attention mechanism is used to dynamically focus on key information in the data, further improving the model's ability to capture historical operation characteristics and prediction accuracy;

[0134] F5, save the time series model with the best verification accuracy as a fault prediction model; the input of the fault prediction model is the equipment operation parameters, equipment status change data, and environmental monitoring data within a period of time, and the output is the predicted faulty equipment and predicted fault time.

[0135] In summary, the power grid analysis system based on topology diagram starts from data collection and model construction, passes through power grid zoning, risk assessment, and fault prediction, forming a complete closed-loop analysis process. The modules cooperate with each other and comprehensively use a variety of algorithms and data processing methods to comprehensively improve the accuracy, reliability and foresight of power grid analysis, providing strong technical support for the safe, stable and efficient operation of the power grid.

[0136] A second aspect of the present invention provides a topology-based power grid analysis method, comprising:

[0137] S1, collect power grid equipment information and topological connection relationship data, and build a device status library and power grid topology model;

[0138] S2, obtaining the importance of several nodes and the density of several edge connections according to the power grid topology model, and using the multi-scale strongly connected region finite partitioning method to partition the power grid according to the importance of several nodes and the density of several edge connections, to obtain several power grid partitions;

[0139] S3, obtaining comprehensive risk indexes of several power grid partitions according to the importance of several nodes and the connection density of several edges, and screening out power grid partitions exceeding a preset risk threshold to obtain risk partitions;

[0140] S4, collecting the operating data of the risk partition, inputting the operating data into the fault prediction model, and obtaining the fault time and faulty equipment predicted by the risk partition; wherein the fault prediction model is constructed based on a deep learning algorithm.

[0141] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0142] Working principle of the present invention:

[0143] Through the model building module, the status data of the equipment is collected to build a real-time updated equipment status library, and the power grid topology model is built based on the equipment status library and the topological link data of the power grid;

[0144] The power grid partitioning module is used to calculate and process the edges and nodes of the power grid topology model to obtain the node importance, and the initial partition center is selected according to the node sorting. Then, a multi-scale strongly connected area finite partitioning method combining the Dijkstra algorithm and the depth-first algorithm is used to obtain a scientific and reasonable power grid partition.

[0145] In the risk assessment module, the grid topology data of each partition is analyzed, combined with the real-time electrical parameters in the equipment status library, to obtain the comprehensive risk index of each grid partition, and the partitions exceeding the preset threshold are screened out and output as risk partitions;

[0146] Through the fault prediction module, the operating data of the risk zone is collected, and the fault prediction model built based on the deep learning algorithm is used to predict the fault of the risk zone, and the predicted fault time and faulty equipment are obtained to provide early warning information for power grid operation and maintenance.

[0147] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A power grid analysis system based on a topology diagram, characterized in that: include: Model building module: used to collect power grid equipment information and topological connection relationship data, build equipment status library and power grid topology model; Power grid partitioning module: used to obtain the importance of several nodes and the density of several edge connections according to the power grid topology model, and use the multi-scale strongly connected area finite partitioning method to partition the power grid according to the importance of several nodes and the density of several edge connections to obtain several power grid partitions; Risk assessment module: used to obtain the comprehensive risk index of several power grid partitions according to the importance of several nodes and the connection density of several edges, and screen out the power grid partitions whose comprehensive risk index exceeds the preset risk threshold to obtain risk partitions; Fault prediction module: used to collect the operating data of the risk partition, input the operating data into the fault prediction model, and obtain the fault time and faulty equipment predicted by the risk partition; wherein, the fault prediction model is built based on a deep learning algorithm.

2. A power grid analysis system based on a topology diagram according to claim 1, characterized in that: The step of constructing a device status library includes: A1, collect information on several devices and their status in the power grid to obtain a basic information base on device status; wherein the status information includes device identification, type, parameters, switch status, voltage, current and power; A2, collects equipment status change data in real time, obtains equipment status auxiliary information through remote signaling, verifies and analyzes equipment status change data and equipment status auxiliary information, and obtains real-time equipment status data after analysis; A3, compare and update the real-time device status data with the existing data in the device status basic information library to obtain a real-time updated device status library.

3. A power grid analysis system based on topology diagram according to claim 1, characterized in that: The method for obtaining the power grid topology model includes: B1, obtaining the topological connection relationship data of the power grid, parsing and converting the topological connection relationship data, and obtaining the converted topological data structure; wherein the nodes of the topological data structure are the power grid equipment, and the edges represent the connection lines; B2, associate the device status library with the topology data structure, verify and update the node devices and electrical parameters of the topology data structure, and obtain the power grid topology model.

4. A power grid analysis system based on topology diagram according to claim 1, characterized in that: The obtaining of several node importances and several edge connection densities according to the power grid topology model includes: C1, the edge connection density Wij of edge eij between node i and node j is calculated according to the formula Wij = (Iij)^2 / [(Uij)^2×(Rij)^2+(Xij)^2]; where Iij represents the current of edge eij, Uij represents the voltage difference between the two ends of edge eij, Rij and Xij represent the resistance and reactance of the connection line corresponding to edge eij respectively; C2, calculate the average weight of the edge connecting node vi according to the formula AW(vi) = (1 / |N(vi)|) × ∑Wij; where |N(vi)| represents the degree of the node, that is, the number of neighboring nodes of node vi, and the summed Wij represents the closeness of the edge connection between node vi and its neighbor node vj; C3, calculate the power consumption weight of node vi according to the formula r(vi)=PG(vi) / (PG+PL); where PG represents the total power generation of all power generation equipment in the power grid, PL represents the total load capacity consumed by all power consumption equipment in the power grid, and PG(vi) represents the power consumption of node vi; C4, calculate the importance of node vi according to the formula Ip(vi)=a1×|N(vi)|+a2×AW(vi)+a3×r(vi); where a1, a2, and a3 represent the weight coefficients of each item, and a1+a2+a3=1.

5. The power grid analysis system based on topology diagram according to claim 1, characterized in that: The multi-scale strongly connected region finite partitioning method comprises: D1, sort several nodes according to their importance Ip(vi), and select m nodes as several initial partition centers C = {c1, c2, ..., cm}; D2, using Dijkstra algorithm to calculate the shortest path p(vi,cj) from several nodes to several initial partition centers, and calculate the distance from several nodes vi to several initial partition centers cj according to the shortest path, node importance and edge connection density, and obtain several distances d(vi,cj); D3, for each node vi, according to the logical formula vi∈Pk if d(vi,ck)(vi,ck)=min{d(vi,cj),j=1,2,3,...,m}, assign vi to the partition to which the partition center with the smallest distance belongs, and obtain the preliminary partition result; where Pk represents the partition, k=1,2,3,...,m; D4, adjust the preliminary partitioning results based on strong connectivity to obtain several power grid partitions.

6. A power grid analysis system based on topology diagram according to claim 5, characterized in that: The calculation formulas for the several distances d(vi,cj) are: d(vi,cj)=λ∑wkl+(1-λ)|Ip(vi)-Ip(cj)|; wherein wkl represents the connection density of edge ekl, ekl∈p(vi,cj), and p(vi,cj) represents the shortest distance from vi to cj, and λ represents the preset weight.

7. A power grid analysis system based on topology diagram according to claim 5, characterized in that: The adjustment of the preliminary partitioning result based on strong connectivity includes: D4-1, starting from any node vi∈Vk in the current partition, use the depth-first search algorithm to determine whether vi can access all other nodes in the current partition; if yes, mark the current partition as strongly connected, and node vi does not need to be partitioned; if no, jump to D4-2; where Vk represents the set of nodes in the current partition, and the center of the current partition is ck; D4-2, backtrack the depth-first search process, record the nodes where branches appear in the search path and cannot continue to visit other unvisited nodes, and obtain several key nodes va; D4-3, calculate the distance d(va,cj) from the key node to the center of other partitions, assign the key node to the partition to which the partition center with the smallest distance belongs, and obtain the adjusted partition; where cj∈C and cj≠ck; D4-4, repeat D4-1 to D4-3 until each partition is marked as strongly connected.

8. A power grid analysis system based on topology diagram according to claim 1, characterized in that: The method for obtaining the comprehensive risk index of the plurality of power grid sections includes: E1, calculate the average AI(Pk) and standard deviation σI(Pk) of the importance of several nodes in partition Pk; E2, calculate the average value AW(Pk) and standard deviation σW(Pk) of the connection density of several edges in partition Pk; E3. The comprehensive risk index Rk of partition Pk is calculated according to the formula: Rk = b1×(σI(Pk) / AI(Pk))+b2×(σW(Pk) / AW(Pk)); wherein b1 and b2 represent risk weights, and b1+b2=1.

9. A power grid analysis system based on topology diagram according to claim 1, characterized in that: The fault prediction model is built based on a deep learning algorithm and includes: F1, collect historical operation data from the power grid operation history database to obtain a historical operation data set containing equipment operation parameters, equipment status change data, and environmental monitoring data within a certain period of time; wherein the equipment operation parameters include voltage, current, and power, the equipment status change data includes the start time and stop time of the equipment, and the environmental monitoring data includes temperature, humidity, and wind speed; F2, after preprocessing the historical operation data set, annotate it according to the faulty equipment and fault time to obtain the preprocessed data set and annotated set; F3, divide the preprocessed data set annotation set according to a preset ratio to obtain a training set and a validation set; F4, build a time series model based on the deep learning algorithm, input the training set and the validation set into the time series model for iterative training and optimization, and obtain the time series model with the best validation accuracy; F5, save the time series model with the best verification accuracy as a fault prediction model; the input of the fault prediction model is the equipment operation parameters, equipment status change data, and environmental monitoring data within a period of time, and the output is the predicted faulty equipment and predicted fault time.

10. A power grid analysis method based on a topology map, applied to a power grid analysis system based on a topology map as claimed in any one of claims 1 to 9, characterized in that: include: S1, collect power grid equipment information and topological connection relationship data, and build a device status library and power grid topology model; S2, obtaining the importance of several nodes and the density of several edge connections according to the power grid topology model, and using the multi-scale strongly connected region finite partitioning method to partition the power grid according to the importance of several nodes and the density of several edge connections, to obtain several power grid partitions; S3, obtaining comprehensive risk indexes of several power grid partitions according to the importance of several nodes and the connection density of several edges, and screening out power grid partitions exceeding a preset risk threshold to obtain risk partitions; S4, collecting the operating data of the risk partition, inputting the operating data into the fault prediction model, and obtaining the fault time and faulty equipment predicted by the risk partition; wherein the fault prediction model is constructed based on a deep learning algorithm.

Images