A method and system for cross-fusion mining of information in substations

Abstract

The invention discloses a method and system for cross-fusion mining of information in substations, which belongs to the field of smart grid anomaly detection, including: obtaining grid equipment information, grid state data, environmental data and grid fault records, fusing them after processing abnormal values, and obtaining original grid fusion data; discretizing the original grid fusion data using a decision tree model to obtain grid fusion data; constructing an undirected graph of grid equipment according to grid equipment topological structure data, and dividing the undirected graph into multiple communities; determining a Bayesian network model of each community using a scoring search method; training the Bayesian network model to obtain a conditional probability table of the community; calculating the fault code at the current moment according to the conditional probability table and judging the abnormal situation of the grid equipment; the invention combines community division and Bayesian network, and can realize efficient and accurate fault detection and positioning through cross-fusion of multi-source data, thereby improving the operation and maintenance management efficiency of the grid.

Description

Technical Field

[0001] The present invention relates to the field of smart grid anomaly detection, and in particular to a method and system for cross-fusion mining of substation information. Background Art

[0002] With the development of the power system, power companies have higher and higher requirements for power supply reliability and quality. During the operation of the power system, fault detection and diagnosis are important links to ensure the safe and stable operation of the power system. However, some problems existing in the current power system make fault detection and diagnosis work face challenges. First, the data sources in the power system are diverse, including power load data, equipment operation data, user power consumption behavior data, environmental data, etc. These data are usually stored in different systems and platforms, forming data islands, which are difficult to analyze and utilize comprehensively. Secondly, due to the possible transmission errors and equipment failures in the process of power system data collection, the collected data may have quality problems such as redundancy, missing, and inconsistency, which affect the accuracy and reliability of data analysis.

[0003] Traditional fault detection and diagnosis methods mainly rely on predefined rules and models, such as rule-based methods, model-based methods, signal processing-based methods, etc. However, these methods have some limitations, such as relying on expert experience and knowledge; complex model construction, large amount of calculation, and insufficient real-time performance; they are sensitive to noise, and the data processing algorithms are complex, making it difficult to cope with the variability of the power system operating environment.

[0004] Although data-driven methods, such as machine learning and deep learning, can automatically extract fault features from big data and build intelligent detection and diagnosis models, these methods are highly dependent on data, while the quality and coverage of data are insufficient, which affects the training and application effects of the model. In addition, existing methods often lack effective fusion and comprehensive utilization of multi-source data, making it difficult to fully grasp the operating status of the power system.

[0005] A Chinese patent application, with publication number CN116881675A, is titled A method for monitoring the state of power equipment based on a Bayesian algorithm. It proposes to collect parameter information and state information of power equipment, conduct association analysis, construct primary and secondary directed acyclic graphs, and establish a Bayesian network model. This method uses posterior probability to update the association relationship. When the parameter or state information changes, the state of power equipment and abnormal nodes are predicted through directed acyclic graphs and conditional probability tables, so as to achieve accurate monitoring of the state of power equipment and abnormal prediction, and improve prediction accuracy and equipment maintenance efficiency; however, when the number of equipment increases significantly, large-scale predictions will lead to reduced prediction accuracy.

[0006] A Chinese patent application, with publication number CN110796368A, entitled "Method and device for dynamic risk assessment of community distribution network based on Bayesian network", proposes to obtain distribution feeder short-circuit status information by inputting the collected current and voltage preset data into the distribution feeder short-circuit fault diagnosis model. Then, weather information, operating years information of community distribution system components and distribution feeder short-circuit status information are input into the power outage risk assessment model to obtain the community power outage risk assessment result. However, this solution is limited to assessing the power outage risk of the distribution network and cannot monitor the abnormal conditions of power equipment in real time. Summary of the invention

[0007] In order to solve the above technical problems, a technical solution adopted by the present invention is to provide a method for cross-fusion mining of information in substations, the method comprising:

[0008] S10. Obtaining grid equipment information and grid status data from the power monitoring system, obtaining environmental data from the meteorological system, and obtaining grid fault records from the power maintenance system;

[0009] S20. Use the moving average method to process the outliers of the grid state data, and fuse the outlier-processed grid state data with the grid equipment information, the environmental data, and the grid fault record to obtain the original grid fusion data;

[0010] S30. Discretize the original grid fusion data using a decision tree model, map the original grid fusion data to discrete intervals, and obtain grid fusion data;

[0011] S40. Obtain the topological structure data of the power grid equipment, construct an undirected graph of the power grid equipment according to the topological structure data, divide the undirected graph using the Girvan-Newman algorithm to obtain multiple communities, create a regional data set for each community, and store the power grid fusion data into the regional data set of each community according to the affiliation;

[0012] S50. construct an initial Bayesian network model for each community, and use a scoring search method to determine the Bayesian network model for each community;

[0013] S60. Inputting the grid integration data at the historical moment in the regional data set of each community into the Bayesian network model of the community, training the Bayesian network model, and obtaining a conditional probability table for each community;

[0014] S70. Use the conditional probability table of each community to calculate the power grid fusion data at the current moment in the regional data set, obtain the fault code at the current moment, and determine the abnormal situation of the power grid equipment at the current moment according to the fault code.

[0015] Furthermore, the S10 includes:

[0016] The power grid equipment information includes: {equipment number, equipment type};

[0017] The equipment types include: {primary distribution transformer, secondary distribution transformer, circuit breaker, distribution cabinet, busbar, power generation equipment, energy storage equipment, electric vehicle charging pile, end-user equipment, renewable energy equipment};

[0018] The grid status data includes: {device number, voltage, current, power, switch status, power factor, time};

[0019] The environmental data includes: {humidity, temperature, time};

[0020] The power grid fault record includes: {equipment number, fault cause, time}.

[0021] Furthermore, the S20 includes:

[0022] S21. Calculate the smoothed voltage of the grid state data using the moving average method, and the calculation formula is: ,in Indicates the device number. Indicated in time, Indicates the number of time points included in calculating the average value. An intermediate variable representing the summation symbol, Indicates that the device number in the power grid status data is The equipment is The voltage at the moment, Indicates the device number is The equipment is The smoothed voltage at the time;

[0023] S22. Continue to use the moving average method to calculate the smoothed current and smoothed power of the grid state data, Indicates the device number is The equipment is The smoothed current at the moment, Indicates the device number is The equipment is The smoothed power at the time;

[0024] S23. Determine in sequence whether all voltages, currents, and powers in the grid status data are abnormal. , then the voltage in the grid status data The value of If , then the current in the grid status data The value of If , then the power in the grid status data The value of The value of ; where Indicates the device number is The equipment is The voltage at the moment, Indicates the voltage abnormality threshold, Indicates the device number is The equipment is The current at the moment, Indicates the current abnormality threshold, Indicates the device number is The equipment is The power of the moment, Indicates the power abnormality threshold;

[0025] S24. According to the device number, the grid status data after abnormal value processing is connected with the grid equipment information to obtain the grid equipment status data, and the grid equipment status data includes: {device number, device type number, voltage, current, power, switch state, power factor, time}; the device type number is represented by numbers 1 to 10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 respectively represent the primary side distribution transformer, the secondary side distribution transformer, the circuit breaker, the distribution cabinet, the busbar, the power generation equipment, the energy storage equipment, the electric vehicle charging pile, the end user equipment, and the renewable energy equipment;

[0026] S25. Count the number of all fault causes in the power grid fault record and sort the fault causes in descending order according to the number, use the sorting sequence number as the fault code to replace the fault cause in the power grid fault record, and connect the power grid equipment status data with the environmental data and the power grid fault record according to the equipment number and time to obtain the original power grid fusion data, wherein the original power grid fusion data includes: {equipment number, equipment type number, voltage, current, power, switch status, power factor, temperature, humidity, time, fault code}, wherein the fault cause that is not matched in the power grid equipment status data is filled with the fault code 0, indicating that the fault cause is normal.

[0027] Furthermore, the step S30 includes:

[0028] The device type number and voltage, device type number and current, device type number and power, device type number and power factor, device type number and temperature, and device type number and humidity in the original power grid fusion data are respectively used as inputs of the decision tree, and the fault code is used as the output of the decision tree to build a decision tree, and the segmentation points in the decision tree model are obtained. According to the segmentation points, the voltage, current, power, power factor, temperature, and humidity are respectively divided into multiple discrete intervals, and the discrete intervals are represented by unique discrete values. The voltage, current, power, power factor, temperature, and humidity in the original power grid fusion data are mapped to the discrete intervals to obtain power grid fusion data; the power grid fusion data includes: {device number, device type number, discrete voltage, discrete current, discrete power, switch state, discrete power factor, discrete temperature, discrete humidity, time, fault code}.

[0029] Furthermore, the S40 includes:

[0030] S41. Define the power grid equipment as a graph node, the connection relationship between the equipment is the undirected edge of the graph, and construct an undirected graph of the power grid equipment, where the weights of the undirected edges are all 1;

[0031] S42. Use Dijkstra algorithm to calculate all possible shortest paths between each pair of nodes in the undirected graph, and calculate the edge betweenness of each edge. The calculation formula of the edge betweenness is:

[0032] ;

[0033] in, represents the edges in the undirected graph, represents the node set of the undirected graph, and represents two different nodes in the undirected graph, Representation Node and nodes The total number of all shortest paths between Representation Node and nodes Passing by the edge The total number of all shortest paths, Represents edge The edge betweenness of

[0034] S43. Delete the edge with the highest betweenness in the undirected graph, update the undirected graph, and determine whether there are still edges in the updated undirected graph. If so, go to S44; if not, go to S45;

[0035] S44. Create the adjacency matrix of the updated undirected graph , the adjacency matrix 1 indicates that there is an edge between nodes, and 0 indicates that there is no edge between nodes. The modularity of the updated undirected graph is calculated. , go to S43, the modularity The calculation formula is:

[0036] ;

[0037] in, represents the set of edges of the updated undirected graph, represents the set size of the edges of the updated undirected graph, and represents two different nodes in the updated undirected graph, represents the node set of the updated undirected graph, Represents the adjacency matrix Middle Nodes and The element value at the corresponding position of the node, and Respectively represent nodes and nodes The degree, and Respectively represent nodes and nodes The connected components that belong to represents a binary function, if and The connected components they belong to are the same, then ,otherwise ;

[0038] S45. Obtain the maximum modularity among the modularities of all updated undirected graphs, take the undirected graph corresponding to the maximum modularity as a partition graph, take each connected component of the partition graph as a community, create a regional data set for each community, and store the power grid fusion data in the regional data set of each community according to the relationship between the device number in the power grid fusion data and the community.

[0039] Furthermore, the S50 includes:

[0040] S51. For each community, create 100 initial matrices with 10 rows and 10 columns, randomly fill the 100 initial matrices with 0 and 1, and require that the values ​​of the 10th column in the 100 initial matrices are not all 0, then set all the elements on the main diagonal of the 100 initial matrices to 0, use the 100 initial matrices as adjacency matrices to construct 100 directed graphs, use the topological sorting method to determine whether the 100 directed graphs have loops, and if a directed graph has a loop, randomly shuffle the elements except the main diagonal in the initial matrix corresponding to the directed graph, until 100 initial matrices are obtained and the 100 initial matrices can construct acyclic directed graphs;

[0041] S52. The acyclic directed graphs constructed by the 100 initial matrices are respectively used as the initial Bayesian network models, the information criterion of all the initial Bayesian network models is calculated, the initial matrices corresponding to the 50 initial Bayesian network models with the highest information criterion are selected as the hybridization matrix, and the initial Bayesian network model with the largest information criterion is used as the candidate Bayesian network model;

[0042] S53. Randomly draw with replacement from the 50 hybridization matrices, select 50 pairs of hybridization matrices in total, and hybridize the 50 pairs of hybridization matrices respectively to obtain 50 sub-matrices, wherein the hybridization represents randomly selecting elements from a pair of hybridization matrices, and each element in the sub-matrix is ​​selected from a pair of hybridization matrices;

[0043] S54. Using the 50 sub-matrices as adjacency matrices to construct 50 directed graphs, using a topological sorting method to determine in turn whether the 50 directed graphs have loops, if so, deleting the sub-matrices corresponding to the directed graphs with loops, and performing random sampling with replacement from the 50 hybridization matrices, selecting a pair of hybridization matrices for hybridization to obtain a sub-matrix, until 50 sub-matrices are obtained and the 50 sub-matrices can construct acyclic directed graphs;

[0044] S55. The acyclic directed graph constructed by the 50 sub-matrices is used as a hybrid Bayesian network model, the information criterion of all hybrid Bayesian network models is calculated, and the hybrid Bayesian network model with the largest information criterion is used as a candidate Bayesian network model, and the information criterion of the candidate Bayesian network model and the alternative Bayesian network model are compared, and the network model corresponding to the largest information criterion in each community is used as the preferred Bayesian network model of the community;

[0045] S56. Loop through steps S51 to S55, compare the information criterion of the preferred Bayesian network model obtained in each loop with the information criterion of the preferred Bayesian network model obtained in the previous loop, and use the preferred Bayesian network model corresponding to the larger value of the information criterion as the Bayesian network model of the community. If the number of loops reaches 10 and the Bayesian network model of the community remains unchanged, the loop terminates. Otherwise, the loop terminates when the number of loops reaches 200.

[0046] Furthermore, the S60 includes:

[0047] For each community, the attributes in the regional data set are used as nodes of the Bayesian network model of the community, and the attributes include: equipment type number, discrete voltage, discrete current, discrete power, switch state, discrete power factor, discrete temperature, discrete humidity, time, fault code, which are respectively represented as nodes of the Bayesian network model ; Input the power grid fusion data at the historical moment in the regional data set into the Bayesian network model for training, calculate the conditional probability of the Bayesian network model, and obtain the conditional probability table of each community, the conditional probability table includes the probability of a certain node under all possible parent node combinations, the parent node represents a node pointing to another node through a directed edge in the Bayesian network model, and the calculation formula of the conditional probability is:

[0048] ;

[0049] in, represents the first nodes, , Representation Node The corresponding attribute is in the region set Kind of value, is a positive integer, Represents the node All parent nodes of Represents the node The value combination of the parent node; Represents the node The value is When its parent node value combination is The joint probability when Represents the node The parent node value combination is The marginal probability when Indicates that at the node The parent node value combination is When The value is The conditional probability when .

[0050] Furthermore, the S70 includes:

[0051] S71. For each community, obtain the grid fusion data in the regional data set at the current moment, and input the value in the grid fusion data into the node of the Bayesian network model , calculate the current time node according to the conditional probability table The probability under each value is calculated as:

[0052] ;

[0053] in, represents the nodes of the Bayesian network model, Representation Node Some possible value of Representation Node The value combination of Indicates that at the node The value combination of When The value is The conditional probability when ; Represents the node The value is When The value combination is The joint probability when Represents the node The value combination is The marginal probability when

[0054] The calculation formula of the joint probability is:

[0055] ;

[0056] in, Representation Node The value is The probability of Represents the current time node The value of Indicates that at a given node The parent node of In the case of Value is The conditional probability is obtained by querying the conditional probability table;

[0057] The calculation formula of the edge probability is:

[0058] ;

[0059] in, Represents the node Some possible value of ;

[0060] S72. For each community, obtain the node at the current moment The maximum probability value under each value, the value corresponding to the maximum probability value As the node at the current moment The value of the node The value of is used to obtain the fault code of the power grid fusion data in the regional data set at the current moment; if the fault code is not 0, the faulty device is found according to the device number of the power grid fusion data, and the fault cause of the faulty device at the current moment is obtained according to the correspondence between the fault code and the fault cause.

[0061] Further, the S52 includes:

[0062] The information criterion is calculated as follows:

[0063] ;

[0064] in, represents the maximum likelihood estimate of the initial Bayesian network model, represents the number of parameters in the initial Bayesian network model, Represents the hyperparameters of the initial Bayesian network model, with a value range of , represents the information criterion of the initial Bayesian network model.

[0065] Furthermore, a cross-fusion mining system for substation information includes:

[0066] Data acquisition module: used to obtain power grid equipment information and power grid status data from the power monitoring system, environmental data from the meteorological system, and power grid fault records from the power maintenance system;

[0067] Data fusion module: used to process the outliers of the power grid status data, and then fuse the power grid status data after outlier processing with the power grid equipment information, the environmental data, and the power grid fault records to obtain original power grid fusion data; and then use a decision tree model to discretize the original power grid fusion data to obtain power grid fusion data;

[0068] Community division module: used to obtain the topological structure data of the power grid equipment, then construct an undirected graph of the power grid equipment according to the topological structure data, divide the undirected graph using the Girvan-Newman algorithm to obtain multiple communities, and finally create a regional data set for each community, and store the power grid fusion data into the regional data set of each community according to the affiliation;

[0069] Bayesian network module: used to determine the Bayesian network model of each community; input the power grid fusion data at the historical moment in the regional data set of each community into the Bayesian network model of the community, train the Bayesian network model, and obtain the conditional probability table of each community;

[0070] Anomaly mining module: used to calculate the power grid fusion data at the current moment in the regional data set of each community, obtain the fault code at the current moment, and judge the abnormal situation of the power grid equipment at the current moment according to the fault code.

[0071] The beneficial effects of the present invention are:

[0072] 1. The present invention manages massive amounts of area data by dividing the community, and the data is calculated and analyzed on-site in the community, which can improve data security and reduce the delay and overhead of long-distance data transmission.

[0073] 2. The present invention fully considers the potential factors that cause failures in power grid equipment, cross-integrates multi-source data, provides comprehensive and accurate information for fault mining, and utilizes the interpretability and efficient reasoning and computing capabilities of the Bayesian network model to achieve accurate and real-time fault detection and positioning, improve the operation and maintenance efficiency of the power grid, and promote the further development of smart grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a flow chart of a method for cross-fusion mining of information in substations provided by the present invention.

[0075] Figure 2 It is a flow chart of community division for a method of cross-fusion mining of information in substations provided by the present invention.

[0076] Figure 3 It is a module diagram of a substation information cross-fusion mining system provided by the present invention. DETAILED DESCRIPTION

[0077] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0078] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present invention, rather than all of the embodiments.

[0079] Figure 1 : is a flow chart of a method for cross-fusion mining of information in substations provided by an embodiment of the present invention, the method comprising:

[0080] S10. Obtaining grid equipment information and grid status data from the power monitoring system, obtaining environmental data from the meteorological system, and obtaining grid fault records from the power maintenance system;

[0081] S20. Use the moving average method to process the outliers of the grid state data, and fuse the outlier-processed grid state data with the grid equipment information, the environmental data, and the grid fault record to obtain the original grid fusion data;

[0082] S30. Discretize the original grid fusion data using a decision tree model, map the original grid fusion data to discrete intervals, and obtain grid fusion data;

[0083] S40. Obtain the topological structure data of the power grid equipment, construct an undirected graph of the power grid equipment according to the topological structure data, divide the undirected graph using the Girvan-Newman algorithm to obtain multiple communities, create a regional data set for each community, and store the power grid fusion data into the regional data set of each community according to the affiliation;

[0084] S50. construct an initial Bayesian network model for each community, and use a scoring search method to determine the Bayesian network model for each community;

[0085] S60. Inputting the grid integration data at the historical moment in the regional data set of each community into the Bayesian network model of the community, training the Bayesian network model, and obtaining a conditional probability table for each community;

[0086] S70. Use the conditional probability table of each community to calculate the power grid fusion data at the current moment in the regional data set, obtain the fault code at the current moment, and determine the abnormal situation of the power grid equipment at the current moment according to the fault code.

[0087] In this embodiment, first, the power grid equipment information and power grid state data are obtained from the power monitoring system, the environmental data are obtained from the meteorological system, and the power grid fault records are obtained from the power maintenance system; the power grid state data is processed for outliers using the moving average method, and the outlier processed power grid state data is fused with the power grid equipment information, environmental data, and power grid fault records to obtain the original power grid fusion data; then the original power grid fusion data is discretized using the decision tree model, and the original power grid fusion data is mapped to discrete intervals to obtain the power grid fusion data; then the topological structure data of the power grid equipment is obtained, and an undirected graph of the power grid equipment is constructed according to the topological structure data, and the Girvan-Newman algorithm is used. The undirected graph is divided into multiple communities, and a regional data set is created for each community. The power grid fusion data is stored in the regional data set of each community according to the affiliation. Then, an initial Bayesian network model is constructed for each community, and the Bayesian network model of each community is determined using the scoring search method. Then, the power grid fusion data at historical moments in the regional data set of each community is input into the Bayesian network model of the community, and the Bayesian network model is trained to obtain a conditional probability table for each community. Finally, the conditional probability table of each community is used to calculate the power grid fusion data at the current moment in the regional data set, and the fault code at the current moment is obtained. The abnormal situation of the power grid equipment at the current moment is judged according to the fault code.

[0088] Furthermore, the S10 includes:

[0089] The power grid equipment information includes: {equipment number, equipment type};

[0090] The equipment types include: {primary distribution transformer, secondary distribution transformer, circuit breaker, distribution cabinet, busbar, power generation equipment, energy storage equipment, electric vehicle charging pile, end-user equipment, renewable energy equipment};

[0091] The grid status data includes: {device number, voltage, current, power, switch status, power factor, time};

[0092] The environmental data includes: {humidity, temperature, time};

[0093] The power grid fault record includes: {equipment number, fault cause, time}.

[0094] Furthermore, the S20 includes:

[0095] S21. Calculate the smoothed voltage of the grid state data using the moving average method, and the calculation formula is: ,in Indicates the device number. Indicated in time, Indicates the number of time points included in calculating the average value. An intermediate variable representing the summation symbol, Indicates that the device number in the power grid status data is The equipment is The voltage at the moment, Indicates the device number is The equipment is The smoothed voltage at the time;

[0096] S22. Continue to use the moving average method to calculate the smoothed current and smoothed power of the grid state data, Indicates the device number is The equipment is The smoothed current at the moment, Indicates the device number is The equipment is The smoothed power at the time;

[0097] S23. Determine in sequence whether all voltages, currents, and powers in the grid status data are abnormal. , then the voltage in the grid status data The value of If , then the current in the grid status data The value of If , then the power in the grid status data The value of The value of ; where Indicates the device number is The equipment is The voltage at the moment, Indicates the voltage abnormality threshold, Indicates the device number is The equipment is The current at the moment, Indicates the current abnormality threshold, Indicates the device number is The equipment is The power of the moment, Indicates the power abnormality threshold;

[0098] S24. According to the device number, the grid status data after abnormal value processing is connected with the grid equipment information to obtain the grid equipment status data, and the grid equipment status data includes: {device number, device type number, voltage, current, power, switch state, power factor, time}; the device type number is represented by numbers 1 to 10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 respectively represent the primary side distribution transformer, the secondary side distribution transformer, the circuit breaker, the distribution cabinet, the busbar, the power generation equipment, the energy storage equipment, the electric vehicle charging pile, the end user equipment, and the renewable energy equipment;

[0099] S25. Count the number of all fault causes in the power grid fault record and sort the fault causes in descending order according to the number, use the sorting sequence number as the fault code to replace the fault cause in the power grid fault record, and connect the power grid equipment status data with the environmental data and the power grid fault record according to the equipment number and time to obtain the original power grid fusion data, wherein the original power grid fusion data includes: {equipment number, equipment type number, voltage, current, power, switch status, power factor, temperature, humidity, time, fault code}, wherein the fault cause that is not matched in the power grid equipment status data is filled with the fault code 0, indicating that the fault cause is normal.

[0100] Furthermore, the step S30 includes:

[0101] The device type number and voltage, device type number and current, device type number and power, device type number and power factor, device type number and temperature, and device type number and humidity in the original power grid fusion data are respectively used as inputs of the decision tree, and the fault code is used as the output of the decision tree to build a decision tree, and the segmentation points in the decision tree model are obtained. According to the segmentation points, the voltage, current, power, power factor, temperature, and humidity are respectively divided into multiple discrete intervals, and the discrete intervals are represented by unique discrete values. The voltage, current, power, power factor, temperature, and humidity in the original power grid fusion data are mapped to the discrete intervals to obtain power grid fusion data; the power grid fusion data includes: {device number, device type number, discrete voltage, discrete current, discrete power, switch state, discrete power factor, discrete temperature, discrete humidity, time, fault code}.

[0102] refer to Figure 2 In another embodiment of the present invention, the S40 includes:

[0103] First, define the power grid equipment as a graph node, and the connection relationship between the equipment as the undirected edge of the graph, and construct an undirected graph of the power grid equipment, where the weights of the undirected edges are all 1; use the Dijkstra algorithm to calculate all possible shortest paths between each pair of nodes in the undirected graph, calculate the edge betweenness of each edge, delete the edge with the highest edge betweenness in the undirected graph, and update the undirected graph. If there are still edges in the updated undirected graph, create an adjacency matrix of the updated undirected graph and calculate its modularity, and then continue to delete the edges with the highest edge betweenness in the updated undirected graph and calculate its modularity until there are no edges in the updated undirected graph, then obtain the maximum modularity among the modularities of all updated undirected graphs, use the undirected graph corresponding to the maximum modularity as a partition graph, use each connected component of the partition graph as a community, create a regional data set for each community, and store the power grid fusion data in the regional data set of each community according to the relationship between the device number in the power grid fusion data and the community;

[0104] Specifically, the calculation formula of the edge betweenness is:

[0105] ;

[0106] in, represents the edges in the undirected graph, represents the node set of the undirected graph, and represents two different nodes in the undirected graph, Representation Node and nodes The total number of all shortest paths between Representation Node and nodes Passing by the edge The total number of all shortest paths, Represents edge The edge betweenness of

[0107] Specifically, the modularity The calculation formula is:

[0108] ;

[0109] in, represents the set of edges of the updated undirected graph, represents the set size of the edges of the updated undirected graph, and represents two different nodes in the updated undirected graph, represents the node set of the updated undirected graph, Represents the adjacency matrix Middle Nodes and The element value at the corresponding position of the node, and Respectively represent nodes and nodes The degree, and Respectively represent nodes and nodes The connected components that belong to represents a binary function, if and The connected components they belong to are the same, then ,otherwise .

[0110] Furthermore, the S50 includes:

[0111] S51. For each community, create 100 initial matrices with 10 rows and 10 columns, randomly fill the 100 initial matrices with 0 and 1, and require that the values ​​of the 10th column in the 100 initial matrices are not all 0, then set all the elements on the main diagonal of the 100 initial matrices to 0, use the 100 initial matrices as adjacency matrices to construct 100 directed graphs, use the topological sorting method to determine whether the 100 directed graphs have loops, and if a directed graph has a loop, randomly shuffle the elements except the main diagonal in the initial matrix corresponding to the directed graph, until 100 initial matrices are obtained and the 100 initial matrices can construct acyclic directed graphs;

[0112] S52. The acyclic directed graphs constructed by the 100 initial matrices are respectively used as the initial Bayesian network models, the information criterion of all the initial Bayesian network models is calculated, the initial matrices corresponding to the 50 initial Bayesian network models with the highest information criterion are selected as the hybridization matrix, and the initial Bayesian network model with the largest information criterion is used as the candidate Bayesian network model;

[0113] S53. Randomly draw with replacement from the 50 hybridization matrices, select 50 pairs of hybridization matrices in total, and hybridize the 50 pairs of hybridization matrices respectively to obtain 50 sub-matrices, wherein the hybridization represents randomly selecting elements from a pair of hybridization matrices, and each element in the sub-matrix is ​​selected from a pair of hybridization matrices;

[0114] S54. Using the 50 sub-matrices as adjacency matrices to construct 50 directed graphs, using a topological sorting method to determine in turn whether the 50 directed graphs have loops, if so, deleting the sub-matrices corresponding to the directed graphs with loops, and performing random sampling with replacement from the 50 hybridization matrices, selecting a pair of hybridization matrices for hybridization to obtain a sub-matrix, until 50 sub-matrices are obtained and the 50 sub-matrices can construct acyclic directed graphs;

[0115] S55. The acyclic directed graph constructed by the 50 sub-matrices is used as a hybrid Bayesian network model, the information criterion of all hybrid Bayesian network models is calculated, and the hybrid Bayesian network model with the largest information criterion is used as a candidate Bayesian network model, and the information criterion of the candidate Bayesian network model and the alternative Bayesian network model are compared, and the network model corresponding to the largest information criterion in each community is used as the preferred Bayesian network model of the community;

[0116] S56. Loop through steps S51 to S55, compare the information criterion of the preferred Bayesian network model obtained in each loop with the information criterion of the preferred Bayesian network model obtained in the previous loop, and use the preferred Bayesian network model corresponding to the larger value of the information criterion as the Bayesian network model of the community. If the number of loops reaches 10 and the Bayesian network model of the community remains unchanged, the loop terminates. Otherwise, the loop terminates when the number of loops reaches 200.

[0117] Furthermore, the S60 includes:

[0118] For each community, the attributes in the regional data set are used as nodes of the Bayesian network model of the community, and the attributes include: equipment type number, discrete voltage, discrete current, discrete power, switch state, discrete power factor, discrete temperature, discrete humidity, time, fault code, which are respectively represented as nodes of the Bayesian network model ; Inputting the power grid fusion data at the historical moment in the regional data set into the Bayesian network model for training, calculating the conditional probability of the Bayesian network model, and obtaining a conditional probability table for each community, wherein the conditional probability table includes the probability of a certain node under all possible parent node combinations, and the parent node represents a node in the Bayesian network model that points to another node through a directed edge;

[0119] Specifically, the calculation formula of the conditional probability is:

[0120] ;

[0121] in, represents the first nodes, , Representation Node The corresponding attribute is in the region set Kind of value, is a positive integer, Represents the node All parent nodes of Represents the node The value combination of the parent node; Represents the node The value is When its parent node value combination is The joint probability when Represents the node The parent node value combination is The marginal probability when Indicates that at the node The parent node value combination is When The value is The conditional probability when .

[0122] Furthermore, the S70 includes:

[0123] S71. For each community, obtain the grid fusion data in the regional data set at the current moment, and input the value in the grid fusion data into the node of the Bayesian network model , calculate the current time node according to the conditional probability table The probability under each value is calculated as:

[0124] ;

[0125] in, represents the nodes of the Bayesian network model, Representation Node Some possible value of Representation Node The value combination of Indicates that at the node The value combination of When The value is The conditional probability when ; Represents the node The value is When The value combination is The joint probability when Represents the node The value combination is The marginal probability when

[0126] Specifically, the calculation formula of the joint probability is:

[0127] ;

[0128] in, Representation Node The value is The probability of Represents the current time node The value of Indicates that at a given node The parent node of In the case of Value is The conditional probability is obtained by querying the conditional probability table;

[0129] Specifically, the calculation formula of the edge probability is:

[0130] ;

[0131] in, Represents the node Some possible value of ;

[0132] S72. For each community, obtain the node at the current moment The maximum probability value under each value, the value corresponding to the maximum probability value As the node at the current moment The value of the node The value of is used to obtain the fault code of the power grid fusion data in the regional data set at the current moment; if the fault code is not 0, the faulty device is found according to the device number of the power grid fusion data, and the fault cause of the faulty device at the current moment is obtained according to the correspondence between the fault code and the fault cause.

[0133] Further, the S52 includes:

[0134] The information criterion is calculated as follows:

[0135] ;

[0136] in, represents the maximum likelihood estimate of the initial Bayesian network model, represents the number of parameters in the initial Bayesian network model, Represents the hyperparameters of the initial Bayesian network model, with a value range of , represents the information criterion of the initial Bayesian network model.

[0137] A system for cross-fusion mining of information oriented to a substation area provided by an embodiment of the present invention is introduced below. The system for cross-fusion mining of information oriented to a substation area described below and the method for cross-fusion mining of information oriented to a substation area described above can refer to each other.

[0138] refer to Figure 3 , a cross-fusion mining system for station area information, including:

[0139] Data acquisition module: used to obtain power grid equipment information and power grid status data from the power monitoring system, environmental data from the meteorological system, and power grid fault records from the power maintenance system;

[0140] Data fusion module: used to process the outliers of the power grid status data, and then fuse the power grid status data after outlier processing with the power grid equipment information, the environmental data, and the power grid fault records to obtain original power grid fusion data; and then use a decision tree model to discretize the original power grid fusion data to obtain power grid fusion data;

[0141] Community division module: used to obtain the topological structure data of the power grid equipment, then construct an undirected graph of the power grid equipment according to the topological structure data, divide the undirected graph using the Girvan-Newman algorithm to obtain multiple communities, and finally create a regional data set for each community, and store the power grid fusion data into the regional data set of each community according to the affiliation;

[0142] Bayesian network module: used to determine the Bayesian network model of each community; input the power grid fusion data at the historical moment in the regional data set of each community into the Bayesian network model of the community, train the Bayesian network model, and obtain the conditional probability table of each community;

[0143] Anomaly mining module: used to calculate the power grid fusion data at the current moment in the regional data set of each community, obtain the fault code at the current moment, and judge the abnormal situation of the power grid equipment at the current moment according to the fault code.

[0144] In this embodiment, the data acquisition module obtains power grid equipment information and power grid status data from the power monitoring system, obtains environmental data from the meteorological system, and obtains power grid fault records from the power maintenance system; the data fusion module performs outlier processing on the power grid status data, and then performs table fusion on the power grid status data after outlier processing with the power grid equipment information, the environmental data, and the power grid fault records to obtain original power grid fusion data; then, a decision tree model is used to discretize the original power grid fusion data to obtain power grid fusion data; the community division module obtains the topological structure data of the power grid equipment, and then constructs an undirected graph of the power grid equipment according to the topological structure data, and uses Girvan-N The Ewman algorithm divides the undirected graph to obtain multiple communities, and finally creates a regional data set for each community, and stores the power grid fusion data into the regional data set of each community according to the affiliation; the Bayesian network module determines the Bayesian network model of each community; the power grid fusion data at historical moments in the regional data set of each community is input into the Bayesian network model of the community, and the Bayesian network model is trained to obtain a conditional probability table for each community; finally, the power grid fusion data at the current moment in the regional data set of each community is calculated in the anomaly mining module to obtain the fault code at the current moment, and the abnormal situation of the power grid equipment at the current moment is judged according to the fault code.

[0145] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0146] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for cross-fusion mining of information in substations, characterized in that: include: S10. Obtaining grid equipment information and grid status data from the power monitoring system, obtaining environmental data from the meteorological system, and obtaining grid fault records from the power maintenance system; S20. Use the moving average method to process the outliers of the grid state data, and fuse the outlier-processed grid state data with the grid equipment information, the environmental data, and the grid fault record to obtain the original grid fusion data; S30. Discretize the original grid fusion data using a decision tree model, map the original grid fusion data to discrete intervals, and obtain grid fusion data; S40. Obtain the topological structure data of the power grid equipment, construct an undirected graph of the power grid equipment according to the topological structure data, divide the undirected graph using the Girvan-Newman algorithm to obtain multiple communities, create a regional data set for each community, and store the power grid fusion data into the regional data set of each community according to the affiliation; S50. construct an initial matrix for each community, construct an initial Bayesian network model based on the initial matrix, select a hybrid matrix and an alternative Bayesian network model according to the information criterion of the initial Bayesian network model, hybridize the hybrid matrix to obtain a submatrix, use a topological sorting method to check the existence of a ring in the submatrix, if the submatrix has a ring, delete the submatrix, continue hybridization until all submatrices can construct acyclic directed graphs, construct a hybrid Bayesian network model based on the submatrix, select the best model from the hybrid Bayesian network model and the alternative Bayesian network model according to the information criterion as the preferred Bayesian network model of the community, iterate cyclically, and determine the Bayesian network model of each community according to the information criterion of the preferred Bayesian network model obtained in each iteration; S60. For each community, the attributes in the regional data set are used as nodes of the Bayesian network model of the community, the grid integration data in the regional data set of each community at a historical moment is input into the Bayesian network model of the community, the Bayesian network model is trained, the conditional probability of the Bayesian network model is calculated, and a conditional probability table of each community is obtained; S70. Use the conditional probability table of each community to calculate the power grid fusion data at the current moment in the regional data set, obtain the fault code at the current moment, and determine the abnormal situation of the power grid equipment at the current moment according to the fault code.

2. A method for cross-fusion mining of information in substations as claimed in claim 1, characterized in that: The S10 comprises: The power grid equipment information includes: {equipment number, equipment type}; The equipment types include: {primary distribution transformer, secondary distribution transformer, circuit breaker, distribution cabinet, busbar, power generation equipment, energy storage equipment, electric vehicle charging pile, end-user equipment, renewable energy equipment}; The grid status data includes: {device number, voltage, current, power, switch status, power factor, time}; The environmental data includes: {humidity, temperature, time}; The power grid fault record includes: {equipment number, fault cause, time}.

3. The method for cross-fusion mining of information in substations according to claim 1, characterized in that: The S20 includes: S21. Calculate the smoothed voltage of the grid state data using the moving average method, and the calculation formula is: ,in Indicates the device number. Indicated in time, Indicates the number of time points included in calculating the average value. An intermediate variable representing the summation symbol, Indicates that the device number in the power grid status data is The equipment is The voltage at the moment, Indicates the device number is The equipment is The smoothed voltage at the time; S22. Continue to use the moving average method to calculate the smoothed current and smoothed power of the grid state data, Indicates the device number is The equipment is The smoothed current at the moment, Indicates the device number is The equipment is The smoothed power at the time; S23. Determine in sequence whether all voltages, currents, and powers in the grid status data are abnormal. , then the voltage in the grid status data The value of If , then the current in the grid status data The value of If , then the power in the grid status data The value of The value of ; where Indicates the device number is The equipment is The voltage at the moment, Indicates the voltage abnormality threshold, Indicates the device number is The equipment is The current at the moment, Indicates the current abnormality threshold, Indicates the device number is The equipment is The power of the moment, Indicates the power abnormality threshold; S24. According to the device number, the grid status data after abnormal value processing is connected with the grid equipment information to obtain the grid equipment status data, and the grid equipment status data includes: {device number, device type number, voltage, current, power, switch state, power factor, time}; the device type number is represented by numbers 1 to 10, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 respectively represent the primary side distribution transformer, the secondary side distribution transformer, the circuit breaker, the distribution cabinet, the busbar, the power generation equipment, the energy storage equipment, the electric vehicle charging pile, the end user equipment, and the renewable energy equipment; S25. Count the number of all fault causes in the power grid fault record and sort the fault causes in descending order according to the number, use the sorting sequence number as the fault code to replace the fault cause in the power grid fault record, and connect the power grid equipment status data with the environmental data and the power grid fault record according to the equipment number and time to obtain the original power grid fusion data, wherein the original power grid fusion data includes: {equipment number, equipment type number, voltage, current, power, switch status, power factor, temperature, humidity, time, fault code}, wherein the fault cause that is not matched in the power grid equipment status data is filled with the fault code 0, indicating that the fault cause is normal.

4. The method for cross-fusion mining of information in substations according to claim 1, characterized in that: The S30 includes: The device type number and voltage, device type number and current, device type number and power, device type number and power factor, device type number and temperature, and device type number and humidity in the original power grid fusion data are respectively used as inputs of the decision tree, and the fault code is used as the output of the decision tree to build a decision tree, and the segmentation points in the decision tree model are obtained. According to the segmentation points, the voltage, current, power, power factor, temperature, and humidity are respectively divided into multiple discrete intervals, and the discrete intervals are represented by unique discrete values. The voltage, current, power, power factor, temperature, and humidity in the original power grid fusion data are mapped to the discrete intervals to obtain power grid fusion data; the power grid fusion data includes: {device number, device type number, discrete voltage, discrete current, discrete power, switch state, discrete power factor, discrete temperature, discrete humidity, time, fault code}.

5. The method for cross-fusion mining of information in substations according to claim 1, characterized in that: The S40 comprises: S41. Define the power grid equipment as a graph node, the connection relationship between the equipment is the undirected edge of the graph, and construct an undirected graph of the power grid equipment, where the weights of the undirected edges are all 1; S42. Use Dijkstra algorithm to calculate all possible shortest paths between each pair of nodes in the undirected graph, and calculate the edge betweenness of each edge. The calculation formula of the edge betweenness is: ； in, represents the edges in the undirected graph, represents the node set of the undirected graph, and represents two different nodes in the undirected graph, Representation Node and nodes The total number of all shortest paths between Representation Node and nodes Passing by the edge The total number of all shortest paths, Represents edge The edge betweenness of S43. Delete the edge with the highest betweenness in the undirected graph, update the undirected graph, and determine whether there are still edges in the updated undirected graph. If so, go to S44; if not, go to S45; S44. Create the adjacency matrix of the updated undirected graph , the adjacency matrix 1 indicates that there is an edge between nodes, and 0 indicates that there is no edge between nodes. The modularity of the updated undirected graph is calculated. , go to S43, the modularity The calculation formula is: ； in, represents the set of edges of the updated undirected graph, represents the set size of the edges of the updated undirected graph, and represents two different nodes in the updated undirected graph, represents the node set of the updated undirected graph, Represents the adjacency matrix Middle Nodes and The element value at the corresponding position of the node, and Respectively represent nodes and nodes The degree, and Respectively represent nodes and nodes The connected components that belong to represents a binary function, if and The connected components they belong to are the same, then ,otherwise ; S45. Obtain the maximum modularity among the modularities of all updated undirected graphs, take the undirected graph corresponding to the maximum modularity as a partition graph, take each connected component of the partition graph as a community, create a regional data set for each community, and store the power grid fusion data in the regional data set of each community according to the relationship between the device number in the power grid fusion data and the community.

6. The method for cross-fusion mining of information in substations according to claim 1, characterized in that: The S50 comprises: S51. For each community, create 100 initial matrices with 10 rows and 10 columns, randomly fill the 100 initial matrices with 0 and 1, and require that the values ​​of the 10th column in the 100 initial matrices are not all 0, then set all the elements on the main diagonal of the 100 initial matrices to 0, use the 100 initial matrices as adjacency matrices to construct 100 directed graphs, use the topological sorting method to determine whether the 100 directed graphs have loops, and if a directed graph has a loop, randomly shuffle the elements except the main diagonal in the initial matrix corresponding to the directed graph, until 100 initial matrices are obtained and the 100 initial matrices can construct acyclic directed graphs; S52. The acyclic directed graphs constructed by the 100 initial matrices are respectively used as the initial Bayesian network models, the information criterion of all the initial Bayesian network models is calculated, the initial matrices corresponding to the 50 initial Bayesian network models with the highest information criterion are selected as the hybridization matrix, and the initial Bayesian network model with the largest information criterion is used as the candidate Bayesian network model; S53. Randomly draw with replacement from the 50 hybridization matrices, select 50 pairs of hybridization matrices in total, and hybridize the 50 pairs of hybridization matrices respectively to obtain 50 sub-matrices, wherein the hybridization represents randomly selecting elements from a pair of hybridization matrices, and each element in the sub-matrix is ​​selected from a pair of hybridization matrices; S54. Using the 50 sub-matrices as adjacency matrices to construct 50 directed graphs, using a topological sorting method to determine in turn whether the 50 directed graphs have loops, if so, deleting the sub-matrices corresponding to the directed graphs with loops, and performing random sampling with replacement from the 50 hybridization matrices, selecting a pair of hybridization matrices for hybridization to obtain a sub-matrix, until 50 sub-matrices are obtained and the 50 sub-matrices can construct acyclic directed graphs; S55. The acyclic directed graph constructed by the 50 sub-matrices is used as a hybrid Bayesian network model, the information criterion of all hybrid Bayesian network models is calculated, and the hybrid Bayesian network model with the largest information criterion is used as a candidate Bayesian network model, and the information criterion of the candidate Bayesian network model and the alternative Bayesian network model are compared, and the network model corresponding to the largest information criterion in each community is used as the preferred Bayesian network model of the community; S56. Loop through steps S51 to S55, compare the information criterion of the preferred Bayesian network model obtained in each loop with the information criterion of the preferred Bayesian network model obtained in the previous loop, and use the preferred Bayesian network model corresponding to the larger value of the information criterion as the Bayesian network model of the community. If the number of loops reaches 10 and the Bayesian network model of the community remains unchanged, the loop terminates. Otherwise, the loop terminates when the number of loops reaches 200.

7. The method for cross-fusion mining of information in substations according to claim 1, characterized in that: The S60 includes: For each community, the attributes in the regional data set are used as nodes of the Bayesian network model of the community, and the attributes include: equipment type number, discrete voltage, discrete current, discrete power, switch state, discrete power factor, discrete temperature, discrete humidity, time, fault code, which are respectively represented as nodes of the Bayesian network model ; Input the power grid fusion data at the historical moment in the regional data set into the Bayesian network model for training, calculate the conditional probability of the Bayesian network model, and obtain the conditional probability table of each community, the conditional probability table includes the probability of a certain node under all possible parent node combinations, the parent node represents a node pointing to another node through a directed edge in the Bayesian network model, and the calculation formula of the conditional probability is: ; in, represents the first nodes, , Representation Node The corresponding attribute is in the region set Kind of value, is a positive integer, Represents the node All parent nodes of Represents the node The value combination of the parent node; Represents the node The value is When its parent node value combination is The joint probability when Represents the node The parent node value combination is The marginal probability when Indicates that at the node The parent node value combination is When The value is The conditional probability when .

8. The method for cross-fusion mining of information in substations according to claim 1, characterized in that: The S70 includes: S71. For each community, obtain the grid fusion data in the regional data set at the current moment, and input the value in the grid fusion data into the node of the Bayesian network model , calculate the current time node according to the conditional probability table The probability under each value is calculated as: ; in, represents the nodes of the Bayesian network model, Representation Node Some possible value of Representation Node The value combination of Indicates that at the node The value combination of When The value is The conditional probability when ; Represents the node The value is When The value combination is The joint probability when Represents the node The value combination is The marginal probability when The calculation formula of the joint probability is: ； in, Representation Node The value is The probability of Represents the current time node The value of Indicates that at a given node The parent node of In the case of Value is The conditional probability is obtained by querying the conditional probability table; The calculation formula of the edge probability is: ； in, Represents the node Some possible value of ; S72. For each community, obtain the node at the current moment The maximum probability value under each value, the value corresponding to the maximum probability value As the node at the current moment The value of the node The value of is used to obtain the fault code of the power grid fusion data in the regional data set at the current moment; if the fault code is not 0, the faulty device is found according to the device number of the power grid fusion data, and the fault cause of the faulty device at the current moment is obtained according to the correspondence between the fault code and the fault cause.

9. The method for cross-fusion mining of information in substations according to claim 6, characterized in that: The S52 includes: The information criterion is calculated as follows: ； in, represents the maximum likelihood estimate of the initial Bayesian network model, represents the number of parameters in the initial Bayesian network model, Represents the hyperparameters of the initial Bayesian network model, with a value range of , represents the information criterion of the initial Bayesian network model.

10. A cross-fusion mining system for information in substations, characterized in that: Implementing a method for cross-fusion mining of information in substations as described in any one of claims 1 to 9, comprising: Data acquisition module: used to obtain power grid equipment information and power grid status data from the power monitoring system, environmental data from the meteorological system, and power grid fault records from the power maintenance system; Data fusion module: used to process the outliers of the power grid status data, and then fuse the power grid status data after outlier processing with the power grid equipment information, the environmental data, and the power grid fault records to obtain original power grid fusion data; and then use a decision tree model to discretize the original power grid fusion data to obtain power grid fusion data; Community division module: used to obtain the topological structure data of the power grid equipment, then construct an undirected graph of the power grid equipment according to the topological structure data, divide the undirected graph using the Girvan-Newman algorithm to obtain multiple communities, and finally create a regional data set for each community, and store the power grid fusion data into the regional data set of each community according to the affiliation; Bayesian network module: used to determine the Bayesian network model of each community; input the power grid fusion data at the historical moment in the regional data set of each community into the Bayesian network model of the community, train the Bayesian network model, and obtain the conditional probability table of each community; Anomaly mining module: used to calculate the power grid fusion data at the current moment in the regional data set of each community, obtain the fault code at the current moment, and judge the abnormal situation of the power grid equipment at the current moment according to the fault code.

Images