A Grid Fault Inversion Dynamic Correction Method and System Based on a Knowledge Graph

Abstract

The present invention discloses a method and system for dynamically correcting power grid fault inversion based on a knowledge graph. The method includes: constructing a power grid fault knowledge graph database; parsing power grid fault messages and using the parsing results as parameters of a power grid fault inversion model to query and compare with the power grid fault knowledge graph database to construct a power grid fault inversion model; constructing a power grid fault inversion dynamic correction model to correct the parameters of the power grid fault inversion model to obtain a final power grid fault inversion model. The advantages of the present invention are: improving the accuracy of power grid fault inversion.

Description

Technical Field

[0001] The present invention relates to the field of power grid fault inversion, and specifically relates to a dynamic correction method and system for power grid fault inversion based on a knowledge graph. Background Art

[0002] At present, with the continuous improvement of the power grid intelligence level, the continuous increase in the power grid interconnection degree, and the sharp increase in the new energy grid connection volume, the possibility of cascading faults caused by power grid faults has further increased. With the more complex new energy grid connection volume and new energy grid connection topology structure, the types of power grid faults are more complex, and the changes in power grid electrical quantities caused by power grid faults are more difficult to analyze. Quickly identifying the types of power grid faults and realizing real-time inversion of power grid faults play an important role in the real-time scheduling of power grid operation management and maintaining the safety and stability of the power grid. Therefore, rapid real-time power grid fault inversion considering the strong coupling of new energy grid connection, the flexible and changeable power grid topology, and the further complexity of power grid fault types is of great significance for the safety and stability of the power grid.

[0003] Chinese Patent Publication No. CN118040600A discloses establishing a distribution network fault inversion model according to the new operation data generated during the model deduction process combined with the fault historical operation data; according to the distribution network operation model and the distribution network fault inversion model, combined with the real-time operation data of the distribution network, conducting deduction of the relay protection setting value strategy and fault inversion of the distribution network, which can improve the operation efficiency and stability of the distribution network and provide a strong guarantee for the safe operation of the power system. However, due to the complex new energy grid connection structure and the large number of complex power grid faults in the current situation, the fault inversion model generated by this patent application often cannot be accurately modeled, resulting in inaccurate fault inversion results output. Summary of the Invention

[0004] The technical problem to be solved by the present invention lies in the problem that the fault inversion results output by the existing power grid fault inversion model are not accurate enough.

[0005] The present invention solves the above technical problem by the following technical means: A dynamic correction method for power grid fault inversion based on a knowledge graph, including:

[0006] S1. Construct a power grid fault knowledge graph database;

[0007] S2. Analyze the power grid fault message and use the analysis result as the parameters of the power grid fault inversion model to query and compare with the power grid fault knowledge graph database, and construct a power grid fault inversion model;

[0008] S3. Construct a dynamic correction model for power grid fault inversion to correct the parameters of the power grid fault inversion model, and obtain the final power grid fault inversion model.

[0009] The present invention constructs a power grid fault inversion dynamic correction model to correct the parameters of the power grid fault inversion model, which is beneficial to accurate modeling in the case of complex new energy grid connection structures and numerous complex power grid faults, so that the output fault inversion results are relatively accurate.

[0010] Furthermore, S1 includes:

[0011] Taking primary power grid equipment as entities in the power grid fault knowledge graph database, setting several classes such as relay protection device information, power grid fault types, power flow information, equipment information, and power grid fault information, and setting multiple attributes for each class to form a triple of entity-class-attribute. The historical power grid data is implemented in one-to-one correspondence in the form of the triple, thereby constructing a power grid fault knowledge graph database.

[0012] Furthermore, S2 includes:

[0013] Using a bidirectional long short-term memory network to parse real-time power grid fault messages and historical power grid fault messages. The parsing results include primary power grid equipment entity information, fault distance, power grid fault types, and the magnitude of transition resistance. The parsing results are used as power grid fault inversion model parameters, and the power grid fault inversion model parameters are corresponded one by one with the power grid fault knowledge graph database to complete the matching of the power grid fault knowledge graph. Simulate the power grid fault within the range of the power grid subnet involved in the power grid fault according to the matching results of the power grid fault knowledge graph, thereby establishing a power grid fault inversion model.

[0014] Furthermore, the real-time power grid fault messages and historical power grid fault messages include real-time power grid fault SOE information messages and real-time relay protection information management system action reports, historical power grid fault SOE information messages, and historical relay protection information management system action reports.

[0015] Furthermore, the use of a bidirectional long short-term memory network to parse real-time power grid fault messages and historical power grid fault messages includes:

[0016] Label the real-time power grid fault messages and historical power grid fault messages using the power grid fault knowledge graph in BIO format. The labeled results are input into the bidirectional long short-term memory network, and the output results of the bidirectional long short-term memory network are input into the CRF layer, and the CRF layer outputs the parsing results.

[0017] Furthermore, S3 includes:

[0018] S31. Establish a power grid topology adjacency matrix using the matching results of the power grid fault knowledge graph; the matching results of the power grid fault knowledge graph refer to the matching results between the power grid fault inversion model parameters and the power grid fault knowledge graph database;

[0019] S32. Simulate the power grid fault according to the matching result of the power grid fault knowledge graph to obtain the power grid fault simulation waveforms of different nodes in the power grid;

[0020] S33. Construct a power grid fault waveform feature matrix based on the power grid fault simulation waveforms, and input the power grid fault waveform feature matrix and the power grid topology adjacency matrix into a multi-layer GCN network connected in sequence to obtain the output matrix of the power grid fault inversion model parameters;

[0021] S34. Calculate the mean value of the Pearson correlation coefficients between the power grid fault simulation waveforms of different nodes in the power grid and the actual action report waveforms as the loss function, and determine whether the loss function meets the requirements. If it does not meet the requirements, go to S35; if it meets the requirements, go to S36;

[0022] S35. Use the output matrix of the power grid fault inversion model parameters to correct the power grid fault inversion model parameters, and return to execute S2;

[0023] S36. Obtain the final power grid fault inversion model.

[0024] Furthermore, the power grid topology adjacency matrix represents the connection relationship between the power grid topology nodes. Each row and each column are power grid topology nodes. Each element in the power grid topology adjacency matrix is represented by a number to indicate whether the nodes are connected. Each row of the power grid fault waveform feature matrix is the time point of the power grid fault simulation waveform, and each column is the power grid topology node. Each element in the power grid fault waveform feature matrix is the discretized data of the corresponding time point of the power grid fault simulation waveform corresponding to the power grid topology node.

[0025] Furthermore, the determination of whether the loss function meets the requirements means that it meets the requirements when the loss function value reaches the minimum, otherwise it does not meet the requirements.

[0026] Furthermore, the method of using the output matrix of the power grid fault inversion model parameters to correct the power grid fault inversion model parameters is to add the parameters in the original power grid fault inversion model parameters and the output matrix of the power grid fault inversion model parameters as the power grid fault inversion model parameters in the next stage, and return to execute S2.

[0027] The present invention also provides a power grid fault inversion dynamic correction system based on a knowledge graph, including:

[0028] A graph library construction module for constructing a power grid fault knowledge graph library;

[0029] An inversion model construction module for parsing the power grid fault message and querying and comparing the parsing result as the power grid fault inversion model parameters with the power grid fault knowledge graph library to construct a power grid fault inversion model;

[0030] An inversion model correction module is used to construct a power grid fault inversion dynamic correction model to correct the parameters of the power grid fault inversion model, and obtain the final power grid fault inversion model.

[0031] Furthermore, the graph library construction module is also used for:

[0032] Taking primary power grid equipment as entities of the power grid fault knowledge graph library, setting several classes such as relay protection device information, power grid fault types, power flow information, equipment information, and power grid fault information, and setting multiple attributes for each class to form a triple of entity-class-attribute, and realizing one-to-one correspondence of entities in the form of the triple for the historical data of the power grid, so as to construct the power grid fault knowledge graph library.

[0033] Furthermore, the inversion model construction module is also used for:

[0034] Using a bidirectional long short-term memory network to parse real-time power grid fault messages and historical power grid fault messages. The parsing results include primary power grid equipment entity information, fault distance, power grid fault types, and the magnitude of transition resistance. The parsing results are used as parameters of the power grid fault inversion model, and the parameters of the power grid fault inversion model are made to correspond one-to-one with the power grid fault knowledge graph library to complete the matching of the power grid fault knowledge graph. Simulate the power grid fault within the power grid subnet range involved in the power grid fault according to the matching results of the power grid fault knowledge graph, so as to establish the power grid fault inversion model.

[0035] Furthermore, the real-time power grid fault messages and historical power grid fault messages include real-time power grid fault SOE information messages and real-time action reports of the relay protection information management system, historical power grid fault SOE information messages and historical action reports of the relay protection information management system.

[0036] Furthermore, the use of the bidirectional long short-term memory network to parse real-time power grid fault messages and historical power grid fault messages includes:

[0037] Label the real-time power grid fault messages and historical power grid fault messages using the power grid fault knowledge graph, input the labeling results into the bidirectional long short-term memory network, and input the output results of the bidirectional long short-term memory network into the CRF layer, and the CRF layer outputs the parsing results.

[0038] Furthermore, the inversion model correction module is also used for:

[0039] S31. Establish a power grid topology adjacency matrix using the matching results of the power grid fault knowledge graph; the matching results of the power grid fault knowledge graph refer to the matching results between the parameters of the power grid fault inversion model and the power grid fault knowledge graph library;

[0040] S32. Simulate the power grid fault according to the matching result of the power grid fault knowledge graph to obtain the power grid fault simulation waveforms of different nodes in the power grid;

[0041] S33. Construct a power grid fault waveform feature matrix based on the power grid fault simulation waveforms, and input the power grid fault waveform feature matrix and the power grid topology adjacency matrix into a multi-layer GCN network connected in sequence to obtain an output matrix of power grid fault inversion model parameters;

[0042] S34. Calculate the mean value of the Pearson correlation coefficients between the power grid fault simulation waveforms of different nodes in the power grid and the actual action report waveforms as the loss function, and determine whether the loss function meets the requirements. If it does not meet the requirements, go to S35; if it meets the requirements, go to S36;

[0043] S35. Use the output matrix of power grid fault inversion model parameters to correct the power grid fault inversion model parameters, and return to execute the inversion model construction module;

[0044] S36. Obtain the final power grid fault inversion model.

[0045] Furthermore, the power grid topology adjacency matrix represents the connection relationship between power grid topology nodes. Each row and each column are power grid topology nodes. Each element in the power grid topology adjacency matrix is represented by a number to indicate whether the nodes are connected. Each row of the power grid fault waveform feature matrix is the time point of the power grid fault simulation waveform, and each column is the power grid topology node. Each element in the power grid fault waveform feature matrix is the discretized data of the corresponding time point of the power grid fault simulation waveform corresponding to the power grid topology node.

[0046] Furthermore, determining whether the loss function meets the requirements means that it meets the requirements when the loss function value reaches the minimum, otherwise it does not meet the requirements.

[0047] Furthermore, the method of using the output matrix of power grid fault inversion model parameters to correct the power grid fault inversion model parameters is to add the parameters in the original power grid fault inversion model parameters and the output matrix of power grid fault inversion model parameters as the power grid fault inversion model parameters in the next stage, and return to execute the inversion model construction module.

[0048] The advantages of the present invention are as follows:

[0049] (1) The present invention constructs a power grid fault inversion dynamic correction model to correct the power grid fault inversion model parameters, which is beneficial to accurate modeling in the case of complex new energy grid connection structures and many complex power grid faults, so that the output fault inversion results are more accurate.

[0050] (2) Based on the actual connection relationship between power grid topologies, the present invention sets the relevant information of power grid faults and relay protection as necessary tags and attributes, and establishes a power grid fault knowledge graph database, which can provide a data model basis for rapid real-time power grid fault inversion.

[0051] (3) By real-time parsing the power grid fault message information, the present invention converts the unstructured data in the form of text information into structured data in the form of keywords, and performs entity matching between the power grid fault information and the power grid fault knowledge graph database, realizes the rapid search of the power grid entities involved in the power grid fault and the change information of the power grid fault electrical quantities, and establishes a power grid fault model, providing a technical reference for power grid operation management personnel.

[0052] (4) Considering the connection relationship between the power grid fault waveform and the power grid topology, the present invention makes full use of a large amount of historical power grid fault data, and uses a graph convolutional neural network (GCN) to construct a dynamic correction model for power grid fault inversion. The input layer of the GCN receives the power grid node fault waveform features and the power grid topology adjacency matrix of the power grid topology graph, aggregates and transforms the node features through the GCN, learns the dependency relationship and fault propagation mode between nodes, and enhances the expression ability of the power grid fault inversion dynamic correction model. By combining the power grid fault knowledge graph with the graph convolutional neural network, the close combination of waveform features and topological features can be considered, and the fault waveform is used as the feature matrix. At the same time, the parameter output matrix is the basis for correcting the power grid fault parameters, and the mean of the Pearson coefficients between waveforms is used as the loss function, which can continuously correct the power grid fault simulation model, realize the consistency between the power grid fault inversion data and the actual power grid fault information, continuously improve the accuracy of power grid fault inversion, and provide a basis for the safe and stable operation of the power grid. Description of the Drawings

[0053] Figure 1 It is a flowchart of a dynamic correction method for power grid fault inversion based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0054] Figure 2 It is a schematic diagram of a knowledge graph triple in a dynamic correction method for power grid fault inversion based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0055] Figure 3 It is a schematic diagram of the process of constructing a power grid fault inversion model in a dynamic correction method for power grid fault inversion based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0056] Figure 4 It is a schematic diagram of structured data in a dynamic correction method for power grid fault inversion based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0057] Figure 5Schematic diagram of the process in which the long short-term memory network converts unstructured data into the required structured data in a power grid fault inversion dynamic correction method based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0058] Figure 6 Schematic diagram of the process of constructing a power grid fault inversion dynamic correction model in a power grid fault inversion dynamic correction method based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0059] Figure 7 Flowchart of dynamically correcting a power grid fault inversion model in a power grid fault inversion dynamic correction method based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0060] Figure 8 Schematic diagram of a power grid fault waveform feature matrix in a power grid fault inversion dynamic correction method based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0061] Figure 9 Schematic diagram of a power grid topology adjacency matrix in a power grid fault inversion dynamic correction method based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0062] Figure 10 Process diagram of power grid fault inversion dynamic correction for A-phase grounding fault in a power grid fault inversion dynamic correction method based on a knowledge graph disclosed in Embodiment 1 of the present invention;

[0063] Figure 11 Process diagram of power grid fault inversion dynamic correction for AB-phase interphase short circuit in a power grid fault inversion dynamic correction method based on a knowledge graph disclosed in Embodiment 1 of the present invention. Detailed implementation manners

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Apparently, the described embodiments are some, rather than all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0065] Embodiment 1

[0066] As Figure 1 shown, Embodiment 1 of the present invention provides a power grid fault inversion dynamic correction method based on a knowledge graph, including the following steps:

[0067] S1. Construct a power grid fault knowledge graph library; the specific process is as follows:

[0068] Knowledge graph technology has the ability to analyze multi-label big data and infer knowledge, and is currently widely used in the field of power systems. Due to the continuous improvement of the intelligent level of the power grid, the increasing degree of power grid interconnection, and the sharp increase in the grid connection volume of new energy, the possibility of cascading failures caused by power grid failures has further increased. With the more complex grid connection volume and grid connection topology of new energy, the types of power grid failures are more complex, and the changes in power grid electrical quantities caused by power grid failures are more difficult to analyze. Knowledge graph technology can use the graph model to retrieve the real-time data of big data grid failures, has the ability of power grid failure knowledge reasoning, and is more suitable for the power grid failure inversion dynamic correction method of the present invention.

[0069] At present, a large amount of power grid historical data is contained in the power grid operation data. A large amount of power grid relay protection and power grid failure information can be obtained through the power grid real-time operation database and the power grid relay protection information system platform, and a mapping model of power grid primary equipment entities, power grid relay protection, and power grid failure knowledge graphs is established. The power grid failure knowledge graph library is described in the form of triples, and each piece of under-frequency load shedding scheme knowledge can be decomposed into the following form: (subject, predicate, object). The subject is an individual, and the individual is an instance of a class. The predicate connects two individuals. The object is either an individual or an instance of a data type. The power grid failure knowledge graph library is established relying on power grid entities. Therefore, based on the actual connection relationship between power grid topologies, the present invention sets the relevant information of power grid failures and relay protection as necessary tags and attributes, and establishes a power grid failure knowledge graph library, which can provide a data model basis for rapid real-time power grid failure inversion.

[0070] The present invention first constructs a power grid failure knowledge graph library according to the power grid historical operation data and the power grid failure historical information in the power grid relay protection information system platform. As Figure 2 shown, the present invention takes the power grid primary equipment as the entity of the power grid failure knowledge graph library, and sets five categories: relay protection device information, power grid failure type, power flow information, equipment information, and power grid failure information. The relay protection device information is set with attributes such as device name, device manufacturer, device version, and protection information. The power grid failure type is set with three attributes: external fault, internal fault, and fault electrical quantity. The power flow information is set with attributes such as voltage, current, active power, and reactive power. The equipment information includes equipment basic information, circuit breaker opening and closing impedance, circuit breaker tripping and closing time, and line distribution impedance. The power grid failure information is set with six attributes: action time, protection action name, fault distance, phase selection result, transition resistance, and power grid failure classification. By constructing a power grid failure knowledge graph library, the power grid failure can be directly mapped to the power grid failure knowledge graph library, providing data support for power grid failure inversion. In this embodiment, a graph database is used to implement one-to-one correspondence of entities in the form of triples of entity-class-attribute for power grid historical data, and a power grid failure knowledge graph library is automatically constructed.

[0071] S2. Analyze the power grid fault message, and use the analysis result as the parameter of the power grid fault inversion model to query and compare with the power grid fault knowledge graph database, and construct the power grid fault inversion model. The specific process is as follows:

[0072] As Figure 3 shown, at present, the level of power grid intelligence is constantly improving, and the real-time information of power grid faults is directly uploaded to the power system dispatching data platform through the dispatching data network. The present invention uses a bidirectional long short-term memory network to convert the unstructured data of the real-time information of power grid faults into the required structured data. As Figure 4 shown, the unstructured data refers to the data in the form of the original text, and the structured data refers to the keyword data extracted from the text. The structured data is matched with the knowledge graph database, and the data such as the power grid fault type, transition resistance, and fault distance parsed are corresponding to the power grid fault knowledge graph database, so as to quickly and real-time establish a power grid fault inversion simulation model and complete the fast dynamic modeling of power grid faults.

[0073] Use a bidirectional long short-term memory network to analyze the real-time SOE information message of power grid faults, the real-time operation report of the relay protection information management system, the historical SOE information message of power grid faults, and the historical operation report of the relay protection information management system, that is, analyze the real-time power grid fault message and the historical power grid fault message. For the SOE information message, it is mainly divided into information such as time sequence, object, and action. The relay protection information management system mainly includes information such as action time, protection action name, fault location measurement, phase selection result, and transition resistance value. The unstructured data is converted into the required structured data through the long short-term memory network. The specific process is as Figure 5 shown. The real-time power grid fault message and the historical power grid fault message are in text form, and when the present invention performs fault matching, it needs to be matched with the knowledge graph, which requires keywords. Therefore, it is necessary to convert the text into keywords. Therefore, the present invention uses the power grid fault knowledge graph to perform BIO annotation on the real-time power grid fault message and the historical power grid fault message. The purpose of annotation is to mark the entities that appear in the text and are the same as those in the knowledge graph database. The entities are also divided into different classes and attributes. Therefore, the annotation result is input into the bidirectional long short-term memory network (BiLSTM). The bidirectional long short-term memory network classifies and scores the marked entities, that is, the bidirectional long short-term memory network scores whether the marked entities belong to the attributes set in the power grid fault knowledge graph database respectively (the attributes include device name, device manufacturer, device version, protection information, external fault, internal fault, fault electrical quantity, etc., see Figure 2 , which will not be elaborated here), and the scoring result output by the bidirectional long short-term memory network is input into the CRF layer. The CRF layer takes the attribute corresponding to the highest score result as the analysis result. The analysis result includes the entity information of the primary power grid equipment, the fault distance, the power grid fault type, and the size of the transition resistance.

[0074] The real-time message named entity recognition result of the power grid fault, that is, the above parsing result, corresponds one-to-one with the power grid fault knowledge graph, completes the matching of the power grid fault knowledge graph, and then establishes a power grid fault inversion model according to the matching result of the power grid fault knowledge graph in the range of the power grid subnet involved in the power grid fault according to the power grid fault time sequence. The essence of establishing the power grid fault inversion model is to model and display the power grid fault corresponding to the power grid fault knowledge graph in the power grid subnet range, so that the whole process can invert the power grid fault according to the power grid fault message. At present, with the continuous increase of the new energy grid-connected capacity of the power grid, the power grid fault chain tripping is becoming more and more complex. The present invention considers the power grid fault expansion modeling, uses the power grid fault knowledge graph to quickly and dynamically query the power grid action entities of the same substation and adjacent substations under the power grid fault time sequence, and realizes the accurate expansion modeling of the power grid fault inversion.

[0075] S3. Construct a power grid fault inversion dynamic correction model to correct the parameters of the power grid fault inversion model, and obtain the final power grid fault inversion model; the specific process is as follows:

[0076] The above-mentioned power grid fault inversion model often cannot directly and accurately model once in the case of the complex new energy grid-connected structure and many complex power grid faults. Therefore, the present invention establishes a power grid fault inversion dynamic correction model. To realize the dynamic correction of the power grid fault inversion model, an automatic correction model for power grid fault inversion based on the power grid fault knowledge graph library is constructed. The power grid fault waveform is directly related to the power grid topology. The present invention considers the connection relationship between the power grid fault waveform and the power grid topology, makes full use of a large amount of power grid historical fault data, constructs a power grid fault inversion dynamic correction model based on the graph convolutional neural network (GCN). The input layer of the graph convolutional neural network receives the power grid node fault waveform features and the power grid topology adjacency matrix of the power grid topology graph, aggregates and transforms the node features through the graph convolutional layer, learns the dependency relationship and fault propagation mode between nodes, adopts multiple layers of GCN to enhance the expression ability of the power grid fault inversion dynamic correction model, and at the same time records the final correction result in the power grid fault knowledge graph library to provide technical support for power grid dispatching operation management personnel. Specifically as Figure 6As shown, first, according to the steps of S1 and S2, use the constructed power grid fault knowledge graph database to automatically construct the original power grid fault inversion model, and generate the fault waveforms of the fault points. Then, perform noise reduction, filtering, and normalization processing on the fault waveforms of the power grid fault inversion model and the historical fault recorder waveforms. Next, establish a power grid fault inversion dynamic correction model based on the graph convolutional neural network (GCN). The connection relationship between the nodes of the power grid fault graph database is the input connection relationship of the graph convolutional neural network. Among them, the nodes represent electrical equipment such as buses and transformers, the edges represent transmission lines, and the attributes on the edges represent line parameters and connection relationships. The time-series waveform data of the nodes of the power grid fault inversion model after data preprocessing is the input feature vector of the graph convolutional neural network. The primary equipment entity information, fault distance, power grid fault type, and transition resistance size of each power grid node are used as the output vector. At the same time, each output vector is the dynamic correction value based on the parameters of the original power grid fault inversion model. The sum of the Pearson correlation coefficients between the fault waveforms of the power grid fault inversion model of each power grid node and the fault waveforms of the fault recorder in the historical data is the loss function of the graph convolutional neural network, and the power grid fault inversion model is continuously corrected dynamically. As Figure 7 shown, the process of correcting the parameters of the power grid fault inversion model by the power grid fault inversion dynamic correction model described above can be divided into the following steps:

[0077] (1) Use the method steps of S1 and S2 to establish the original power grid fault inversion model according to the matching results of the above power grid fault knowledge graph, and establish a power grid topology adjacency matrix using the matching results of the power grid fault knowledge graph; the matching results of the power grid fault knowledge graph refer to the matching results between the parameters of the power grid fault inversion model and the power grid fault knowledge graph database;

[0078] (2) Simulate the power grid fault according to the matching results of the power grid fault knowledge graph to obtain the power grid fault simulation waveforms of different nodes in the power grid;

[0079] (3) Construct a power grid fault waveform feature matrix, and input the power grid fault waveform feature matrix and the power grid topology adjacency matrix into a multi-layer GCN network to obtain the output matrix of the power grid fault inversion model parameters; the multi-layer GCN network is multiple GCN networks connected in sequence. The first layer of the multi-layer GCN network receives the power grid fault waveform feature matrix and the power grid topology adjacency matrix, and the last layer of the multi-layer GCN network outputs the output matrix of the power grid fault inversion model parameters.

[0080] Among them, the horizontal axis (i.e., each row) of the power grid fault waveform feature matrix is the time point of the fault waveform, and the vertical axis (i.e., each column) is the power grid topology node. Each element in the power grid fault waveform feature matrix is the discretized data of the corresponding time point of the fault waveform corresponding to the power grid topology node. The power grid topology adjacency matrix represents the connection relationship between power grid topology nodes. Both the horizontal and vertical axes are power grid topology nodes, corresponding to the vertical axis of the power grid fault waveform feature matrix. Each element in the power grid topology adjacency matrix is represented by the number 0 or 1 to indicate whether the nodes are connected or not. As Figure 8 shown, the fault current of the 110 kV bus node at time t1 is 0.23 A, and the fault current at time t2 is 0.35 A, and so on to form the power grid fault waveform feature matrix. As Figure 9 shown, 1 in the power grid topology adjacency matrix indicates that two power grid topologies are connected, and 0 indicates that two power grid topologies are not connected.

[0081] (4) Calculate the mean Pearson correlation coefficient of the simulation waveforms and action report waveforms of different power grid nodes as the loss function, and determine whether the mean Pearson correlation coefficient loss function meets the requirements. If it does not meet the requirements, go to step (5); if it meets the requirements, go to step (6); the formula for calculating the mean Pearson correlation coefficient is as follows:

[0082]

[0083]

[0084] In the formula, and are the discrete sequence values of the simulation waveform and the action report waveform respectively, is the Pearson correlation coefficient of the waveform of the first power grid node, is the Pearson correlation coefficient of the waveform of the th power grid node, and its calculation method is the same as , which will not be elaborated here. is the mean Pearson correlation coefficient of the waveforms of different power grid nodes, is the covariance of x and y, is 's standard deviation, is 's standard deviation, is the number of power grid nodes.

[0085] (5) Use the power grid fault inversion model parameter output matrix to correct the original power grid fault inversion model parameters. The specific method is to add the original power grid fault inversion model parameters and the parameters in the power grid fault inversion model parameter output matrix, which is the power grid fault inversion model parameter for the next stage, and then go to S2;

[0086]

[0087] In the formula, are the parameters of the corrected power grid fault inversion model, are the parameters of the original power grid fault inversion model, are the parameters corresponding to the original power grid fault inversion model parameters in the output matrix of the power grid fault inversion model parameters.

[0088] (6) Obtain the final power grid fault inversion model.

[0089] Taking the typical faults of phase A ground fault and phase AB interphase fault on the 110 kV power grid line as examples, a power grid fault inversion model is established by using the power grid fault knowledge graph, and the dynamic correction process of the power grid fault inversion is demonstrated. By continuously correcting the power grid fault distance, fault type, transition resistance size, etc., through Figure 10 It can be seen that during the continuous dynamic correction process of the phase A ground fault, it continuously coincides with the fault recording waveform. Through Figure 11 It can be seen that as the fault inversion of the phase AB interphase fault is continuously dynamically corrected, it is continuously consistent with the waveform recorded by the fault recorder. Finally, the fault waveform is consistent with the waveform of the actual power grid line fault recorder, and the power grid fault inversion is finally completed.

[0090] Embodiment 2

[0091] Based on Embodiment 1, Embodiment 2 of the present invention further provides a power grid fault inversion dynamic correction system based on a knowledge graph, including:

[0092] A graph library construction module for constructing a power grid fault knowledge graph library;

[0093] An inversion model construction module for parsing power grid fault messages and querying and comparing the parsing results with the power grid fault knowledge graph library as the parameters of the power grid fault inversion model to construct a power grid fault inversion model;

[0094] An inversion model correction module for constructing a power grid fault inversion dynamic correction model to correct the parameters of the power grid fault inversion model to obtain the final power grid fault inversion model.

[0095] Specifically, the graph library construction module is further used for:

[0096] Taking the primary equipment of the power grid as the entity of the power grid fault knowledge graph library, setting several classes such as relay protection device information, power grid fault type, power flow information, equipment information, and power grid fault information, and setting multiple attributes for each class to form a triple of entity-class-attribute, and realizing one-to-one correspondence of entities in the form of the triple for the power grid historical data, thereby constructing a power grid fault knowledge graph library.

[0097] Specifically, the inversion model construction module is further used for:

[0098] The real-time power grid fault messages and historical power grid fault messages are parsed using a bidirectional long short-term memory network. The parsing results include the physical information of primary power grid equipment, fault distance, power grid fault type, and the magnitude of the transition resistance. The parsing results are used as the parameters of the power grid fault inversion model. The parameters of the power grid fault inversion model are corresponded one by one with the power grid fault knowledge graph database to complete the matching of the power grid fault knowledge graph. The power grid fault is simulated according to the matching result of the power grid fault knowledge graph within the power grid subnet involved in the power grid fault, thereby establishing a power grid fault inversion model.

[0099] More specifically, the real-time power grid fault messages and historical power grid fault messages include real-time power grid fault SOE information messages and real-time action reports of the relay protection information management system, historical power grid fault SOE information messages, and historical action reports of the relay protection information management system.

[0100] More specifically, the parsing of the real-time power grid fault messages and historical power grid fault messages using the bidirectional long short-term memory network includes:

[0101] The real-time power grid fault messages and historical power grid fault messages are labeled with BIO using the power grid fault knowledge graph. The labeling results are input into the bidirectional long short-term memory network, and the output results of the bidirectional long short-term memory network are input into the CRF layer, and the CRF layer outputs the parsing results.

[0102] More specifically, the inversion model correction module is also used for:

[0103] S31. Establish a power grid topology adjacency matrix using the matching result of the power grid fault knowledge graph; the matching result of the power grid fault knowledge graph refers to the matching result between the parameters of the power grid fault inversion model and the power grid fault knowledge graph database;

[0104] S32. Simulate the power grid fault according to the matching result of the power grid fault knowledge graph to obtain the power grid fault simulation waveforms of different nodes in the power grid;

[0105] S33. Construct a power grid fault waveform feature matrix according to the power grid fault simulation waveforms, and input the power grid fault waveform feature matrix and the power grid topology adjacency matrix into a multi-layer GCN network connected in sequence to obtain the output matrix of the power grid fault inversion model parameters;

[0106] S34. Calculate the mean of the Pearson correlation coefficients between the power grid fault simulation waveforms of different nodes in the power grid and the actual action report waveforms as the loss function, and judge whether the loss function meets the requirements. If it does not meet the requirements, go to S35; if it meets the requirements, go to S36;

[0107] S35. Correct the parameters of the power grid fault inversion model using the output matrix of the power grid fault inversion model parameters, and return to execute the inversion model construction module;

[0108] S36. Obtain the final power grid fault inversion model.

[0109] More specifically, the power grid topology adjacency matrix represents the connection relationship between power grid topology nodes. Each row and each column are power grid topology nodes. Each element in the power grid topology adjacency matrix is represented by a number to indicate whether there is a connection between nodes. Each row of the power grid fault waveform feature matrix is the time point of the power grid fault simulation waveform, and each column is a power grid topology node. Each element in the power grid fault waveform feature matrix is the discretized data of the power grid fault simulation waveform corresponding to the time point corresponding to the power grid topology node.

[0110] More specifically, determining whether the loss function meets the requirements means that it meets the requirements when the value of the loss function reaches the minimum, otherwise it does not meet the requirements.

[0111] More specifically, the method of using the power grid fault inversion model parameter output matrix to correct the power grid fault inversion model parameters is to add the original power grid fault inversion model parameters and the parameters in the power grid fault inversion model parameter output matrix as the power grid fault inversion model parameters for the next stage, and return to execute the inversion model construction module.

[0112] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A knowledge graph-based power grid fault inversion dynamic correction method, characterized in that: include: S1. Build a knowledge graph library of power grid faults; S2, parsing the power grid fault message and using the parsing result as the power grid fault inversion model parameter to query and compare with the power grid fault knowledge graph library to build a power grid fault inversion model; S3, constructing a power grid fault inversion dynamic correction model to correct the power grid fault inversion model parameters to obtain the final power grid fault inversion model; S3 includes: S31. Establishing a power grid topology adjacency matrix using the power grid fault knowledge graph matching result; the power grid fault knowledge graph matching result refers to the matching result between the power grid fault inversion model parameters and the power grid fault knowledge graph library; S32, simulating a power grid fault according to the power grid fault knowledge graph matching result, and obtaining power grid fault simulation waveforms at different nodes in the power grid; S33, constructing a power grid fault waveform feature matrix according to the power grid fault simulation waveform, and inputting the power grid fault waveform feature matrix and the power grid topology adjacency matrix into a multi-layer GCN network connected in sequence to obtain a power grid fault inversion model parameter output matrix; S34, calculating the mean value of the Pearson correlation coefficient between the grid fault simulation waveform and the actual action report waveform at different nodes in the grid as the loss function, and judging whether the loss function meets the requirements, if not, going to S35, if meeting the requirements, going to S36; S35, using the power grid fault inversion model parameter output matrix to correct the power grid fault inversion model parameters, and returning to execute S2; S36. Obtain the final power grid fault inversion model.

2. According to the knowledge graph-based power grid fault inversion dynamic correction method of claim 1, it is characterized in that: S1 includes: The primary equipment of the power grid is taken as the entity of the power grid fault knowledge graph library, and the following classes are set: relay protection device information, power grid fault type, flow information, equipment information, and power grid fault information. Each class is set with multiple attributes to form a triple of entity-class-attribute. The historical data of the power grid is realized in one-to-one correspondence with entities in the form of the triple, thereby constructing a power grid fault knowledge graph library.

3. The power grid fault inversion dynamic correction method based on knowledge graph according to claim 1 is characterized in that S2 include: A bidirectional long short-term memory network is used to parse real-time power grid fault messages and historical power grid fault messages. The parsing results include the entity information of the primary equipment in the power grid, fault distance, power grid fault type, and transition resistance. The parsing results are used as the parameters of the power grid fault inversion model. The power grid fault inversion model parameters are matched one-to-one with the power grid fault knowledge graph library to complete the power grid fault knowledge graph matching. The power grid fault is simulated within the power grid subnet involved in the power grid fault according to the power grid fault knowledge graph matching results, thereby establishing a power grid fault inversion model.

4. A knowledge graph-based power grid fault inversion dynamic correction method according to claim 3, characterized in that: The real-time power grid fault message and the historical power grid fault message include a real-time SOE information message of a power grid fault and a real-time relay protection information management system action report, and a historical power grid fault SOE information message and a historical relay protection information management system action report.

5. The power grid fault inversion dynamic correction method based on knowledge graph according to claim 3 is characterized in that: The method of using a bidirectional long short-term memory network to analyze real-time power grid fault messages and historical power grid fault messages includes: The real-time power grid fault messages and historical power grid fault messages are annotated with BIO using the power grid fault knowledge graph, and the annotation results are input into the bidirectional long short-term memory network. The output results of the bidirectional long short-term memory network are input into the CRF layer, and the CRF layer outputs the parsing results.

6. The method for dynamic correction of power grid fault inversion based on knowledge graph according to claim 1, characterized in that: The power grid topology adjacency matrix represents the connection relationship between power grid topology nodes, each row and each column are power grid topology nodes, and each element in the power grid topology adjacency matrix uses a number to represent whether the nodes are connected or not. Each row of the power grid fault waveform feature matrix is ​​a time point of a power grid fault simulation waveform, and each column is a power grid topology node. Each element in the power grid fault waveform feature matrix is ​​the discretized data of the corresponding time point of the power grid fault simulation waveform corresponding to the power grid topology node.

7. The power grid fault inversion dynamic correction method based on knowledge graph according to claim 1 is characterized in that: The determination of whether the loss function meets the requirement means that the requirement is met when the loss function value reaches the minimum, otherwise the requirement is not met.

8. The power grid fault inversion dynamic correction method based on knowledge graph according to claim 1 is characterized in that: The method of using the power grid fault inversion model parameter output matrix to correct the power grid fault inversion model parameters is to add the original power grid fault inversion model parameters and the parameters in the power grid fault inversion model parameter output matrix as the power grid fault inversion model parameters of the next stage, and return to execute S2.

9. A knowledge graph-based power grid fault inversion dynamic correction system, characterized in that: include: A graph library construction module, used to construct a knowledge graph library for power grid faults; An inversion model building module is used to parse the power grid fault message and use the parsing result as the power grid fault inversion model parameter to query and compare with the power grid fault knowledge graph library to build a power grid fault inversion model; The inversion model correction module is used to construct a power grid fault inversion dynamic correction model to correct the power grid fault inversion model parameters to obtain the final power grid fault inversion model; the inversion model correction module is also used to: S31, using the power grid fault knowledge graph matching results to establish a power grid topology adjacency matrix; The power grid fault knowledge graph matching result refers to the matching result between the power grid fault inversion model parameters and the power grid fault knowledge graph library; S32, simulating a power grid fault according to the power grid fault knowledge graph matching result, and obtaining power grid fault simulation waveforms at different nodes in the power grid; S33, constructing a power grid fault waveform feature matrix according to the power grid fault simulation waveform, and inputting the power grid fault waveform feature matrix and the power grid topology adjacency matrix into a multi-layer GCN network connected in sequence to obtain a power grid fault inversion model parameter output matrix; S34, calculating the mean value of the Pearson correlation coefficient between the grid fault simulation waveform and the actual action report waveform at different nodes in the grid as the loss function, and judging whether the loss function meets the requirements, if not, going to S35, if meeting the requirements, going to S36; S35, using the power grid fault inversion model parameter output matrix to correct the power grid fault inversion model parameters, and returning to execute the inversion model construction module; S36. Obtain the final power grid fault inversion model.

Images