Fault diagnosis method, system and equipment for main water inlet valve of pumped storage power station and storage medium

Abstract

The invention provides a fault diagnosis method, system and device for a main water inlet valve of a pumped storage power station and a medium, and the method comprises the following steps: S1, obtaining fault information of the main water inlet valve of the pumped storage power station from a fault log record, executing an interactive problem on the fault information, analyzing the fault information, and assembling and storing the fault information into a fault knowledge base of the main water inlet valve; s2, performing structured unified representation on the analyzed information, and establishing a main water inlet valve fault knowledge graph; s3, related entities and edges in the fault knowledge graph are extracted to form a relation sub-graph; s4, realizing inference information based on fault knowledge graph information; and S5, acquiring monitoring data in real time, repeating the steps S1 to S4, updating the fault knowledge base and acquiring a fault diagnosis result. According to the method, the correlation degree is calculated by adopting the transfer evolution matrix in knowledge graph construction, the transitivity of parameters of the pumped storage power station is considered, and the overall updating of the matrix value is realized, so that the transfer relationship between nodes and edges is enhanced in knowledge graph calculation, and the relevance of structured representation is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electrical automation, and particularly relates to a method, a system, a device and a storage medium for diagnosing faults of a main inlet valve of a pumped storage power station. Background Art

[0002] With the rapid development of China's economy and society, the power load has increased rapidly, the peak-valley difference has been continuously enlarged, and the power grid has higher and higher requirements for stability. The insufficient peak regulation capacity will become an outstanding problem restricting the development of the power system. Pumped storage power stations, with their unique operating characteristics of peak regulation and valley filling, play the functions of regulating load, promoting energy conservation of the power system and maintaining the safe and stable operation of the power grid, and gradually become an effective and indispensable regulation means in China's power system. Against the background of the continuous increase in the installed capacity of pumped storage power stations, the structure of pumped storage units is becoming increasingly complex and the operating conditions are harsh, which makes the deterioration speed of pumped storage units accelerate and the probability of faults increase, putting forward higher requirements for the management, maintenance, monitoring and diagnosis of the units.

[0003] A pumped storage unit is a complex non-linear dynamic system, and the formation and development of faults during its operation have extremely strong randomness. In the existing technology, traditional fault diagnosis modeling theories and methods have long adopted time-frequency and space-time transformation analysis methods, and it is difficult to accurately describe a large number of uncertain factors mathematically, resulting in the actual condition-based maintenance system being difficult to obtain relatively accurate diagnostic conclusions, which greatly restricts the application of fault diagnosis theories and methods in engineering practice. Summary of the Invention

[0004] The first object of the present invention is to provide a method for diagnosing faults of a main inlet valve of a pumped storage power station in view of the above-mentioned problems.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for diagnosing faults of a main inlet valve of a pumped storage power station includes the following steps:

[0007] S1. Obtain the fault information of the main inlet valve of the pumped storage power station from the fault log record, perform interactive questions on the fault information and parse it, and compile and store it into the main inlet valve fault knowledge base;

[0008] S2. According to the main inlet valve fault knowledge base in step S1, perform a structured unified representation on the parsed information, establish a main inlet valve fault knowledge graph, and the structured unified representation includes a transfer evolution matrix H of the parsed information. The elements in the transfer evolution matrix H are the change coefficients of the information matrix after the structured unified representation in terms of time or space;

[0009] S3. According to the main inlet valve fault knowledge graph in step S2, extract the relevant entities and edges in the fault knowledge graph to form a relationship subgraph;

[0010] S4. According to the relationship subgraph obtained in step S3, establish an abstract Bayesian inference network to realize uncertain inference based on the information of the fault knowledge graph;

[0011] S5. Obtain monitoring data in real time, repeat steps S1 - S4, update the fault knowledge base and obtain the fault diagnosis result.

[0012] While adopting the above technical solutions, the present invention can also adopt or combine the following technical solutions:

[0013] As a preferred technical solution of the present invention: in step S2, the structured unified representation includes a triple structure of feature - attribute - fault form and a transfer evolution matrix H of the triple structure, and the transfer evolution matrix H is used to obtain the confidence of the triple structure relationship;

[0014] The feature represents the text name of the fault;

[0015] The attribute includes the relationship confidence parameter R b and an analysis information matrix W obtained from the feature and the fault based on a semantic parsing model. The relationship confidence R b parameter represents the correlation degree of the relationship in the triple structure, and the formula is as follows:

[0016]

[0017] In the formula, W is the analysis information matrix, H is the transfer evolution matrix, and m and n are the number of rows and columns of the matrix W;

[0018] The fault represents the text name of the fault type;

[0019] The transfer evolution matrix H has the same dimension as the analysis information matrix W.

[0020] As a preferred technical solution of the present invention: in step S2, the main inlet valve fault knowledge graph is a network structure, including nodes and edges. The nodes represent features and faults, and the edges represent the relationship strength R S , and the formula is as follows:

[0021]

[0022] In the formula, A1 represents the first activation function in two dimensions, A2 represents the second activation function in two dimensions, w ij is an element in the analysis information matrix W, h ij represents an element in the transfer evolution matrix H, and C ijis the standard value of the corresponding eigenvalue vector stored in the fault knowledge base, F ij is the standard value of the corresponding fault eigenvalue vector stored in the fault knowledge base, R b is the relationship confidence parameter.

[0023] As a preferred technical solution of the present invention: the network structure in the knowledge graph is a two-layer structure, including a pattern layer and a data layer,

[0024] The nodes in the pattern layer are the names of faults or features, and the training in the pattern layer represents the conditional probability of the content of the two nodes it links;

[0025] The nodes in the data layer are the data vector features of faults or features, and the edges in the data layer represent the statistical value of the corresponding situation of the content of the two nodes it links in all data.

[0026] As a preferred technical solution of the present invention: in step S4, the formula for obtaining the inference information based on the fault knowledge graph information is as follows:

[0027]

[0028] In the formula, F represents the fault type, and X1, X2... represent the features associated with the fault.

[0029] The second object of the present invention is to provide a main inlet valve fault diagnosis system for a pumped storage power station, including the following modules:

[0030] An input module, which is used to obtain the fault information recorded in the fault log;

[0031] An interaction module, which is used to parse the fault information in the fault log, perform semantic parsing, and store the information with the same semantic expression structure as the fault knowledge base into the fault knowledge base;

[0032] A knowledge graph training module, which is used to perform a knowledge graph structured expression on the information in the fault knowledge base to generate a fault diagnosis model, and the fault diagnosis model is used to receive real-time monitoring data and perform training to obtain a fault diagnosis result.

[0033] The knowledge graph training module further includes the following sub-modules:

[0034] A heterogeneous knowledge fusion module, which is used to uniformly represent various structural knowledge in the fault knowledge base;

[0035] A fault knowledge graph construction module, which is used to extract feature names and fault names from triple features to construct nodes of the knowledge graph, and form a two-layer knowledge graph with a network structure through the relationship confidence included in the triples and the edges calculated from the statistics of the triples;

[0036] A fault knowledge graph relationship sub-graph extraction module, which is used to extract the knowledge graph relationship sub-graph in the knowledge graph;

[0037] A fault graph inference module, which constructs an abstract Bayesian inference network based on the relationship sub-graph to deduce the probability of fault occurrence;

[0038] A main inlet valve knowledge graph update module, which is used to update the fault knowledge base.

[0039] The third object of the present invention is to provide an electronic device, which includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory complete mutual communication through the communication bus. Its characteristics are as follows:

[0040] A memory, which is used to store a computer program;

[0041] A processor, which is used to execute the computer program stored on the memory to implement the steps of the main inlet valve fault diagnosis method of the pumped storage power station as described above.

[0042] Another object of the present invention is to provide a non-volatile storage medium, in which an executable program is stored. When the executable program is executed by a processor, the steps of the main inlet valve fault diagnosis method of the pumped storage power station as described above are implemented.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1) In the structured construction of the knowledge graph of the present invention, the concept of a transfer evolution matrix is introduced to perform the calculation of relevance. During the actual operation of the pumped storage power station, relevant parameters have a certain transitivity, that is, when a value in the feature matrix changes, it does not actually mean the change of a single feature value, but the entire matrix value needs to be updated. In the current knowledge graph calculation, this kind of influence is not considered, which makes the calculation of nodes and edges further consider the transfer relationship between data, and significantly improves the relevance of the structured unified representation in the knowledge graph.

[0045] 2) Since the present invention introduces the evolutionary correlation in faults and characteristics, it can more comprehensively grasp the possible deterioration conditions, corresponding faults and characteristics of the main inlet valve, effectively improve the safe and stable operation level and the integrated operation and maintenance intelligent level of the pumped-storage unit, and can give a risk forecast of the main inlet valve fault in advance to avoid the occurrence of sudden faults. Description of the Drawings

[0046] Figure 1 It is a flowchart of the fault diagnosis method for the main inlet valve of the pumped-storage power station provided by the present invention.

[0047] Figure 2 It is a structural diagram for processing the fault knowledge of the main inlet valve.

[0048] Figure 3 It is a double-layer fault knowledge graph.

[0049] Figure 4 It is a schematic diagram for extracting the relationship subgraph.

[0050] In the figure: 1 - data layer of the fault knowledge graph; 2 - schema layer of the fault knowledge graph; 3 - relationship subgraph extracted from the schema layer of the fault knowledge graph. Detailed Embodiment

[0051] The present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0052] As Figure 1 shown, a fault diagnosis method for the main inlet valve of a pumped-storage power station specifically includes the following steps:

[0053] S1. Obtain the fault information of the main inlet valve of the pumped-storage power station from the fault log records, perform interactive questions on the fault information and parse it, and compile and store it in the main inlet valve fault knowledge base;

[0054] Through research and analysis of the existing fault records, equipment structure, and detection data of the pumped-storage power station, a series of characteristics and attributes related to the main inlet valve fault are formulated. Currently, through the collation of materials such as fault compilation, power station design drawings, and power station monitoring system point tables, they are compiled into the main inlet valve fault knowledge base.

[0055] S2. According to the main inlet valve fault knowledge base in step S1, perform a structured unified representation on the parsed information, establish a main inlet valve fault knowledge graph, and the structured unified representation includes the transfer and evolution matrix H of the parsed information. The elements in the transfer and evolution matrix H are the change coefficients of the information matrix after structured unified representation in terms of time or space;

[0056] The fault knowledge base contains rich fault knowledge of various structures, as well as monitoring data of different structures, and even some human judgment conclusions that cannot be monitored online. Through heterogeneous knowledge fusion, multi-source heterogeneous knowledge is structurally unified, and then a fault knowledge graph of the main inlet valve is established based on the unified knowledge.

[0057] The structural unified representation includes a triple structure of feature-attribute-fault form and a transfer evolution matrix H of the triple structure. The transfer evolution matrix H is used to obtain the confidence level of the triple structure relationship.

[0058] The feature represents the text name of the fault.

[0059] For structured data features, the original value can be directly used, and it also includes traditional feature values obtained by traditional calculation methods, as well as deep feature vectors calculated by a deep feature extraction model. The mean value of the feature vectors calculated from all training data corresponding to the corresponding fault is used as the standard value of the feature vector, and the standard value of the feature vector is used as the numerical description of the deep feature. Its relationship confidence level is obtained from the formula as follows:

[0060]

[0061] In the formula, N represents the amount of data available for calculating this feature, and N0 represents the determined data amount basis, which is preferably more than 100.

[0062] For unstructured data features, there are two cases: the first case is the status value, such as "bypass valve opening signal", and its numerical description uses 0,1 variables to indicate whether the status is satisfied; the second is the threshold value, such as "the leakage water volume of the ball valve sealing input cavity is greater than the required value", and its numerical description uses an inequality to represent the size relationship with the threshold value, such as "the leakage water volume of the ball valve sealing input cavity is greater than the required value > 0.95L / min". Its relationship confidence level is initially determined through expert discussion, and a corresponding probability value is assigned to represent the confidence level of the triple relationship extracted in this statement. For example, "accidental" is assigned 0.001, "inevitable" is assigned 1, and the default value is assigned 0.8. The above fault knowledge processing structure is as Figure 2 shown.

[0063] The attribute contains the relationship confidence parameter R b and the parsing information matrix W obtained from the feature and the fault based on the semantic parsing model. The relationship confidence R b The parameter represents the correlation degree of the relationship in the triple structure, and the formula is as follows:

[0064]

[0065] In the formula, W is the parsing information matrix, H is the transfer evolution matrix, and m and n are the number of rows and columns of the matrix W.

[0066] The fault represents the text name of the fault type;

[0067] For example, "Base seat displacement - (0.48, >50mm) - Main valve self-excited vibration", "Main valve vibration frequency - (0.96, >360ms) - Main valve self-excited vibration", and "Highest pressure value of penstock - (0.001, >4.55MPa) - Main valve seal fault".

[0068] The transfer evolution matrix H has the same dimension as the analytical information matrix W. The transfer evolution matrix H is a matrix composed of multiple evolution coefficients with the same dimension as the information matrix W. The matrix values can be obtained from empirical constants or are variable constants calculated by functions related to time and space. Thus, by multiplying the eigenvalue matrix by the transfer evolution matrix, the fault transfer evolution relationship under each fault type can be obtained, and a more accurate eigenvalue matrix can be obtained. The rank between the multiplied eigenvalue matrices can express the correlation degree of the relationship, and this is encapsulated as an attribute parameter in the attribute.

[0069] The knowledge graph of the main inlet valve fault is a network structure, including nodes and edges. The nodes represent features and faults, and the edges represent the relationship strength R S , and the feature names and fault names are directly extracted from the triple features of feature - attribute - fault form to construct the nodes of the knowledge graph. The edge is the relationship strength between nodes, that is, the concurrent probability of node activation, which is calculated through the relationship confidence contained in the attribute in the triple and the statistic of the triple. The formula is as follows:

[0070]

[0071] In the formula, A1 represents the first activation function in two dimensions, A2 represents the second activation function in two dimensions, w ij is an element in the analytical information matrix W, h ij represents an element in the transfer evolution matrix H, C ij is the standard value of the corresponding eigenvalue vector saved in the fault knowledge base, F ij is the standard value of the corresponding fault eigenvalue vector saved in the fault knowledge base, R b is the relationship confidence parameter.

[0072] The first activation function is the hyperbolic tangent function, and the second activation function is the Softmax function.

[0073] S3. According to the knowledge graph of the main inlet valve fault in step S2, relevant entities and edges in the fault knowledge graph are extracted through a deep feature extraction model to form a relationship subgraph;

[0074] Define the activation function of the edge according to the numerical description of the feature. The numerical description includes: status value, threshold value, and standard value, and the specific definition of the activation function.

[0075] As Figures 3 - 4 shown, the network structure in the knowledge graph is a two-layer structure, including a pattern layer and a data layer.

[0076] The nodes in the pattern layer are the names of faults or features, and the training in the pattern layer represents the conditional probability of the content of the two nodes it links.

[0077] The nodes in the data layer are the data vector features of faults or features, and the edges in the data layer represent the statistical values of the corresponding situations of the content of the two nodes it links in all data.

[0078] The pattern layer is constructed based on text data features, and the data layer is constructed based on vibration data features and text data features. The traditional feature nodes in the data layer and the text feature nodes with the same meaning jointly correspond to the corresponding nodes in the pattern layer. For example, the "oscillation value" and "oscillation limit exceeded" in the data layer jointly correspond to the "oscillation limit exceeded" in the pattern layer. The data layer statistically calculates the relationship strength between the data volume and the edges and then updates the relationship strength of the edges in the pattern layer. When the accuracy of the calculation model of the vibration data features reaches the limit condition P b When, update the corresponding nodes and edges in the pattern layer, P b It is advisable to set it to 0.8 - 1.

[0079] The purpose of knowledge graph reasoning diagnosis is to find the corresponding equipment faults based on the existing knowledge graph and real-time state features. The reasoning process obtains the corresponding feature entities and attribute values in the graph through a deep learning model and traditional feature extraction, directly obtains the numerical data, status values, and over-limit values of other detection devices, and obtains deep features by performing feature extraction on the numerical data through a deep extraction model. Using the deep features, status values, and over-limit values as targets, retrieve their names in the knowledge graph pattern layer and extract the relevant entities and edges in the graph to form a subgraph.

[0080] S4. According to the relationship subgraph obtained in step S3, establish an abstract Bayesian inference network, and use the Bayesian chain rule, Bayes' theorem, and the principle of conditional independence to achieve uncertain reasoning based on the information of the fault knowledge graph.

[0081] The Bayesian rule can be expressed as:

[0082]

[0083] Derive the formula for obtaining inference information based on the information of the fault knowledge graph from the Bayesian rule as follows:

[0084]

[0085] In the formula, F represents the fault type, and X1, X2... represent the features associated with the fault. The probability of the fault can be deduced based on the information of the edges in the knowledge graph, and the possibility of various faults can be judged according to the magnitude of the probability value.

[0086] S5. Obtain the monitoring data in real time, repeat steps S1 - S4, update the fault knowledge base, and obtain the fault diagnosis result.

[0087] To update the fault diagnosis knowledge graph, the fault knowledge base needs to be updated first. The update of the fault knowledge base is divided into two aspects. On the one hand, it is through the data import and analysis of the production real-time monitoring platform, and on the other hand, an artificial input interface is provided. The updated data is fused to obtain the knowledge triple form that can be directly used in the knowledge graph. For the existing nodes in the graph, data statistics can be directly carried out and updated; for the newly added data, the confidence levels of the original detection data and the unstructured data input manually are set manually, and the structured data deep feature extraction model is trained with the updated data on the original basis, and the model classification accuracy test value is used as the confidence level. When the confidence level reaches the limit value P b At this time, the structure is updated to the pattern layer, including the feature nodes of the model and the edges connected to them. Calculate the relationship strength R' of all edges with the updated data S , and then perform weighted averaging with the relationship strength of the edges in the pattern layer to obtain the new relationship strength value R S , which is the updated relationship strength, and its formula is as follows:

[0088]

[0089] Among them represents the data volume corresponding to the fault f in the updated data, R ij represents the relationship strength of the edges in the original pattern layer, and A represents any activation function. S

[0090] Through the above method, an open and extensible fault diagnosis knowledge base framework is established, the fault diagnosis knowledge base and the fault sample standard library are accumulated and updated in real time, a fault tree diagnosis model based on fault reasoning is constructed, the early potential fault warning and intelligent diagnosis of the pumped storage power station unit are realized, providing theoretical guidance and technical support for the formulation of the main inlet valve maintenance strategy, and ensuring the safety and reliability of the operation of the main inlet valve.

[0091] In a specific factual manner of a main inlet valve application, the system of the present invention for implementing the fault diagnosis method of the main inlet valve of a pumped storage power station includes a main inlet valve fault data information acquisition module, a main inlet valve fault tree construction module, and a main inlet valve fault tree analysis module; the main inlet valve fault data information acquisition module, the main inlet valve fault tree construction module, and the main inlet valve fault tree analysis module are connected in series in sequence; the main inlet valve fault data information acquisition module is used to acquire the fault data information of the main inlet valve of the pumped storage power station and compile the main inlet valve fault knowledge base; the main inlet valve fault tree construction module is used to classify the main inlet valve fault diagnosis knowledge base, define the top event, intermediate events, and bottom events, and construct a tree-shaped main inlet valve fault tree step by step; the main inlet valve fault tree analysis module is used to perform qualitative analysis, quantitative analysis, and bottom event probability determination on the main inlet valve fault tree in combination with the main inlet valve body data, unit operation data, and oil pressure device data to determine the fault warning and intelligent diagnosis of the main inlet valve.

[0092] The present invention also provides a fault diagnosis system for the main inlet valve of a pumped storage power station, including the following modules:

[0093] An input module, which is used to acquire the fault information recorded in the fault log.

[0094] An interaction module, which is used to parse the fault information in the fault log, perform semantic parsing to obtain information with the same semantic expression structure as the fault knowledge base and store it in the fault knowledge base, and is also used to generate a dialog box for interaction and confirmation with the user to further correct the expression of the fault information.

[0095] A knowledge graph training module, which is used to perform a knowledge graph structured expression on the information in the fault knowledge base to generate a fault diagnosis model, and the fault diagnosis model is used to receive real-time monitoring data and perform training to obtain a fault diagnosis result.

[0096] The knowledge graph training module further includes the following sub-modules:

[0097] A heterogeneous knowledge fusion module, which is used to uniformly represent various structural knowledge in the fault knowledge base.

[0098] A fault knowledge graph construction module, which is used to extract the feature names and fault names from the triple features to construct the nodes of the knowledge graph, and form a two-layer knowledge graph with a network structure through the edges calculated from the relationship confidence in the triples and the statistics of the triples.

[0099] A fault knowledge graph relationship sub-graph extraction module, which is used to extract the knowledge graph relationship sub-graph from the knowledge graph.

[0100] The fault graph inference module constructs an abstract Bayesian inference network based on the relational sub-graph to deduce the probability of fault occurrence;

[0101] The main inlet valve knowledge graph update module is used to update the fault knowledge base.

[0102] The present invention also provides an electronic device, which includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory complete mutual communication through the communication bus.

[0103] The memory is used to store a computer program;

[0104] The processor is used to execute the computer program stored on the memory to implement the steps of the fault diagnosis method for the main inlet valve of the pumped-storage power station as described above.

[0105] The present invention also provides a non-transitory readable storage medium, which is a non-volatile storage medium. The non-volatile storage medium stores an executable program. When the executable program is executed by the processor, it implements the steps of the fault diagnosis method for the main inlet valve of the pumped-storage power station as described above.

[0106] So far, the technical solution of the present invention has been described in combination with the specific experimental process shown in the drawings. However, the protection scope of the present invention is not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for diagnosing faults of a main water inlet valve of a pumped storage power station, characterized in that: The steps include: S1. Obtain the fault information of the main water inlet valve of the pumped storage power station from the fault log record, perform interactive questions on the fault information and parse it, and compile and store it into the main water inlet valve fault knowledge base; S2. According to the main water inlet valve fault knowledge base in step S1, the parsed information is represented in a structured unified manner to establish a main water inlet valve fault knowledge graph, wherein the structured unified representation includes a transfer evolution matrix H of the parsed information, and the elements in the transfer evolution matrix H are the variation coefficients of the information matrix after the structured unified representation in time or space; S3, according to the main water inlet valve fault knowledge graph in step S2, extract the relevant entities and edges in the fault knowledge graph to form a relationship subgraph; S4. According to the relationship subgraph obtained in step S3, an abstract Bayesian reasoning network is established to realize uncertain reasoning based on fault knowledge graph information; S5. Acquire monitoring data in real time, repeat steps S1-S4, update the fault knowledge base and obtain fault diagnosis results.

2. The method according to claim 1, characterized in that: In step S2, the structured unified representation includes a triple structure in the form of feature-attribute-fault and a transfer evolution matrix H of the triple structure, and the transfer evolution matrix H is used to obtain the confidence of the triple structure relationship; The feature represents the text name of the fault; The attributes include the relationship confidence parameter R b and the parsing information matrix W obtained by the features and the fault according to the semantic parsing model, the relationship confidence R b The parameter represents the degree of correlation of the relationship in the triple structure, and the formula is as follows: Where W is the analytical information matrix, H is the transfer evolution matrix, m and n are the number of rows and columns of the matrix W; The fault represents the text name of the fault type; The transfer evolution matrix H has the same dimension as the analytical information matrix W.

3. The method according to claim 1, characterized in that: In step S2, the main water inlet valve fault knowledge graph is a network structure, including nodes and edges, wherein the nodes represent features and faults, and the edges represent the relationship strength R between nodes. S , the formula is as follows: In the formula, A1 represents the first two-dimensional activation function, A2 represents the second two-dimensional activation function, and w ij To analyze the elements in the information matrix W, h ij represents the elements in the transfer evolution matrix H, C ij is the standard value of the corresponding eigenvalue vector stored in the fault knowledge base, F ij is the standard value of the corresponding fault feature value vector stored in the fault knowledge base, R b is the relationship confidence parameter.

4. The method according to claim 3, characterized in that: The network structure in the knowledge graph is a two-layer structure, including a model layer and a data layer. The nodes in the pattern layer are names of faults or features, and the training in the pattern layer represents the conditional probability of the contents of the two nodes to which it is linked; The nodes in the data layer are data vector features of faults or features, and the edges in the data layer represent the statistical values ​​of the corresponding situations of the contents of the two nodes linked by it in all data.

5. The method according to claim 1, characterized in that: In step S4, the formula for obtaining the inference information based on the fault knowledge graph information is as follows: In the formula, F represents the fault type, and X1, X2, ... represent the features associated with the fault.

6. A pumped storage power station main water inlet valve fault diagnosis system, characterized in that: Includes the following modules: An input module, the input module is used to obtain fault information recorded in the fault log; An interactive module, the interactive module is used to parse the fault information in the fault log, perform semantic analysis, and obtain information with the same semantic expression structure as the fault knowledge base and store it in the fault knowledge base; A knowledge graph training module is used to perform knowledge graph structured expression on the information in the fault knowledge base to generate a fault diagnosis model, and the fault diagnosis model is used to receive real-time monitoring data and perform training to obtain fault diagnosis results.

7. A pumped storage power station main water inlet valve fault diagnosis system, characterized in that: The knowledge graph training module also includes the following submodules: A heterogeneous knowledge fusion module, which is used to unify the representation of various structural knowledge in the fault knowledge base; A fault knowledge graph construction module, which is used to extract feature names and fault names from triple features to construct nodes of the knowledge graph, and to construct a double-layer knowledge graph of a network structure through edges obtained by calculating the relationship confidence contained in the triples and the statistics of the triples; A fault knowledge graph relationship subgraph extraction module, wherein the fault knowledge graph relationship subgraph extraction module is used to extract a knowledge graph relationship subgraph in the knowledge graph; A fault graph reasoning module, which is based on a relational subgraph and constructs an abstract Bayesian reasoning network to deduce the probability of a fault occurring; A main water inlet valve knowledge graph updating module, wherein the main water inlet valve knowledge graph updating module is used to update a fault knowledge base.

8. An electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, characterized in that: A memory, the memory being used to store a computer program; A processor, wherein the processor is used to execute a computer program stored in a memory to implement the steps of a method for diagnosing a main water inlet valve fault in a pumped storage power station as described in any one of claims 1 to 5.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores an executable program, and when the executable program is executed by the processor, the steps of the method for diagnosing faults of the main water inlet valve of a pumped storage power station are implemented as claimed in any one of claims 1 to 5.

Images