Method and system for constructing fault diagnosis knowledge graph of auxiliary equipment of hydropower plant

Abstract

The application provides a kind of hydroelectric plant auxiliary equipment fault diagnosis knowledge graph construction method and system, belong to knowledge graph technical field.The method includes: obtaining the multi-source text data of hydroelectric plant auxiliary equipment and pre-processing, form multi-source data set;Based on the multi-source data set, entity and relationship are identified, and a basic semantic triple set is constructed;For each entity in the basic semantic triple set, the operating state vector of the entity is constructed;According to the operating state vector, the rationality of each basic semantic triple in the basic semantic triple set is scored, and a scoring result is obtained;According to the scoring result, the basic semantic triple is screened, and a knowledge triple set is obtained;According to the knowledge triple set, a fault diagnosis knowledge graph is constructed.The knowledge graph constructed by the application integrates static semantic relationship and dynamic operating state, and can provide a high credibility knowledge base for subsequent fault reasoning.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to a method and system for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants. Background Technology

[0002] In the operation and maintenance of auxiliary equipment in hydropower plants, fault diagnosis and knowledge management have long relied on manual experience, traditional signal processing methods, and document- and rule-based knowledge base systems. However, the complex structure and variable operating environment of auxiliary equipment, along with the interrelationships between parameters of its components, lead to significant differences in the distribution of operating parameters such as temperature, pressure, and current under different operating conditions. Traditional fault detection methods based on fixed thresholds or steady-state assumptions often experience a significant decrease in reliability when operating conditions change, making them difficult to adapt to dynamic operating environments.

[0003] In recent years, knowledge graph technology has been introduced into the field of equipment knowledge management to organize semantic information such as equipment, components, fault phenomena, and operating parameters, supporting structured queries and basic reasoning. However, existing knowledge graph constructions mostly focus on extracting entities and relationships from text to form static triple structures, failing to effectively incorporate dynamic information and express dynamic relationships dependent on operating conditions. Regarding graph quality assessment, existing methods primarily rely on static indicators such as semantic similarity, statistical frequency, and expert rules to judge the credibility of triples, lacking a unified dynamic evaluation mechanism. Due to the absence of state constraints, existing graphs struggle to accurately identify pseudo-relationships and cannot support high-confidence graph completion and adaptive updates.

[0004] In conclusion, the current operation and maintenance of auxiliary equipment in hydropower plants still needs to explore intelligent methods that can deeply integrate real-time operational data and conduct unified dynamic evaluations in terms of fault diagnosis and knowledge management, so as to improve the adaptability of fault detection and the effectiveness of knowledge management. Summary of the Invention

[0005] This invention provides a method and system for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants. It addresses the shortcomings of existing fault diagnosis knowledge graphs in effectively expressing dynamic correlations dependent on operating conditions, achieves deep integration of dynamic operating data and static semantic information, and improves the reliability of the fault diagnosis knowledge graph through a dynamic scoring mechanism.

[0006] In a first aspect, the present invention provides a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant, comprising: Multi-source text data of auxiliary equipment in a hydropower plant is acquired and preprocessed to form a multi-source dataset; wherein, the multi-source text data includes at least the equipment's operating status data; Based on the multi-source dataset, entity and relation recognition is performed, and a basic semantic triplet set is constructed. For each entity in the set of basic semantic triples, an operational state vector of the entity is constructed based on the operational state data corresponding to the entity. Based on the running state vector, the rationality of each basic semantic triple in the set of basic semantic triples is scored to obtain the scoring result; Based on the scoring results, basic semantic triples are filtered to obtain a set of knowledge triples. A fault diagnosis knowledge graph is constructed based on the set of knowledge triples.

[0007] According to the present invention, a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant includes, in the step of scoring the rationality of each basic semantic triplet in the set of basic semantic triplets based on the operating state vector, the method includes: For each basic semantic triple in the set of basic semantic triples, calculate the semantic relevance between the head entity and the tail entity in the basic semantic triple; Based on the running state vectors corresponding to the head entity and the tail entity respectively, calculate the state consistency and trend difference between the head entity and the tail entity; wherein, the state consistency characterizes the degree of difference between the running state data of the head entity and the tail entity, and the trend difference characterizes the degree of difference between the trends of the running state data of the head entity and the tail entity over time. Based on the semantic relevance The consistency of the state and the difference of the trend are used to score the rationality of the basic semantic triples, and the score result is obtained.

[0008] According to the present invention, a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant includes calculating the state consistency and trend difference between the head entity and the tail entity based on the operating state vectors corresponding to the head entity and the tail entity, respectively. Based on the running state vectors corresponding to the head entity and the tail entity respectively, the temporal state vectors of the head entity and the tail entity in a preset time period are extracted by a temporal convolutional network. Based on the difference between the temporal state vectors, the state consistency between the head entity and the tail entity is calculated. Based on the running state vectors corresponding to the head entity and the tail entity, the trend features of the head entity and the tail entity in a preset time period are extracted by the sliding window difference method. Based on the differences between the trend features, the trend difference between the head entity and the tail entity is calculated.

[0009] According to the present invention, a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant is provided. The rationality of the basic semantic triples is scored based on the semantic relevance, the state consistency, and the trend difference to obtain a scoring result, including: Based on the state consistency and the trend difference, the confidence factor of the basic semantic triple is determined; wherein, the confidence factor characterizes the degree of matching of the running state data between the head entity and the tail entity of the basic semantic triple and the reliability of the running state data. The rationality of the basic semantic triples is scored based on the semantic relevance, state consistency, trend difference, and confidence factor to obtain the scoring results.

[0010] According to the present invention, a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant is provided, wherein the basic semantic triples are filtered based on the scoring results to obtain a set of knowledge triples, including: From the set of basic semantic triples, select basic semantic triples whose scores are greater than a preset semantic relevance threshold and whose confidence factors are greater than a preset confidence threshold, and obtain multiple selected basic semantic triples to form a knowledge triple set.

[0011] According to the method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant provided by the present invention, after the basic semantic triplet filtering based on the scoring results to obtain a knowledge triplet set, the method further includes: For missing triples in the basic semantic triple set that have missing entities, multiple candidate entities that meet preset conditions with the entities in the missing triples are selected from the basic semantic triple set to obtain a candidate entity set; wherein, the preset conditions include state consistency less than a predetermined state similarity threshold and trend difference less than a predetermined trend similarity threshold. The reasonableness of each candidate entity in the candidate entity set and the candidate triple formed by the missing triple is scored, and the confidence factor of the candidate triple is calculated. Select multiple candidate triples with the highest scores and confidence factors greater than the preset confidence threshold; Multiple candidate triples selected are added to the knowledge triple set.

[0012] According to the present invention, a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant, wherein the pre-determined state similarity threshold is determined by taking the mean of the average distance within the cluster after clustering the running state vectors corresponding to all entities in the plurality of selected basic semantic triples. The predetermined trend similarity threshold is determined based on the distribution of the trend differences.

[0013] Secondly, the present invention also provides a knowledge graph construction system for fault diagnosis of auxiliary equipment in hydropower plants, comprising: The data acquisition module is used to acquire multi-source text data of auxiliary equipment in hydropower plants and preprocess it to form a multi-source dataset; wherein, the multi-source text data includes at least the equipment's operating status data; The triplet construction module is used to identify entities and relationships based on the multi-source dataset and construct a basic semantic triplet set. The rationality assessment module is used to construct an operational state vector for each entity in the set of basic semantic triples based on the operational state data corresponding to the entity; and to score the rationality of each basic semantic triple in the set of basic semantic triples based on the operational state vector to obtain a score result. The triplet filtering module is used to filter basic semantic triplets based on the scoring results to obtain a set of knowledge triplets. The knowledge graph construction module is used to construct a fault diagnosis knowledge graph based on the set of knowledge triples.

[0014] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the knowledge graph construction method for fault diagnosis of hydropower plant auxiliary equipment as described above.

[0015] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the knowledge graph construction method for fault diagnosis of hydropower plant auxiliary equipment as described above.

[0016] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the knowledge graph construction method for fault diagnosis of hydropower plant auxiliary equipment as described above.

[0017] The beneficial effects of the technical solutions provided by some embodiments of the present invention include at least the following: 1) The present invention provides a method and system for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants. By incorporating dynamic information such as the real-time operating status of equipment into the knowledge graph structure, the present invention achieves deep integration of dynamic operating data and static semantic information. This can effectively represent the changes in the operating status of the same semantic relationship at different operating stages, enabling the knowledge graph to express dynamic correlations dependent on operating conditions, and significantly enhancing the consistency and interpretability of knowledge in the physical world. Based on this, the present invention dynamically quantifies and evaluates the rationality of each basic semantic triple according to the operating status vector of each entity, and filters the basic semantic triples based on the evaluation results, thereby obtaining a higher quality set of knowledge triples and improving the reliability of the constructed fault diagnosis knowledge graph.

[0018] 2) This invention designs a dynamic scoring mechanism, which calculates the semantic relevance, state consistency, and trend difference between the head and tail entities for each basic semantic triple. These indicators are used to dynamically quantify the semantic consistency, state matching degree, data change trend, and uncertainty between the head and tail entities. Finally, these indicators are combined to quantitatively evaluate the rationality and confidence of the basic triple from multiple dimensions, and to achieve effective screening of triples, thereby constructing a high-quality knowledge graph that can be adaptively updated according to changes in working conditions.

[0019] 3) This invention uses preset semantic relevance thresholds and preset confidence thresholds to screen basic semantic triples from both semantic and state dimensions. This can automatically eliminate unreasonable or redundant relationships, thereby improving the quality of the knowledge triple set and the final knowledge graph.

[0020] 4) For triples with missing entities, this invention determines the candidate entity set through state consistency index and trend difference index, scores the rationality of the candidate triples and calculates their confidence factor, and can predict potential reasonable associations in the case of missing information, thereby achieving automatic completion of potential missing knowledge and maintaining the accuracy, coverage and real-time adaptability of the knowledge graph. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is one of the flowcharts illustrating a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant, provided by the present invention. Figure 2This is the second flowchart of a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant, provided by the present invention. Figure 3 This is a schematic diagram of the structure of a knowledge graph construction system for fault diagnosis of auxiliary equipment in hydropower plants provided by the present invention; Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] Hydropower plant auxiliary systems generate a large amount of operational logs, maintenance records, procedures, and parameter measurement data during long-term operation. This data contains rich experiential knowledge and information related to operating conditions. However, existing technologies have significant limitations in utilizing this information to construct knowledge graphs suitable for fault diagnosis of hydropower plant auxiliary equipment.

[0025] First, this information is scattered, heterogeneous, and lacks structured expression. Traditional knowledge bases are built only based on semantic triples, which cannot incorporate dynamic information such as device operating status data into the graph structure. The same semantic relationship cannot reflect the change in its causal strength at different operating stages, which makes the knowledge unable to play a role in real-time scenarios. Secondly, there is a lack of automatic completion capabilities for potentially missing knowledge. Most existing completion methods for missing semantic triples rely mainly on semantic similarity or statistical rules, neglecting state constraints and uncertainty control. As a result, they are prone to introducing incorrect completions in high-uncertainty regions, affecting the quality of the graph. Finally, the existing knowledge graph construction process lacks a quantitative evaluation mechanism for the rationality of ternary combinatorial relationships, often resulting in redundant, erroneous, or semantically correct but functionally mismatched associations. This makes it difficult for knowledge graphs to achieve reliable adaptive evolution and ensure consistency with the physical world.

[0026] Therefore, this invention proposes a method and system for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants. By introducing the operating status data of equipment or components and using reasonable evaluation methods to quantify the credibility and manage the uncertainty of triples, the accuracy, interpretability and online maintainability of the knowledge graph are improved.

[0027] Example 1 Please see Figure 1 ,Figure 1 One of the flowcharts for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant, provided as an embodiment of the present invention, includes: S101. Obtain multi-source text data of auxiliary equipment in hydropower plants and preprocess it to form a multi-source dataset; S102. Entity and relation recognition based on multi-source datasets, and construction of a basic semantic triplet set; S103. For each entity in the set of basic semantic triples, construct the entity's running state vector based on the running state data corresponding to the entity. S104. Based on the running state vector, score the rationality of each basic semantic triple in the basic semantic triple set and obtain the score result. S105. Based on the scoring results, basic semantic triples are filtered to obtain a set of knowledge triples. S106. Construct a fault diagnosis knowledge graph based on the set of knowledge triples.

[0028] This invention incorporates dynamic information such as the real-time operating status of equipment into a knowledge graph structure, which can effectively represent the changes in the operating status of the same semantic relationship at different operating stages. This enables the knowledge graph to express dynamic correlations dependent on operating conditions, significantly enhancing the consistency and interpretability of knowledge in the physical world. Based on this, this invention dynamically quantifies and evaluates the rationality of each basic semantic triple according to the operating status vector of each entity, and filters the basic semantic triples based on the evaluation results, resulting in a higher quality set of knowledge triples, thereby improving the reliability of the constructed fault diagnosis knowledge graph.

[0029] In step S101 of this embodiment, structured and unstructured data sources are collected from the hydropower plant auxiliary system to form multi-source text data, which is then preprocessed to obtain a multi-source dataset.

[0030] Step S101 may include the following sub-steps: S101-1. Obtain multi-source text data. In this embodiment, multi-source text data may include: equipment information in the hydropower plant auxiliary system, equipment operating status data (i.e., real-time monitoring data of equipment operation), inspection and maintenance record documents, and may also include fault reports, fault logs, system log records, and other documents.

[0031] S101-2. Preprocess the multi-source text data.

[0032] S101-2-1. Perform preprocessing such as data cleaning on the equipment's operating status data, including outlier filtering and missing value imputation.

[0033] Specifically, ① for stable parameters (such as bearing temperature, vibration, motor current, etc.), an anomaly detection method based on interquartile range (IQR) is used to eliminate occasional outliers.

[0034] For example: A circulating water pump collects a 10-minute temperature sequence (°C) on a certain day as [50, 51, 52, 51, 49, 50, 51, 87, 50]. Then the first quartile Q1 = 50, the third quartile Q3 = 51.5, and the interquartile range IQR = 1.5. Therefore, the reasonable interval is [50...]. [1.5×1.5, 51.5 + 1.5×1.5] = [47.75, 53.75], but the 87℃ in this temperature sequence is clearly out of range, so this outlier is removed.

[0035] ② For non-stationary time series data with trends, periods, or multiple operating modes (such as cooling water temperature, flow rate, pressure, etc.), the rolling window Z-score method can be used for anomaly detection.

[0036] For example, the time series data of the outlet temperature change of a generator stator cooler during a certain period is [42, 43, 44, 46, 48, 49, 50, 51, 53, 52]. If IQR is used, 53℃ will be misjudged as an anomaly, but it is actually a normal temperature rise caused by high load. Therefore, the rolling window Z-score method is used for anomaly detection. The window size can be window=5, and the window mean u≈47 and the window standard deviation σ≈3 are calculated. The Z-score corresponding to 53℃ is Z(53) = (53 47) / 3 ≈ 2.0, let the Z-score threshold be Z. thresh =3, Z(53) <Z thresh If the temperature is 53℃, then it is considered a normal fluctuation and will not be excluded.

[0037] ③ For short-term missing data (e.g., <10 minutes) of equipment, the Locally Weighted Scatterplot Smoothing (LOESS) method can be used to impute missing values ​​in the time series. For example, as shown in Table 1, the temperature (°C) monitoring data of a certain cooler shows periodic fluctuations: Table 1 Temperature monitoring data of a certain cooler

[0038] The temperature value at 10:02 was missing. The missing value was estimated to be approximately 49.5℃ using the LOESS method based on the trend of the neighborhood data. This missing value was then used to fill in the missing value.

[0039] ④ For long-term missing data of the device (e.g., >=10 minutes), the Transformer prediction model can be used to estimate and impute missing values.

[0040] For example, if the cooling water flow rate of a generator is missing for 10 consecutive minutes, the historical generator cooling water flow rate sequence (flow rate sequence of the past 30 minutes) is input into the Transformer prediction model. The model can predict the average value of the missing interval and fill the missing interval with the average value.

[0041] S101-2-2. The following cleaning strategy can be adopted for inspection and maintenance record documents: a) Text normalization processing.

[0042] The same meaning may be written in different ways, so it is necessary to normalize the synonyms of professional terms. Table 2 shows an example of text normalization.

[0043] Table 2 Examples of Text Normalization Processing

[0044] As shown in Table 2, different original Chinese text expressions can be normalized into English representations, and a unified encoding method can be used to ensure the uniqueness of knowledge graph entities.

[0045] b) Regularized field extraction.

[0046] For example, an original maintenance record reads "#3 cooling water pump (number: P3-001) vibration 6.5mm / s, bearing model 6205, replaced on 2024-06-10", and Table 3 shows its regularized extraction results.

[0047] Table 3. Examples of Regularized Field Extraction

[0048] S101-3. Construct a multivariate dataset by combining expert experience.

[0049] The collected and processed data, combined with expert experience, form a basic equipment information dataset, an operational status dataset, and an inspection and maintenance text dataset. These datasets can be stored in CSV (Comma-Separated Values) format.

[0050] (1) Equipment basic information dataset, as shown in Table 4, is a partial example of the equipment basic information dataset.

[0051] Table 4 shows some examples of the equipment basic information dataset.

[0052] (2) Running status dataset, as shown in Table 5, is a partial example of the running status dataset.

[0053] Table 5 shows some examples of the running status dataset.

[0054] (3) Inspection and maintenance text dataset, as shown in Table 6, are some examples of inspection and maintenance text datasets.

[0055] Table 6 shows some examples of the inspection and maintenance text dataset.

[0056] In step S102 of this embodiment, based on the multi-source dataset obtained in step S101, entities and relations are extracted, basic semantic triples are constructed, and a set of basic semantic triples is formed.

[0057] Step S102 may specifically include the following sub-steps: S102-1, Identify all possible entities from the multi-source dataset.

[0058] Since the multi-source datasets are all in CSV format, for the equipment basic information dataset and the operating status dataset, field names can be directly mapped to entities. For records in the inspection and maintenance text dataset, an architecture or method based on a BERT pre-trained model can be used to automate the Named Entity Recognition (NER) task.

[0059] For example, suppose the inspection and maintenance text dataset contains the following content (Table 7): Table 7. Examples of some fields and values ​​in the inspection and maintenance text dataset.

[0060] The above methods can be used to identify entities such as: cooling water pump A, inlet filter, outlet pressure, pressure fluctuations, and blockages.

[0061] Extract all possible entities from the multi-source dataset, and finally output a unified entity set: E={e i}, e i Represents the i-th entity.

[0062] S102-2, Perform relation extraction.

[0063] Dependency tree analysis was performed on the text in the inspection and maintenance text dataset to extract subject-verb-object, modifier-type, and other structures, and to identify relationships such as "equipment-component", "equipment-fault", and "component-fault".

[0064] For example, taking the maintenance record "Cooling water pump A experiences pressure fluctuations, suspected to be due to inlet filter blockage" in Table 7 as an example, the relationship extraction results shown in Table 8 can be extracted.

[0065] Table 8. Example of relation extraction results

[0066] Extract all possible relations from multi-source datasets, and finally output a unified relation set: R = {r k}, r k This represents the k-th relation.

[0067] S102-3. Combine the identified entities and relations into basic semantic triples.

[0068] Specifically, based on expert experience, for each pair of entities (e i , e j Check if the extracted relation r exists. k If it exists, then construct the basic semantic triple (e i , r k , e j ); You can also set templates, such as (Device A, hasFault, Fault B). If "Device A has fault B" is identified, the basic semantic triple can be recorded according to the set template. For example, based on Tables 7 and 8, the following basic semantic triples can be extracted: (Cooling water pump A, hasFault, pressure fluctuation) (Pressure fluctuations, suspected cause: inlet filter blockage) (Cooling water pump A, hasPart, inlet filter) (Inlet filter, hasFault, clogged) The final output is a unified set of basic semantic triples: S = {s' ij |(e i ,r k ,e j )}, where s' ij Represents the entity e i ,、Tail entity e j and the relationship between them r k The basic triplet (e) i ,r k ,e j ).

[0069] In step S103 of this embodiment, in order to ensure that the device's operating status data can support the evaluation of the basic semantic triples and subsequent fault reasoning, the entity's operating status monitoring parameters are mapped to a standardized operating status vector.

[0070] Specifically, for each entity e i Construct the operating state vector V. For example, the operating state characteristics of a certain water pump A include the parameters shown in Table 9.

[0071] Table 9. Examples of operating status characteristics of a certain water pump A

[0072] Based on the parameters in Table 9, a 10-dimensional operating state vector v can be formed: v=[p_mean, p_std, p_slope, I_mean, I_std, T_mean, T_max, vib_rms, vib_peak, trend_slope].

[0073] Suppose that the sampling results of a certain water pump A over the last 10 minutes are recorded in the operating status dataset as follows: Average outlet pressure: 0.566 MPa Standard deviation of outlet pressure: 0.015 MPa Slope of outlet pressure change: +0.0012 / minute Average motor current: 33.1 A Motor current standard deviation: 0.9 A Average bearing temperature: 49℃ Maximum bearing temperature: 51℃ Vibration RMS: 1.83 mm / s Peak vibration: 2.1 mm / s Overall trend slope: +0.02 The running state vector is v = [0.566, 0.015, 0.0012, 33.1, 0.9, 49, 51, 1.83, 2.1, 0.02, 0.92].

[0074] Because parameters such as temperature, current, and pressure have different dimensions, they need to be standardized. For example, min-max standardization can be used to obtain a standardized operating state vector. v_norm = [0.66, 0.04, 0.55, 0.71, 0.36, 0.62, 0.68, 0.31, 0.34, 0.48,0.92] Understandably, in practical applications, for each entity e... i The original time-series data at its sampling frequency should be recorded, for example, sampling once every 1 minute, to obtain a running state vector v at the current sampling time point t. t The final output is each entity e. i Unified running state vector set V i-t ={v t}, where t is the sampling time point.

[0075] In step S104 of this embodiment, the rationality of the basic semantic triples is quantitatively evaluated based on the running state vectors corresponding to each entity in each basic semantic triple, and a scoring result is obtained.

[0076] For example, the rationality of the basic semantic triple can be quantified based on the transformation relationship between the head entity and tail entity's runtime state vectors. If the basic semantic triple (head entity, relation, tail entity) is rational, then the runtime state vector of the head entity, after a specific transformation based on the relation between entities, should be aligned with the runtime state vector of the tail entity in the semantic space. Therefore, its rationality can be evaluated based on the degree of alignment in this semantic space.

[0077] For example, a small neural network model can be used to pre-learn the transformation matrix relating the running state vectors of the head entity and the tail entity. For a basic semantic triple to be evaluated, the transformation result of the head entity's running state vector after passing through this transformation matrix is ​​calculated. Then, the cosine similarity between this transformation result and the tail entity's running state vector is calculated. This cosine similarity is used as the score for the basic semantic triple; the higher the similarity, the higher the reasonableness score of the basic semantic triple. This method allows for rapid quantitative evaluation of the reasonableness of basic semantic triples, providing support for subsequent basic semantic triple selection.

[0078] In step S105 of this embodiment, basic semantic triples are filtered based on the scoring results of step S104 to obtain a set of knowledge triples for constructing a fault diagnosis knowledge graph.

[0079] Specifically, a threshold method can be used for filtering, such as filtering out basic triples whose scores are greater than a preset score threshold to form a set of knowledge triples.

[0080] In step S106 of this embodiment, a fault diagnosis knowledge graph is constructed using the set of knowledge triples obtained in step S105.

[0081] Specifically, the ontology layer of the fault diagnosis knowledge graph is first constructed based on entity category, relation category, and attribute category. Then, based on the knowledge triple set, knowledge storage and graph construction are carried out. For example, the Neo4j graph database can be used to map the knowledge triples to nodes and edges of the knowledge graph to construct the fault diagnosis knowledge graph.

[0082] This invention addresses the auxiliary systems of hydropower plants and proposes a method for constructing a fault diagnosis knowledge graph. It introduces operational state vectors into traditional knowledge graphs, and integrates log records, equipment operational status data, and textual knowledge from multiple sources. Furthermore, it scores and filters the rationality of basic semantic triples, enabling the final knowledge graph to reflect the correlation of operating conditions and the evolution characteristics of operational states among different devices, thereby significantly improving knowledge representation capabilities and dynamic consistency.

[0083] Example 2 Please see Figure 2 , Figure 2 A second flowchart illustrating a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant, provided as an embodiment of the present invention, includes: S201. Obtain multi-source text data of auxiliary equipment in hydropower plants and preprocess it to form a multi-source dataset; S202. Entity and relation recognition based on multi-source datasets, and construction of a basic semantic triplet set; S203. For each entity in the set of basic semantic triples, construct the entity's running state vector based on the running state data corresponding to the entity. S204. For each basic semantic triple in the set of basic semantic triples, calculate the semantic relevance between the head entity and the tail entity in the basic semantic triple; S205. Based on the running state vectors corresponding to the head entity and the tail entity respectively, extract the temporal state vectors of the head entity and the tail entity in the preset time period through a temporal convolutional network, and calculate the state consistency between the head entity and the tail entity based on the difference between the temporal state vectors. S206. Based on the running state vectors corresponding to the head entity and the tail entity respectively, extract the trend features of the head entity and the tail entity in the preset time period using the sliding window difference method, and calculate the trend difference between the head entity and the tail entity based on the difference between the trend features. S207. Determine the confidence factor of the basic semantic triples based on state consistency and trend differences; S208. The rationality of the basic semantic triples is scored based on semantic relevance, state consistency, trend difference and confidence factor to obtain the scoring results; S209. Based on the scoring results, basic semantic triples are filtered to obtain a set of knowledge triples. S210. Complete the missing triples in the basic semantic triple set that have missing entities, and add them to the knowledge triple set. S211. Construct a fault diagnosis knowledge graph based on the set of knowledge triples; S212. Based on the confidence factor updated in time series, the fault diagnosis knowledge graph is dynamically updated.

[0084] This invention proposes a scoring mechanism that can jointly quantify semantic relevance, temporal trend, state consistency, and data confidence under dynamic operating conditions, and use it for redundant relationship elimination, potential missing knowledge completion, and feedback updates. For each basic semantic triple, it calculates the semantic relevance, state consistency, and trend difference between its head and tail entities. These indicators dynamically quantify the semantic consistency, state matching degree, data change trend, and uncertainty between the head and tail entities. Finally, it comprehensively evaluates the rationality and confidence of the basic triple from multiple dimensions, achieving effective triple selection and automatic completion of potential missing knowledge, thereby constructing a high-quality knowledge graph that can adaptively update with changing operating conditions.

[0085] Steps S201 to S203 in this embodiment can be referred to steps S101 to S103 in Embodiment 1, and will not be repeated here.

[0086] In the embodiments of the present invention, a dynamic context relationship scoring mechanism is proposed to quantify the rationality of basic semantic triples and provide a basis for subsequent fault reasoning and other work. It corresponds to the following steps S204~S208. The dynamic context relationship scoring mechanism will be described below in conjunction with the specific implementation steps.

[0087] In step S204 of this embodiment, for each basic semantic triple in the set of basic semantic triples, the semantic relevance between its head entity and tail entity is calculated.

[0088] Step S204 may specifically include the following sub-steps: S204-1, Semantic Embedding Coding.

[0089] For example, the basic semantic triple S={s' ij |(e i ,r k ,e j The head entity e in )} i Let it be denoted as h, and the tail entity be e. j Let t be the encoding of the head entity, relation, and tail entity of the basic semantic triple using a bilinear interaction structure. The entity embedding dimension is denoted as . (This dimension value can be 64, 128, 256 or 512).

[0090] Specifically, Xavier Uniform can be used to initialize the semantic embedding vectors of the head entities respectively. Tail entity semantic embedding vector , , .in, It represents a uniform distribution within the interval [a, b].

[0091] The relationship between the head entity and the tail entity can also be initialized using the Xavier-Normal method, which initializes the relationship weight matrix. , The relationship interaction vector is initialized using a standard normal distribution. , , Represents a standard normal distribution; for the relation bias term Initialized to zero .

[0092] S204-2, Parameter Update.

[0093] Based on the semantic embedding encoding above, the formula for calculating semantic relevance can be expressed as:

[0094] in, Semantic relevance, relation weight matrix Relational interaction vectors and relational bias terms These parameters all need to be updated and optimized before they can be used to calculate semantic relevance.

[0095] For example, in an embodiment of the present invention, a semantic relation encoder based on a graph neural network (GNN) can be used to calculate semantic relevance. The graph neural network is trained using end-to-end backpropagation, and the AdamW optimizer is used for iterative optimization to realize the relation weight matrix. Relational interaction vectors and relational bias terms The parameters are updated and optimized to output the prediction results of semantic relevance.

[0096] Specifically, a batch of triples is prepared in advance, their semantic relevance is calculated, and positive samples are constructed. The semantic relevance labels of the positive samples are... For each pair of triples, randomly replace the head or tail entity and calculate its semantic relevance. Construct negative samples, with the semantic relevance label for each negative sample being... The learning rate can be set to [value] based on these positive and negative samples to train the graphical neural network. The weight decays to loss function for:

[0097] After the above training, the resulting graph neural network can predict the semantic relevance of the input triples.

[0098] S204-3, Calculate semantic relevance.

[0099] Input the semantic embedding encoding result of the basic semantic triples in S204-1 into the graph neural network trained in S204-2, and output the semantic relevance of the basic semantic triples.

[0100] In step S205 of this embodiment, timing state encoding is performed based on the running state vectors corresponding to the head entity and the tail entity, and the state consistency between the head entity and the tail entity is calculated.

[0101] Step S205 may specifically include the following sub-steps: S205-1, Based on the running state vectors corresponding to the head entity and tail entity respectively Construct the running state matrices of the head entity and the tail entity for a preset time period, respectively. .

[0102] in, T represents the time step, which is the length of the observation window for the preset time period; This represents the dimension of the running state vector at each time step, i.e., the number of running state parameters of the entity.

[0103] For example, if water pump A obtains operating status data for 60 minutes in step S203 as shown in Table 10, then the observation window length is... T = 60, dimension of the running state vector = 3 (three state variables: temperature, flow rate, and pressure).

[0104] Table 10 Example of operating status data for water pump A over 60 minutes

[0105] Based on the above operating status data of water pump A at various time points, construct the operating status matrix of water pump A: .

[0106] S205-2. Extract the temporal state vectors corresponding to the running state matrices of the head entity and the tail entity respectively using a Temporal Convolutional Network (TCN). :

[0107] The temporal state vector is extracted to calculate the state consistency between the head entity and the tail entity.

[0108] S205-3. Based on the differences between the temporal state vectors, calculate the state consistency between the head entity and the tail entity. :

[0109] This state consistency characterizes the degree of difference between the operational state data corresponding to the head entity and the tail entity.

[0110] In step S206 of this embodiment, the trend features corresponding to the running state matrices of the head entity and the tail entity are extracted respectively, and the trend difference between the head entity and the tail entity is calculated.

[0111] Specifically, the sliding window difference of the running state matrix of the head entity is calculated separately. Sliding window difference of the running state matrix of the tail entity Table 10 shows the operating status matrix of water pump A. For example, calculate the sliding window difference. The formula is:

[0112] Where m and n are the running state matrices respectively. The row and column indices are used, and w is the sliding window size, which is determined as needed. If w=1, it is a first-order difference, which can be used to study velocity changes. If w=5, it is a difference of 5 time points, which can be used to capture short-term trends. If w=30, it is a difference of 30 time points, which can be used to capture long-term trends.

[0113] After calculating the sliding window difference of the running state matrices of the head and tail entities, feature extraction is performed using the linear transformation layer of the TCN to obtain the trend features corresponding to the running state matrices of the head and tail entities. :

[0114] Calculate trend characteristics Trend differences between :

[0115] This trend difference characterizes the degree of difference between the trends in the operational status data of the head and tail entities over time.

[0116] In step S207 of this embodiment, state consistency is... and trend differences The mapping is to confidence factors, and the mapping formula is:

[0117] Among them, confidence factor , These are the weights.

[0118] This confidence factor characterizes the degree of matching between the head and tail entities of the basic semantic triple and the reliability of the runtime status data.

[0119] In step S208 of this embodiment, based on semantic relevance... State consistency Trend differences The rationality of basic semantic triples is scored using indicators such as confidence factor, and the scoring results are obtained.

[0120] For example, a lightweight multilayer perceptron (MLP) can be used as a scoring discriminator to classify semantic relevance. State consistency Trend differences After splicing, combined with confidence factors As input to the scoring discriminator This yields the scoring result output by the scoring discriminator.

[0121]

[0122] in, It is an S-shaped function. The scoring results represent the basic semantic triples. Reasonableness score under current operating conditions.

[0123] To facilitate subsequent processing, the confidence factor can be... and Add basic semantic triples s' ij In this process, we obtain the basic semantic triples of the extended representation. s ij =[ s' ij , , ], and obtain the extended set of basic semantic triples S={ s ij}

[0124] S204-S208 above proposes a dynamic contextual relationship scoring mechanism. By introducing the dynamic operating state of the device, the contextual relationships between entities can be dynamically analyzed. Combined with semantic relevance, the basic semantic triples can be effectively quantitatively evaluated. Furthermore, under this mechanism, semantic relevance calculation, state consistency calculation, trend difference calculation, and rationality scoring can be performed by combining neural network models. These neural network models, combined with corresponding operations, can form a dynamic contextual relationship scoring model. By training and maintaining this dynamic contextual relationship scoring model, and inputting the basic semantic triples into it, the rationality score of the basic semantic triples can be automatically output, achieving rapid and high-quality triple quantitative evaluation.

[0125] In step S209 of this embodiment, the relationships between entities are filtered based on the multi-dimensional scoring results of step S208, and pseudo-associations are eliminated.

[0126] To improve the accuracy of knowledge graph edges and their consistency with the physical world, embodiments of this invention employ a dual threshold mechanism of semantics and state for filtering.

[0127] Specifically, a semantic relevance threshold is first preset. ,in, The reasonableness score distribution calculated for verified true triples is used, and its 95th confidence level is taken. As a semantic relevance threshold, it is used to ensure that only the most semantically significant relationships are retained; a pre-set reliability threshold. =0.7, meaning that at least 70% of the operational state consistency must be met.

[0128] For any basic semantic triple, the relation in the basic semantic triple is considered true and valid only if both of the following conditions are met:

[0129]

[0130] in Basic semantic triples The confidence factor between the head entity and the tail entity will automatically mark the basic semantic triple as "low confidence association" if any condition is not met. The relationship between the corresponding entities will be downweighted or directly removed in subsequent updates.

[0131] The basic semantic triples that satisfy the above conditions are selected from the basic semantic triple set S to form the knowledge triple set.

[0132] This invention filters the relationships of basic semantic triples from both semantic and state dimensions by using preset semantic relevance thresholds and preset confidence thresholds. It can automatically eliminate unreasonable or redundant relationships, thereby improving the quality of the knowledge triple set and the final knowledge graph.

[0133] In step S210 of this embodiment, the missing triples are predicted and completed based on the dynamic context relationship scoring mechanism proposed in this invention.

[0134] In some possible embodiments, missing triples with missing entities in the basic semantic triple set are filled in with entity completion information and added to the knowledge triple set, including: S210-1. For missing triples in the basic semantic triple set that have missing entities, select multiple candidate entities from the basic semantic triple set that meet the preset conditions with the entities in the missing triples to obtain a candidate entity set. S210-2. Score the reasonableness of each candidate entity in the candidate entity set and the candidate triple formed by the missing triple, and calculate the confidence factor of the candidate triple. S210-3: Select multiple candidate triples with the highest scores and confidence factors greater than the preset confidence threshold; add the selected candidate triples to the knowledge triple set.

[0135] In step S210-1 above, the preset conditions include that the state consistency is less than a predetermined state similarity threshold and the trend difference is less than a predetermined trend similarity threshold.

[0136] In some possible embodiments, the predetermined state similarity threshold is determined based on the mean of the average distance within each cluster after clustering the running state vectors corresponding to all entities in multiple selected basic semantic triples; the predetermined trend similarity threshold is determined based on the distribution of trend differences.

[0137] Specifically, all entity state vectors are clustered using the K-means clustering algorithm, and the mean of the average distances within the resulting K clusters is used as a threshold. For example, the state similarity threshold... Where k = 1, 2, ..., K, and K is the total number of categories. For the k-th cluster, This can be the average intra-cluster distance or cluster radius of k clusters. Trend similarity threshold. Can be taken ,in, This represents the 85% confidence level of the distribution of trend differences.

[0138] For a missing triple (h, r, ?) with a missing entity, where "?" represents the missing tail entity, candidate tail entities are selected from the set of basic semantic triples, and the candidate entity set is constrained according to the above preset conditions. :

[0139] in, Representing candidate entities, This represents the L2 norm.

[0140] For the candidate entity set Each candidate entity in The missing triple (h, r, ?) is filled into the missing triple to form a candidate triple. The reasonableness of the candidate triple is scored in the same way as in step S208 to obtain the score result of the candidate entity. : ) Simultaneously, the confidence factor of the candidate triple is calculated, which serves as the confidence factor for the candidate entity. .

[0141] In step S210-3 above, the scoring results corresponding to multiple candidate entities are... With confidence factor Jointly rank the data, selecting the top M scores with the highest ratings and a confidence level of not less than [value missing]. The candidate entities are identified, and the missing triples are completed based on these candidate entities. The knowledge-completed triples (i.e., the candidate triples selected through joint sorting) are then added to the knowledge triple set to form a new knowledge triple set.

[0142] This invention addresses triples with missing entities by using state consistency and trend difference indices to constrain state, determine candidate entity sets, score the rationality of candidate triples, and calculate their confidence factors. This allows for the prediction of potential reasonable associations even with missing information, enabling automatic completion of potentially missing knowledge. It can be applied to scenarios where faults or relationship records are not found in fault logs, thereby maintaining the accuracy, coverage, and real-time adaptability of the graph.

[0143] Step S211 in this embodiment can be referred to S106 in Embodiment 1, and will not be repeated here.

[0144] In step S212 of this embodiment, in order to maintain the real-time accuracy and adaptability of the knowledge graph, dynamic updates of the knowledge graph are continuously performed during the reasoning and feedback phases.

[0145] Specifically, when using the knowledge graph constructed by this invention for fault diagnosis, if the reasoning result of the knowledge graph is verified to be correct by on-site inspection, the confidence factor of the corresponding knowledge triple is increased:

[0146] If the knowledge graph prediction is incorrect, a penalty decay is applied to the corresponding confidence factor:

[0147] in, u、 These are the confidence factors before and after the update, respectively. The learning rate factor can be taken here. =0.2.

[0148] Furthermore, based on the results of actual maintenance verification, newly verified high-confidence samples can be periodically added to the training set to incrementally update the dynamic contextual relationship scoring model and dynamically optimize the model parameters and thresholds. During model training, the weights of semantic and state features are automatically balanced, enabling the continuous evolution of the knowledge graph.

[0149] Based on the above methods, a knowledge graph G=(E,R,S,V) for fault diagnosis of auxiliary equipment in hydropower plants can be constructed. This knowledge graph integrates static semantic relationships and dynamic operating states V, which can provide a highly reliable knowledge foundation for subsequent fault reasoning.

[0150] Example 3 Please see Figure 3 , Figure 3 A schematic diagram of a knowledge graph construction system for fault diagnosis of auxiliary equipment in a hydropower plant, provided as an embodiment of the present invention, is shown. The system includes: The data acquisition module 310 is used to acquire multi-source text data of auxiliary equipment in hydropower plants and preprocess it to form a multi-source dataset; wherein, the multi-source text data includes at least the operating status data of the equipment; Triple construction module 320 is used for entity and relation recognition based on multi-source datasets and to construct basic semantic triple set 330; The rationality assessment module 340 is used to construct the entity's running state vector for each entity in the basic semantic triplet set based on the entity's corresponding running state data; and to score the rationality of each basic semantic triplet in the basic semantic triplet set based on the running state vector to obtain the score result. The triplet filtering module 350 is used to filter basic semantic triplets based on the scoring results to obtain a set of knowledge triplets. The Knowledge Graph Construction Module 360 ​​is used to construct a fault diagnosis knowledge graph based on a set of knowledge triples.

[0151] The knowledge graph construction system for fault diagnosis of auxiliary equipment in hydropower plants described above and the knowledge graph construction method for fault diagnosis of auxiliary equipment in hydropower plants described above can be referred to and correspond to each other.

[0152] Example 4 Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 810, a communications interface 820, a memory 430, and a communication bus 440. The processor 410, communications interface 420, and memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions stored in the memory 430 to execute the knowledge graph construction method for fault diagnosis of auxiliary equipment in a hydropower plant provided in the above embodiments.

[0153] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0154] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a knowledge graph construction method for fault diagnosis of auxiliary equipment in hydropower plants provided in the above-described method embodiments.

[0155] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform a method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant provided by the methods described above.

[0156] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0157] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in a hydropower plant, characterized in that, include: Multi-source text data of auxiliary equipment in a hydropower plant is acquired and preprocessed to form a multi-source dataset; wherein, the multi-source text data includes at least the equipment's operating status data; Based on the multi-source dataset, entity and relation recognition is performed, and a basic semantic triplet set is constructed. For each entity in the set of basic semantic triples, an operational state vector of the entity is constructed based on the operational state data corresponding to the entity. Based on the running state vector, the rationality of each basic semantic triple in the set of basic semantic triples is scored to obtain the scoring result; Based on the scoring results, basic semantic triples are filtered to obtain a set of knowledge triples. A fault diagnosis knowledge graph is constructed based on the set of knowledge triples.

2. The method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants according to claim 1, characterized in that, The step of scoring the rationality of each basic semantic triplet in the basic semantic triplet set based on the running state vector includes: For each basic semantic triple in the set of basic semantic triples, calculate the semantic relevance between the head entity and the tail entity in the basic semantic triple; Based on the running state vectors corresponding to the head entity and the tail entity respectively, calculate the state consistency and trend difference between the head entity and the tail entity; wherein, the state consistency characterizes the degree of difference between the running state data of the head entity and the tail entity, and the trend difference characterizes the degree of difference between the trends of the running state data of the head entity and the tail entity over time. The rationality of the basic semantic triples is scored based on the semantic relevance, the state consistency, and the trend difference, and a scoring result is obtained.

3. The method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants according to claim 2, characterized in that, The step of calculating the state consistency and trend difference between the head entity and the tail entity based on the running state vectors corresponding to the head entity and the tail entity respectively includes: Based on the running state vectors corresponding to the head entity and the tail entity respectively, the temporal state vectors of the head entity and the tail entity in a preset time period are extracted by a temporal convolutional network. Based on the difference between the temporal state vectors, the state consistency between the head entity and the tail entity is calculated. Based on the running state vectors corresponding to the head entity and the tail entity, the trend features of the head entity and the tail entity in a preset time period are extracted by the sliding window difference method. Based on the differences between the trend features, the trend difference between the head entity and the tail entity is calculated.

4. The method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants according to claim 2, characterized in that, Based on the semantic relevance, the state consistency, and the trend difference The rationality of the basic semantic triples is scored to obtain the scoring results, including: Based on the state consistency and the trend difference, the confidence factor of the basic semantic triple is determined; wherein, the confidence factor characterizes the degree of matching of the running state data between the head entity and the tail entity of the basic semantic triple and the reliability of the running state data. The rationality of the basic semantic triples is scored based on the semantic relevance, state consistency, trend difference, and confidence factor to obtain the scoring results.

5. The method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants according to claim 3, characterized in that, The basic semantic triples are filtered based on the scoring results to obtain a set of knowledge triples, including: From the set of basic semantic triples, select basic semantic triples whose scores are greater than a preset semantic relevance threshold and whose confidence factors are greater than a preset confidence threshold, and obtain multiple selected basic semantic triples to form a knowledge triple set.

6. The method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants according to claim 5, characterized in that, After filtering basic semantic triples based on the scoring results to obtain a set of knowledge triples, the process further includes: For missing triples in the basic semantic triple set that have missing entities, multiple candidate entities that meet preset conditions with the entities in the missing triples are selected from the basic semantic triple set to obtain a candidate entity set; wherein, the preset conditions include state consistency less than a predetermined state similarity threshold and trend difference less than a predetermined trend similarity threshold. The reasonableness of each candidate entity in the candidate entity set and the candidate triple formed by the missing triple is scored, and the confidence factor of the candidate triple is calculated. Select multiple candidate triples with the highest scores and confidence factors greater than the preset confidence threshold; Multiple candidate triples selected are added to the knowledge triple set.

7. The method for constructing a knowledge graph for fault diagnosis of auxiliary equipment in hydropower plants according to claim 6, characterized in that, The predetermined state similarity threshold is determined by taking the mean of the average distance within each cluster after clustering the running state vectors corresponding to all entities in the multiple selected basic semantic triples. The predetermined trend similarity threshold is determined based on the distribution of the trend differences.

8. A knowledge graph construction system for fault diagnosis of auxiliary equipment in hydropower plants, characterized in that, include: The data acquisition module is used to acquire multi-source text data of auxiliary equipment in hydropower plants and preprocess it to form a multi-source dataset; wherein, the multi-source text data includes at least the equipment's operating status data; The triplet construction module is used to identify entities and relationships based on the multi-source dataset and construct a basic semantic triplet set. The rationality assessment module is used to construct an operational state vector for each entity in the set of basic semantic triples based on the operational state data corresponding to the entity; and to score the rationality of each basic semantic triple in the set of basic semantic triples based on the operational state vector to obtain a score result. The triplet filtering module is used to filter basic semantic triplets based on the scoring results to obtain a set of knowledge triplets. The knowledge graph construction module is used to construct a fault diagnosis knowledge graph based on the set of knowledge triples.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the knowledge graph construction method for fault diagnosis of auxiliary equipment in hydropower plants as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the knowledge graph construction method for fault diagnosis of auxiliary equipment in hydropower plants as described in any one of claims 1 to 7.

Images