Hydroelectric generating set fault early warning method, system and equipment based on knowledge graph, and medium

Abstract

The invention discloses a hydroelectric generating set fault early warning method, system and equipment based on a knowledge graph and a medium, and belongs to the technical field of hydroelectric generating set fault early warning, and the method comprises the steps: constructing a hydroelectric generating set knowledge graph database, and representing a set structure, operation information and a fault association relationship; collecting real-time operation data of the hydroelectric generating set, and performing data preprocessing; based on the knowledge graph and the collected data, path traversal and node matching are carried out by adopting a multi-modal reasoning mechanism fusing rule reasoning and case reasoning, potential abnormal parameters are identified, and a candidate fault hypothesis set is generated; and when the confidence coefficient of any fault hypothesis exceeds a set threshold value, outputting corresponding fault early warning information. According to the method, a fault hypothesis system is constructed by means of an existing fault propagation path in a knowledge graph and abnormal data captured by real-time monitoring, the confidence coefficient of fault hypothesis is accurately calculated based on the Bayesian network, and timely and accurate fault alarm triggering is achieved.

Description

Technical Field

[0001] This invention relates to the field of hydropower unit fault early warning technology, specifically to a hydropower unit fault early warning method, system, equipment, and medium based on knowledge graph. Background Technology

[0002] Traditional fault diagnosis methods for hydropower units largely rely on rules and expert experience, which are insufficient to meet practical needs when faced with complex and ever-changing operating conditions. In terms of fault early warning, existing systems often rely on simple threshold judgments or statistical methods, failing to delve into the unit's operating data, resulting in insufficient early warning accuracy and frequent false alarms and missed alarms. By constructing a knowledge graph and integrating multi-source information, fault relationships are presented in an intuitive and visual form, solving the problem of insufficient interpretability in traditional methods and providing strong support for the safe and stable operation of hydropower units.

[0003] This invention refines the knowledge graph part and relies on Bayesian networks to accurately calculate the confidence of fault hypotheses, thereby achieving timely and accurate fault alarm triggering. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is: how to solve the problem that traditional hydropower unit fault diagnosis methods rely heavily on rules and expert experience, which are difficult to meet actual needs when faced with complex and ever-changing operating modes; in terms of fault early warning, existing systems are often based on simple threshold judgment or statistical methods, which cannot deeply mine the unit's operating data, resulting in insufficient early warning accuracy and frequent false alarms and missed alarms.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a knowledge graph-based method for early warning of hydropower unit faults, comprising the following steps: constructing a knowledge graph database for hydropower units to characterize unit structure, operating information, and fault correlations; collecting real-time operating data of the hydropower units and performing data preprocessing; based on the knowledge graph and the collected data, employing a multimodal reasoning mechanism that integrates rule-based reasoning and case-based reasoning to perform path traversal and node matching, identifying potential abnormal parameters, and generating a set of candidate fault hypotheses; for each fault hypothesis, constructing a multidimensional Bayesian network containing monitoring parameters, equipment operating conditions, and historical failure rates, calculating the fault confidence by combining the causal weight information between nodes in the knowledge graph, and outputting corresponding fault warning information when the confidence of any fault hypothesis exceeds a set threshold.

[0007] As a preferred embodiment of the knowledge graph-based hydropower unit fault early warning method of the present invention, the knowledge graph database includes information extraction based on historical fault records of hydropower units to construct a hydropower unit ontology knowledge graph database, including data collection and organization, knowledge extraction, knowledge representation, and graph construction and storage steps.

[0008] As a preferred embodiment of the knowledge graph-based hydropower unit fault early warning method of the present invention, the step of collecting real-time operating data of the hydropower unit and performing data preprocessing includes acquiring real-time operating data and performing data standardization processing.

[0009] As a preferred embodiment of the knowledge graph-based hydropower unit fault early warning method of the present invention, the step of identifying potential abnormal parameters and generating a set of candidate fault hypotheses includes matching the collected data and generating fault hypotheses based on the node relationship information in the knowledge graph.

[0010] As a preferred embodiment of the knowledge graph-based hydropower unit fault early warning method of the present invention, the knowledge extraction includes entity extraction and relation extraction; the knowledge representation includes knowledge base mapping of entities and relations and entity linking methods; the knowledge base mapping of entities and relations includes establishing links between entity fields extracted from text and corresponding entities in the knowledge base, and mapping and storing relations extracted from text in the knowledge base through entities at both ends of the relationship; the entity linking method includes entity matching based on terminology dictionary, entity disambiguation enhanced by relation tags, and knowledge base entity fusion.

[0011] As a preferred embodiment of the knowledge graph-based hydropower unit fault early warning method of the present invention, the following steps are included: the entity matching based on the terminology dictionary includes calculating the matching text entity with the candidate entity in the terminology dictionary using string Jaccard similarity calculation, and optimizing the matching accuracy by combining contextual semantic information word vector similarity; the entity disambiguation enhanced by relation tags includes filtering semantically consistent candidate entities using relation tags attached to entity references, and excluding erroneous links through relation constraints; the knowledge base entity fusion includes normalizing the successfully matched entity references and terminology dictionary entries to generate unique knowledge base entity identifiers, and storing the linking results in the knowledge graph database.

[0012] As a preferred embodiment of the knowledge graph-based hydropower unit fault early warning method described in this invention, the following steps are included: generating fault hypotheses involves matching collected operational data with fault propagation paths in the knowledge graph, and outputting a set of candidate fault hypotheses based on rule-based reasoning and case-based reasoning mechanisms; calculating fault confidence involves constructing a multidimensional Bayesian network for each fault hypothesis, including monitoring parameters, equipment operating conditions, and historical failure rates, calculating a joint probability distribution based on the causal relationship weights in the knowledge graph, obtaining the confidence of each fault hypothesis, and triggering a fault early warning when any confidence exceeds a preset threshold.

[0013] This invention constructs a fault hypothesis system and relies on Bayesian networks to accurately calculate the confidence level of fault hypotheses, thereby achieving timely and accurate fault alarm triggering.

[0014] Another objective of this invention is to provide a knowledge graph-based early warning system for hydropower unit faults.

[0015] To address the aforementioned technical problems, this invention provides the following technical solution: a knowledge graph-based hydropower unit fault early warning system, comprising: a knowledge graph construction module, a real-time data acquisition and preprocessing module, a dynamic inference engine construction module, and a system integration and deployment module; the knowledge graph construction module is used to extract information based on historical fault records of hydropower units and construct a hydropower unit ontology knowledge graph database, including data collection and organization, knowledge extraction, knowledge representation, and graph construction and storage steps; the real-time data acquisition and preprocessing module is used to process and preprocess the collected real-time data of the hydropower units; the dynamic inference engine construction module is used to construct a fault hypothesis system by leveraging existing fault propagation paths within the knowledge graph and abnormal data captured by real-time monitoring, and to accurately calculate the confidence level of the fault hypothesis using a Bayesian network, thereby achieving... Timely and accurate fault alarm triggering includes a fault hypothesis generation unit and a confidence calculation unit. The fault hypothesis generation unit is used to perform path traversal and node matching based on the knowledge graph and collected data, using a multimodal reasoning mechanism that integrates rule reasoning and case reasoning to identify potential abnormal parameters and generate a set of candidate fault hypotheses. The confidence calculation unit is used to construct a multidimensional Bayesian network containing monitoring parameters, equipment operating conditions, and historical failure rate factors for each fault hypothesis, and calculate the fault confidence by combining the causal weight information between nodes in the knowledge graph. When the confidence of any fault hypothesis exceeds a set threshold, the corresponding fault warning information is output. The system integration and deployment module is used to integrate the knowledge graph construction module, data acquisition and preprocessing module, and dynamic reasoning engine module to form a complete hydropower unit fault early warning system.

[0016] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the knowledge graph-based hydropower unit fault early warning method.

[0017] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the knowledge graph-based hydropower unit fault early warning method.

[0018] The beneficial effects of this invention are as follows: By leveraging existing fault propagation paths within a knowledge graph and real-time monitoring-captured anomaly data, this invention constructs a fault hypothesis system. Relying on a Bayesian network, it accurately calculates the confidence level of these fault hypotheses, enabling timely and accurate fault alarm triggering. By integrating multi-source information through the construction of a knowledge graph, it solves the problem of insufficient interpretability in traditional methods, providing strong support for the safe and stable operation of hydropower units. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The above is a flowchart of a knowledge graph-based early warning method for hydropower unit faults, provided as an embodiment of the present invention.

[0021] Figure 2 The flowchart shows a knowledge graph construction module for a knowledge graph-based hydropower unit fault early warning method provided in one embodiment of the present invention.

[0022] Figure 3 This is a system scheme module diagram of a knowledge graph-based hydropower unit fault early warning system provided in one embodiment of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0024] Example 1, referring to Figure 1This is the first embodiment of the present invention, which provides a knowledge graph-based method for early warning of hydropower unit faults, including:

[0025] S1. Construct a knowledge graph database for hydropower units to represent the unit structure, operating information, and fault correlations.

[0026] S2. Collect real-time operating data of the hydropower unit and perform data preprocessing.

[0027] S3. Based on the knowledge graph and the collected data, a multimodal reasoning mechanism that integrates rule-based reasoning and case-based reasoning is used to perform path traversal and node matching, identify potential abnormal parameters, and generate a set of candidate fault hypotheses.

[0028] S4. For each fault hypothesis, construct a multidimensional Bayesian network that includes monitoring parameters, equipment operating conditions and historical failure rates. Combine the causal weight information between nodes in the knowledge graph to calculate the fault confidence. When the confidence of any fault hypothesis exceeds the set threshold, output the corresponding fault warning information.

[0029] In this embodiment, without relying on traditional empirical rules or single threshold judgments, a graph model incorporating knowledge of equipment structure, operating status, and fault associations is constructed. Through unified preprocessing of the collected data, effective integration of data and knowledge is achieved. By combining rule-based reasoning and case-based reasoning, potential fault scenarios can be automatically identified from complex data, and the probability of each hypothesis is quantitatively evaluated based on a multidimensional Bayesian network. When the confidence level of a fault hypothesis exceeds a set threshold, a timely warning is output, helping to detect fault risks early and take intervention measures. This method features clear structural expression, interpretable reasoning process, and high alarm triggering accuracy, making it suitable for various types of early warning scenarios for hydropower equipment.

[0030] Example 2, refer to Figure 1 and Figure 2 This is the second embodiment of the present invention. Based on the above embodiments, a method for early warning of hydropower unit faults based on knowledge graphs is provided.

[0031] In this embodiment of the application, step S1 involves extracting information based on historical fault records of hydropower units and constructing a knowledge graph database of hydropower unit ontology, including data collection and organization, knowledge extraction, knowledge representation, and graph construction and storage steps.

[0032] Knowledge extraction includes entity extraction and relation extraction; knowledge representation includes knowledge base mapping of entities and relations and entity linking methods; knowledge base mapping of entities and relations includes linking entity fields extracted from text with corresponding entities in the knowledge base, and mapping and storing relations extracted from text in the knowledge base through the entities at both ends of the relationship; entity linking methods include entity matching based on terminology dictionary, entity disambiguation enhanced by relation tags, and knowledge base entity fusion.

[0033] Entity matching based on a terminology dictionary includes calculating the similarity between the matched text entity and the candidate entity in the terminology dictionary using string Jaccard similarity, and optimizing the matching accuracy by combining contextual semantic information and word vector similarity; entity disambiguation enhanced by relation tags includes filtering semantically consistent candidate entities using relation tags attached to entity references, and eliminating erroneous links through relation constraints; knowledge base entity fusion includes normalizing the successfully matched entity references and terminology dictionary entries to generate unique knowledge base entity identifiers, and storing the linking results in a knowledge graph database.

[0034] Specifically, structural data of each component of the hydropower unit is collected to clarify the physical connection relationship and hierarchical structure between the components, historical fault records of the hydropower unit are compiled, and expert experience rules in the field of hydropower unit fault diagnosis are extracted.

[0035] The knowledge extraction process includes entity extraction and relation extraction. Entity extraction involves identifying the fields to be extracted and analyzing their textual structure, constructing field rule templates based on the results of manual analysis, and extracting entities from structured data according to the rule templates. Relation extraction includes extracting relations from structured data without additional relation extraction, and extracting relations from unstructured data using sentence structure-based relations extraction and plain text label-based relations extraction.

[0036] Relation extraction based on sentence structure includes designing conditional sentence rule templates, matching the antecedent and consequent of conditional sentences, determining the beginning and end of relations, dividing conditional relations for simple conditional sentences by matching antecedent and consequent identifiers using a predefined keyword library, and dividing antecedent and consequent for complex conditional sentences based on specific symbol identifiers; Relation extraction based on plain text tags includes introducing relation tags in the entity annotation stage, using the "BIO" annotation method, and expanding to 15 types of entity-relation joint tags.

[0037] The knowledge representation steps include knowledge base mapping of entities and relations and entity linking methods. Knowledge base mapping of entities and relations includes linking entity fields extracted from text with corresponding entities in the knowledge base, and mapping and storing relations extracted from text in the knowledge base through the entities at both ends of the relationship. Entity linking methods include entity matching based on terminology dictionary, entity disambiguation enhanced by relation tags, and knowledge base entity fusion.

[0038] Entity matching based on a terminology dictionary uses string Jaccard similarity calculation to match text entities with candidate entities in the terminology dictionary, and combines contextual semantic information and word vector similarity to optimize matching accuracy; entity disambiguation enhanced by relation tags uses relation tags attached to entity references to filter semantically consistent candidate entities, and excludes erroneous links through relation constraints; knowledge base entity fusion normalizes the successfully matched entity references and terminology dictionary entries to generate unique knowledge base entity identifiers, and stores the linking results in a knowledge graph database.

[0039] The knowledge graph construction and storage process includes graph structure modeling, graph database storage, and data consistency maintenance. Graph structure modeling uses nodes to represent entities, with node attributes including entity type, name, and association parameters. Edges represent relationships between entities, with edge attributes including relationship type, weight, and time information. Graph database storage uses a graph database to store the knowledge graph, achieving efficient knowledge retrieval and updating through a graph query language, and establishing indexes to optimize the query performance of nodes and edges. Data consistency maintenance sets entity-relationship constraint rules and uses a transaction mechanism to ensure the atomicity and consistency of data writes.

[0040] Specifically, S1.1, data collection and organization.

[0041] Equipment structure information: Collect structural data of each component of the hydropower unit to clarify the physical connection relationship and hierarchical structure between the components.

[0042] Historical Fault Cases: Compile historical fault records of hydropower units, including the time of occurrence of the fault, description of the fault phenomenon, location of the faulty component, fault handling measures, and final resolution.

[0043] Expert rules: Extract expert experience rules in the field of hydropower unit fault diagnosis, including fault tendency judgment rules under specific combinations of operating parameters.

[0044] S1.2 Knowledge Extraction.

[0045] Entity extraction: Extracting entities from hydropower unit fault handling data, specifically including: identifying the fields to be extracted and analyzing their textual description structure; constructing field rule templates based on manual analysis results; and extracting entities from structured data according to the rule templates.

[0046] Relation extraction:

[0047] Structured data relation extraction: Entities and relations in structured data correspond one-to-one, eliminating the need for additional relation extraction.

[0048] Unstructured data relation extraction: Two methods are used: relation extraction based on sentence structure and relation extraction based on plain text labels.

[0049] Based on relation extraction using sentence structure, conditional sentence rule templates are designed to match the antecedent and consequent of conditional sentences, and to determine the beginning and end of the relation, including:

[0050] Processing of simple conditional sentences: Matching antecedent and consequent identifiers using a predefined keyword library to define conditional relationships.

[0051] Conditional compound sentence processing: The antecedent and consequent are divided based on specific symbolic identifiers (such as "if...then..." or "if...then...").

[0052] Based on plain text tag-based relation extraction, relation tags are introduced in the entity annotation stage to achieve joint extraction of entities and relations. Specifically, this includes: on the basis of 5 basic entity tags (such as "equipment" and "action"), further subdividing relation tags (such as "equipment_disposal equipment" and "action_disposal action").

[0053] The "BIO" annotation method is adopted, expanding to 15 types of entity-relation joint labels to improve relation extraction accuracy. Experiments are conducted to evaluate the impact of different relation labels on the entity recognition model, optimizing the model's performance in accuracy, recall, and F1 score.

[0054] S1.3 Knowledge Representation.

[0055] The extracted entities and relationships are stored in a structured manner using a knowledge representation language to facilitate subsequent knowledge base construction and computation. This includes the following steps:

[0056] S1.3.1 Knowledge base mapping of entities and relations.

[0057] Link the entity fields extracted from the text with the corresponding entities in the knowledge base.

[0058] For relationships extracted from text, the relationships are mapped and stored in the knowledge base by associating the entities at both ends of the relationship.

[0059] S1.3.2, Entity Linking Method.

[0060] Traditional entity linking algorithms rely on the discriminative information of entities in the knowledge base, while this invention constructs knowledge base entity links based on the following data:

[0061] Terminology dictionary: Obtain a terminology dictionary containing entity names through term mining.

[0062] Entity references: Extract entity references with relational labels from power grid fault handling data.

[0063] By establishing a mapping relationship between a terminology dictionary and entity references, a complete link between knowledge base entities is achieved. Specific methods include entity matching based on the terminology dictionary, entity disambiguation enhanced by relation tags, and knowledge base entity fusion.

[0064] Entity matching based on a terminology dictionary is performed by using string Jaccard similarity to calculate the matching text entities with candidate entities in the terminology dictionary; the matching accuracy is optimized by combining contextual semantic information and word vector similarity.

[0065] Entity disambiguation enhanced by relation tags uses relation tags attached to entity references (such as "equipment_monitoring object" or "action_handling action") to filter candidate entities with semantic consistency; and excludes erroneous links by relation constraints (such as "circuit breaker" can only be associated with "protection action" type relations).

[0066] The knowledge base entity fusion process normalizes the successfully matched entity references and terminology dictionary entries to generate unique knowledge base entity identifiers; the linking results are stored in the knowledge graph database to support subsequent reasoning and querying.

[0067] S1.4, Map Construction and Storage.

[0068] Based on the entities and relationships extracted in the aforementioned steps, a knowledge graph of hydropower unit faults is constructed using a graph database, specifically including the following steps:

[0069] S1.4.1 Graph structure modeling.

[0070] Entities are represented by nodes, and node attributes include entity type, name, and associated parameters.

[0071] Relationships between entities are represented by edges, and edge attributes include relationship type, weight, and time information.

[0072] S1.4.2 Graph database storage.

[0073] Use graph databases (such as Neo4j and Nebula Graph) to store knowledge graphs.

[0074] Efficient knowledge retrieval and updating can be achieved through graph query languages ​​(such as Cypher).

[0075] Optimize query performance for nodes and edges by creating indexes.

[0076] S1.4.3, Data Consistency Maintenance

[0077] Set entity-relationship constraint rules (e.g., "rotor component" can only be associated with "vibration monitoring parameters" type relationships).

[0078] The atomicity and consistency of data writing are ensured through transaction mechanisms.

[0079] The embodiments of this invention ensure that the constructed graph possesses panoramic, semantic, and evolvable characteristics. Because entities and relationships are clearly defined, the graph structure can support subsequent high-precision path traversal and node reasoning. Therefore, compared to single-source graph construction methods, this invention has significant advantages in terms of graph scalability, expressiveness, and adaptability to causal reasoning.

[0080] In this embodiment of the application, in step S2, real-time operating data of the hydropower unit is collected and data preprocessing is performed, and operating data such as vibration, temperature, pressure, and speed are collected in real time.

[0081] Data preprocessing includes data cleaning and data normalization. Data cleaning is used to remove noisy data from the collected data and fill missing values ​​using interpolation or machine learning-based methods. Data normalization is used to map collected data of different types and scales to a unified range through normalization methods.

[0082] Specifically, real-time data acquisition includes collecting operational data such as vibration, temperature, pressure, and rotational speed.

[0083] Data cleaning removes noisy data from the collected data. For missing values, interpolation methods (such as linear interpolation and spline interpolation) or machine learning-based methods (such as the K-nearest neighbor algorithm) are used to fill in the missing values.

[0084] Data normalization maps collected data of different types and scales to a unified interval through normalization methods (such as min-max normalization and Z-score standardization), which facilitates subsequent model processing.

[0085] In one optional implementation, the collected data originates from an existing industrial monitoring system interface. This interface periodically calls the OPC data channel to obtain information on electrical equipment operating status, power consumption, voltage, and other indicators. During the data preprocessing stage, Z-score standardization is used to map all indicators to a zero-mean, unit-variance distribution. This method allows access to existing data platforms without adding new data acquisition hardware, making it suitable for system-integrated deployment scenarios.

[0086] In another alternative implementation, the acquisition device performs local data processing via an edge computing terminal. The acquired data is preprocessed in real time by the edge nodes before being uploaded to the central inference platform. The edge preprocessing methods include missing value imputation and median normalization based on the K-nearest neighbor algorithm. This method can reduce data transmission bandwidth pressure and is suitable for deployment environments with limited network resources or high real-time requirements.

[0087] This application's implementation combines multi-source, high-frequency, real-time data acquisition with a unified normalization processing mechanism to construct a high-quality data input sequence with strong continuity and uniform format. This input structure significantly enhances the consistency of data expression and interaction efficiency in the inference engine, thereby effectively improving the accuracy of fault identification and the stability of confidence modeling.

[0088] In this embodiment of the application, in step S3, the collected data is matched based on the node relationship information in the knowledge graph and fault hypotheses are generated. The collected running data is matched with the fault propagation path in the knowledge graph, and a set of candidate fault hypotheses is output based on rule reasoning and case reasoning mechanisms.

[0089] Extract the pre-defined fault propagation path topology from the knowledge graph, receive abnormal parameter indicators from the real-time monitoring data stream, map the abnormal parameter indicators to knowledge graph nodes based on the multimodal reasoning mechanism to perform path traversal, and output a set of candidate fault hypotheses.

[0090] Specifically, by leveraging existing fault propagation paths within the knowledge graph and anomaly data captured through real-time monitoring, a fault hypothesis system is constructed. Furthermore, a Bayesian network is used to accurately calculate the confidence level of these fault hypotheses, thereby enabling timely and accurate fault alarm triggering. This includes:

[0091] S3.1 Fault hypothesis generation.

[0092] S3.1.1 Extract the pre-defined fault propagation path topology from the knowledge graph.

[0093] S3.1.2 Receive abnormal parameter indicators from the real-time monitoring data stream, including but not limited to vibration amplitude, temperature gradient and pressure fluctuation.

[0094] S3.1.3 Based on the multimodal reasoning mechanism, abnormal parameter indicators are mapped to knowledge graph nodes for path traversal. The multimodal reasoning mechanism includes rule reasoning and case reasoning.

[0095] Rule-based reasoning applies preset fault association rules to perform logical deduction.

[0096] Case-based reasoning matches similarity features from a historical failure case database.

[0097] S3.1.4 Output a set of candidate fault hypotheses, where each hypothesis includes the fault type and associated equipment location. Exemplary hypotheses include bearing wear, rotor imbalance, or lubricant failure.

[0098] In one optional implementation, the multimodal inference mechanism employs a path scoring method based on graph neural networks. It selects the subgraph structure most relevant to the current monitoring state from the graph, calculates the confidence score of the fault path through node embedding and edge propagation, and then selects the highest-scoring paths as fault hypotheses. This approach is suitable for scenarios with graph training capabilities and a need to model complex nonlinear paths.

[0099] In another alternative implementation, the inference process is entirely based on a rule system, which consists of predefined logical statements. Each rule specifies a trigger condition and a corresponding fault type. When the collected data meets part of the condition of a rule, the corresponding fault type is directly used as a candidate hypothesis. This approach is logically clear, has a fast response time, and is suitable for edge deployments or embedded lightweight models.

[0100] The implementation method of this application, through the fusion of rule-based reasoning and case-based reasoning, can not only cover predefined fault logic chains but also identify historically similar scenarios that are not fully matched, thereby improving the adaptability and interpretability of fault hypothesis generation. Because it fully utilizes knowledge graph structures and historical data accumulation, it can effectively reduce false positives and false negatives, providing a high-quality candidate set for subsequent quantitative reasoning.

[0101] In this embodiment of the application, in step S4, a multidimensional Bayesian network containing monitoring parameters, equipment operating conditions and historical failure rates is constructed for each fault hypothesis. The joint probability distribution is calculated based on the causal relationship weights in the knowledge graph to obtain the confidence level of each fault hypothesis. A fault warning is triggered when any confidence level exceeds a preset threshold.

[0102] For each fault hypothesis, a multidimensional Bayesian network is constructed. The joint probability distribution is calculated through a probabilistic fusion engine, and the quantitative confidence index of each fault hypothesis is output. When the confidence of any hypothesis exceeds a preset threshold, a hierarchical alarm signal is triggered. The root node of the multidimensional Bayesian network represents the fault hypothesis ontology. The child node layer includes real-time monitoring parameter nodes, equipment operating condition parameter nodes, and historical failure rate nodes. The edges establish conditional dependencies based on the causal relationship weights in the knowledge graph. The input layer of the probabilistic fusion engine receives sensor time-series data from the real-time data acquisition module, the knowledge layer calls the prior probability distribution table stored in the knowledge graph, and the computation layer applies the Bayesian backpropagation algorithm to iteratively update the node confidence.

[0103] S4.1 Construct a multidimensional Bayesian network for each fault hypothesis. The network structure includes:

[0104] Root node: Represents the fault hypothesis ontology.

[0105] Sub-node layer: includes real-time monitoring parameter nodes, equipment operating condition parameter nodes, and historical failure rate nodes.

[0106] Edge: Establish conditional dependencies based on the causal relationship weights in the knowledge graph.

[0107] S4.2 Calculate the joint probability distribution using the probability fusion engine, specifically including:

[0108] Input layer: Receives sensor timing data from the real-time data acquisition module.

[0109] Knowledge layer: Calls the prior probability distribution table stored in the knowledge graph.

[0110] Computational layer: The Bayesian backpropagation algorithm is applied to iteratively update the node confidence.

[0111] S4.3 Output the quantitative confidence index of each fault hypothesis. When the confidence of any hypothesis exceeds the preset threshold, a graded alarm signal is triggered.

[0112] Example 3, the third embodiment of the present invention, is scientifically demonstrated through experiments in order to verify the beneficial effects of the present invention.

[0113] The knowledge graph construction module is used to extract information based on the historical fault records of hydropower units and build a knowledge graph database of hydropower unit ontology.

[0114] S1.1 Data collection implementation.

[0115] In accordance with the requirements for data collection and organization, professional personnel were assigned to collect equipment structural information, historical fault cases, and expert rules for the hydropower units, and then systematically organized them. For example, the equipment structural information included 3D model data and assembly diagrams of core components such as the turbine runner, main shaft, and guide bearings; historical fault cases compiled fault records from the past 5 years, with a typical example being: the unit's thrust bearing temperature exceeded the limit alarm, which was diagnosed as severe oil contamination on the surface of the lower wind tunnel, requiring modification to remove oil from the deduced combined bearing; and 15 diagnostic rules were collected, such as "vibration frequency 2 × rotational frequency and temperature > 75℃ → bearing wear probability 85%".

[0116] S1.2 Knowledge Extraction Implementation.

[0117] Entity extraction: The BERT-BiLSTM-CRF model is used to identify entities in the fault reports. For example:

[0118] Text: "Thrust bearing temperature rises to 68℃" → Entities: "Thrust Bearing" (equipment), "Temperature" (parameter), "68℃" (value)

[0119] Relation extraction:

[0120] Rule-based relation extraction: Create 38 conditional sentence templates, such as: "When [parameter] exceeds [threshold], [fault] may occur" → extract relations (parameter, warning indicator, fault).

[0121] Tag-based relation extraction: Define an entity-relationship joint tag system, for example: "Insufficient cooling water flow causes bearing overheating" → (cooling water flow, causes, bearing temperature).

[0122] S1.3 Knowledge Representation and Storage Implementation.

[0123] Entity linking: Linking entities according to the method of the present invention.

[0124] Using Neo4j graph database for storage, the system is built and includes:

[0125] Nodes: Device class and parameter class.

[0126] Edges: include relationships such as "cause", "affect", and "association".

[0127] The real-time data acquisition and preprocessing module is used to process historical hydropower unit fault data collected from the system.

[0128] 128 monitoring points were deployed to collect operational data such as vibration, temperature, pressure, and rotational speed in real time.

[0129] Noisy data was removed from the collected data. For missing values, linear interpolation and spline interpolation were used to fill in the missing values.

[0130] Different types and units of collected data are normalized using the Z-score normalization method to map the data to a unified interval, which facilitates subsequent model processing.

[0131] The dynamic inference engine module extracts the fault propagation path topology from the knowledge graph, receives real-time monitored abnormal parameter indicators, performs path traversal based on a multimodal inference mechanism, and generates a set of candidate fault hypotheses.

[0132] Real-time monitoring detected the following anomalies: Vibration: 2× frequency component amplitude 0.18mm (baseline 0.05mm); Temperature: guide bearing temperature 72℃ (normal <65℃). Through knowledge graph reasoning, traversing three fault propagation paths, hypotheses were generated: rotor dynamic imbalance (62% confidence level), excessive bearing clearance (78% confidence level), and lubricating oil contamination (85% confidence level).

[0133] For each fault hypothesis (such as turbine bearing failure), a multidimensional Bayesian network is constructed. The joint probability distribution is calculated through a probability fusion engine, and the quantitative confidence index of each fault hypothesis is output. When the confidence exceeds a preset threshold, a graded alarm signal is triggered.

[0134] Taking unit bearing failure as an example, a Bayesian network is constructed. Root node: bearing failure; evidence nodes: vibration spectrum characteristics (likelihood probability 0.82) and oil detection results (likelihood probability 0.76); output: lubricating oil contamination confidence level 89.2%, thus triggering a level one alarm.

[0135] System integration and deployment module. This module involves interface design and development for the knowledge graph construction module, data acquisition and preprocessing module, and dynamic inference engine module, enabling data interaction and collaborative work between these modules. All modules are then integrated into a complete hydropower unit fault early warning system, and the system is tested and optimized. Finally, the system is deployed to a real-world hydropower unit operation monitoring environment to monitor the operating status of the hydropower units in real time and provide timely fault warnings.

[0136] Example 4 is the fourth embodiment of the present invention, which differs from the previous two embodiments in that:

[0137] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they 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.

[0138] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0139] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0140] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented in combination with any of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0141] Example 5, refer to Figure 3 This is the fifth embodiment of the present invention. This embodiment provides a knowledge graph-based hydropower unit fault early warning system, including a knowledge graph construction module, a real-time data acquisition and preprocessing module, a dynamic reasoning engine construction module, and a system integration and deployment module.

[0142] The knowledge graph construction module is used to extract information based on the historical fault records of hydropower units and build a knowledge graph database of hydropower unit ontology, including data collection and organization, knowledge extraction, knowledge representation, and graph construction and storage steps.

[0143] The real-time data acquisition and preprocessing module is used to process and preprocess the real-time data acquired from the hydropower unit.

[0144] The dynamic reasoning engine building module is used to construct a fault hypothesis system by leveraging the existing fault propagation paths within the knowledge graph and the abnormal data captured by real-time monitoring. It also relies on Bayesian networks to accurately calculate the confidence of the fault hypotheses, thereby achieving timely and accurate fault alarm triggering. The module includes a fault hypothesis generation unit and a confidence calculation unit.

[0145] The fault hypothesis generation unit is used to perform path traversal and node matching based on knowledge graphs and collected data, and to identify potential abnormal parameters and generate a set of candidate fault hypotheses.

[0146] The confidence calculation unit is used to construct a multi-dimensional Bayesian network for each fault hypothesis, which includes monitoring parameters, equipment operating conditions and historical failure rates. It calculates the fault confidence by combining the causal weight information between nodes in the knowledge graph. When the confidence of any fault hypothesis exceeds the set threshold, the corresponding fault warning information is output.

[0147] The system integration and deployment module is used to integrate the knowledge graph construction module, data acquisition and preprocessing module, and dynamic reasoning engine module to form a complete hydropower unit fault early warning system.

[0148] The data collection and organization steps of the knowledge graph construction module include collecting structural data of each component of the hydropower unit to clarify the physical connection relationship and hierarchical structure between the components, organizing historical fault records of the hydropower unit, and extracting expert experience rules in the field of hydropower unit fault diagnosis.

[0149] The knowledge extraction steps of the knowledge graph construction module include entity extraction and relation extraction. Entity extraction includes determining the fields to be extracted and analyzing their textual expression structure, constructing field rule templates based on the results of manual analysis, and extracting entities from structured data according to the rule templates. Relation extraction includes extracting relations from structured data without additional relation extraction, and extracting relations from unstructured data based on sentence structure and plain text tags.

[0150] Relation extraction based on sentence structure includes designing conditional sentence rule templates, matching the antecedent and consequent of conditional sentences, determining the beginning and end of relations, dividing conditional relations for simple conditional sentences by matching antecedent and consequent identifiers using a predefined keyword library, and dividing antecedent and consequent for complex conditional sentences based on specific symbol identifiers; Relation extraction based on plain text tags includes introducing relation tags in the entity annotation stage, using the "BIO" annotation method, and expanding to 15 types of entity-relation joint tags.

[0151] The knowledge representation steps of the knowledge graph construction module include knowledge base mapping of entities and relations and entity linking methods. Knowledge base mapping of entities and relations includes establishing links between entity fields extracted from text and corresponding entities in the knowledge base, and mapping and storing relations extracted from text in the knowledge base through the entities at both ends of the relationship. Entity linking methods include entity matching based on terminology dictionary, entity disambiguation enhanced by relation tags, and entity fusion in the knowledge base.

[0152] Entity matching based on a terminology dictionary uses string Jaccard similarity calculation to match text entities with candidate entities in the terminology dictionary, and combines contextual semantic information and word vector similarity to optimize matching accuracy; entity disambiguation enhanced by relation tags uses relation tags attached to entity references to filter semantically consistent candidate entities, and excludes erroneous links through relation constraints; knowledge base entity fusion normalizes the successfully matched entity references and terminology dictionary entries to generate unique knowledge base entity identifiers, and stores the linking results in a knowledge graph database.

[0153] The knowledge graph construction module includes graph construction and storage steps, including graph structure modeling, graph database storage, and data consistency maintenance. Graph structure modeling uses nodes to represent entities, with node attributes including entity type, name, and association parameters. Edges represent relationships between entities, with edge attributes including relationship type, weight, and time information. Graph database storage uses a graph database to store the knowledge graph, achieving efficient knowledge retrieval and updating through a graph query language, and establishing indexes to optimize the query performance of nodes and edges. Data consistency maintenance sets entity-relationship constraint rules and uses a transaction mechanism to ensure the atomicity and consistency of data writes.

[0154] The real-time data acquisition and preprocessing module is used to collect operational data such as vibration, temperature, pressure, and rotational speed in real time. Data cleaning is used to remove noise data from the acquired data and fill missing values ​​using interpolation or machine learning-based methods. Data normalization is used to map acquired data of different types and dimensions to a unified range through normalization methods.

[0155] The fault hypothesis generation steps of the dynamic reasoning engine building module include extracting the pre-set fault propagation path topology structure in the knowledge graph, receiving abnormal parameter indicators from the real-time monitoring data stream, mapping the abnormal parameter indicators to knowledge graph nodes for path traversal based on the multimodal reasoning mechanism, and outputting a set of candidate fault hypotheses; the multimodal reasoning mechanism includes a rule reasoning module and a case reasoning module.

[0156] The confidence calculation unit of the dynamic inference engine building module includes the following steps: constructing a multidimensional Bayesian network for each fault hypothesis, calculating the joint probability distribution through a probability fusion engine, and outputting the quantitative confidence index of each fault hypothesis. When the confidence of any hypothesis exceeds a preset threshold, a hierarchical alarm signal is triggered. The root node of the multidimensional Bayesian network represents the fault hypothesis ontology, and the child node layer includes real-time monitoring parameter nodes, equipment operating condition parameter nodes, and historical failure rate nodes. The edges establish conditional dependencies based on the causal relationship weights in the knowledge graph. The input layer of the probability fusion engine receives sensor time-series data from the real-time data acquisition module, the knowledge layer calls the prior probability distribution table stored in the knowledge graph, and the calculation layer applies the Bayesian backpropagation algorithm to iteratively update the node confidence.

[0157] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for early warning of faults in hydropower units based on knowledge graphs, characterized in that: include, Construct a knowledge graph database for hydropower units to represent the unit structure, operating information, and fault correlations; Collect real-time operating data of hydropower units and perform data preprocessing; Based on knowledge graphs and collected data, a multimodal reasoning mechanism that integrates rule-based reasoning and case-based reasoning is used to perform path traversal and node matching, identify potential abnormal parameters, and generate a set of candidate fault hypotheses. For each fault hypothesis, a multidimensional Bayesian network is constructed that includes monitoring parameters, equipment operating conditions, and historical failure rates. The fault confidence is calculated by combining the causal weight information between nodes in the knowledge graph. When the confidence of any fault hypothesis exceeds a set threshold, the corresponding fault warning information is output.

2. The knowledge graph-based hydropower unit fault early warning method as described in claim 1, characterized in that: The knowledge graph database includes information extraction based on historical fault records of hydropower units, constructing a knowledge graph database of hydropower unit ontology, including data collection and organization, knowledge extraction, knowledge representation, and graph construction and storage steps.

3. The knowledge graph-based hydropower unit fault early warning method as described in claim 2, characterized in that: The process involves collecting real-time operating data from the hydropower units and performing data preprocessing, including: Acquire real-time operational data and perform normalization processing on the data.

4. The knowledge graph-based hydropower unit fault early warning method as described in claim 3, characterized in that: The process involves identifying potential abnormal parameters and generating a set of candidate fault hypotheses. include, The collected data is matched and fault hypotheses are generated based on the node relationship information in the knowledge graph.

5. The knowledge graph-based hydropower unit fault early warning method as described in claim 2, characterized in that: The knowledge extraction includes entity extraction and relation extraction; The knowledge representation includes knowledge base mapping of entities and relations and entity linking methods; The knowledge base mapping of entities and relationships includes establishing links between entity fields extracted from text and corresponding entities in the knowledge base, and mapping and storing relationships extracted from text in the knowledge base by associating the entities at both ends of the relationship. The entity linking methods include entity matching based on terminology dictionaries, entity disambiguation enhanced by relation tags, and entity fusion based on knowledge bases.

6. The knowledge graph-based hydropower unit fault early warning method as described in claim 5, characterized in that: The terminology dictionary-based entity matching includes using string Jaccard similarity to calculate the matching text entities and candidate entities in the terminology dictionary, and combining contextual semantic information and word vector similarity to optimize the matching accuracy; The relation tag-enhanced entity disambiguation includes filtering semantically consistent candidate entities using relation tags attached to entity references and excluding erroneous links through relation constraints. The knowledge base entity fusion includes normalizing the successfully matched entity references and terminology dictionary entries to generate unique knowledge base entity identifiers, and storing the linking results in a knowledge graph database.

7. The knowledge graph-based hydropower unit fault early warning method as described in claim 4, characterized in that: The generation of fault hypotheses includes matching the collected operational data with the fault propagation path in the knowledge graph, and outputting a set of candidate fault hypotheses based on rule-based reasoning and case-based reasoning mechanisms. The calculation of fault confidence includes constructing a multidimensional Bayesian network containing monitoring parameters, equipment operating conditions, and historical failure rates for each fault hypothesis, calculating the joint probability distribution based on the causal relationship weights in the knowledge graph, obtaining the confidence of each fault hypothesis, and triggering a fault warning when any confidence exceeds a preset threshold.

8. A knowledge graph-based hydropower unit fault early warning system, employing the knowledge graph-based hydropower unit fault early warning method as described in any one of claims 1 to 7, characterized in that, include: The system includes a knowledge graph construction module, a real-time data acquisition and preprocessing module, a dynamic reasoning engine building module, and a system integration and deployment module. The knowledge graph construction module is used to extract information based on the historical fault records of hydropower units and construct a knowledge graph database of hydropower unit ontology, including data collection and organization, knowledge extraction, knowledge representation, and graph construction and storage steps. The real-time data acquisition and preprocessing module is used to process and preprocess the real-time data acquired from the hydropower unit. The dynamic reasoning engine building module is used to construct a fault hypothesis system by leveraging the existing fault propagation paths within the knowledge graph and the abnormal data captured by real-time monitoring, and to accurately calculate the confidence of the fault hypothesis based on the Bayesian network, thereby achieving timely and accurate fault alarm triggering. It includes a fault hypothesis generation unit and a confidence calculation unit. The fault hypothesis generation unit is used to perform path traversal and node matching based on knowledge graphs and collected data, and to identify potential abnormal parameters and generate a set of candidate fault hypotheses by adopting a multimodal reasoning mechanism that integrates rule reasoning and case reasoning. The confidence calculation unit is used to construct a multidimensional Bayesian network containing monitoring parameters, equipment operating conditions and historical failure rates for each fault hypothesis, and calculate the fault confidence by combining the causal weight information between nodes in the knowledge graph. When the confidence of any fault hypothesis exceeds the set threshold, the corresponding fault warning information is output. The system integration and deployment module is used to integrate the knowledge graph construction module, data acquisition and preprocessing module, and dynamic reasoning engine module to form a complete hydropower unit fault early warning system.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the knowledge graph-based hydropower unit fault early warning method according to any one of claims 1 to 7.

10. A 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 steps of the knowledge graph-based hydropower unit fault early warning method according to any one of claims 1 to 7.

Images