A Reliability Evaluation Method and System for Hydrogen Fuel Cells Based on Multi-Source Heterogeneous Data Analysis
By combining knowledge graphs and deep learning models to analyze multi-source heterogeneous data of hydrogen fuel cells, a knowledge graph of failure and fault propagation is constructed, which solves the problem of difficulty in utilizing unstructured data in existing technologies and realizes efficient and accurate reliability evaluation and risk assessment of hydrogen fuel cell systems.
Patent Information
- Application Number
- CN202611257686.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for evaluating the reliability of hydrogen fuel cells are unable to fully utilize the large amount of unstructured data generated during operation, cannot effectively characterize the causal propagation and correlation paths between complex failure modes, and are difficult to adapt to the dynamic risk assessment needs of multi-condition and multi-dimensional operating characteristics.
By combining knowledge graphs and deep learning semantic analysis models, multi-source heterogeneous data of hydrogen fuel cells are analyzed to construct a knowledge graph of hydrogen fuel cell failure and fault propagation. The risk of cascade failure is calculated by the node centrality index, and the risk index is mapped to the [0,1] interval using the Sigmoid normalization function to achieve a comprehensive reliability score.
It enables dynamic reliability evaluation of hydrogen fuel cell systems, improves data parsing efficiency by approximately 60%, overcomes the limitations of traditional methods, and enhances the accuracy and intelligence of evaluation results, enabling the identification of critical failure nodes and potential high-risk paths.
Smart Images

Figure CN123120791A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hydrogen fuel cell data analysis technology, specifically to a method, system, and readable storage medium for evaluating the reliability of hydrogen fuel cells. Background Technology
[0002] Hydrogen fuel cells, as a key technology for energy transition and green, low-carbon development, have been rapidly adopted in various fields such as low-altitude aircraft, automobiles, power plants, and distributed energy, thanks to their advantages of high energy density, zero emissions, and long range. However, during long-term operation, critical components of hydrogen fuel cells, such as the internal stack, gas supply system, and cooling system, are susceptible to complex operating conditions, leading to various failure modes and potential malfunctions, which can result in performance degradation or even system shutdown. Therefore, conducting reliability assessments and risk predictions is a core issue in ensuring the safe and stable operation of hydrogen fuel cell systems.
[0003] Existing reliability assessment methods are mostly based on experimental statistics, mechanistic modeling, or condition monitoring signal analysis. On the one hand, these methods heavily rely on prior knowledge and structured data, making it difficult to fully utilize the large amount of unstructured data generated during operation, such as test records, failure cases, and maintenance reports. On the other hand, traditional methods fail to adequately reveal the coupling relationships across components and mechanisms, and cannot effectively characterize the causal propagation and correlation paths between complex failure modes, resulting in significant deficiencies in the accuracy and intelligence of the assessment results. Furthermore, given the multi-condition and multi-dimensional operating characteristics of hydrogen fuel cells in different application scenarios, traditional methods are ill-suited to the dynamic risk assessment needs in complex environments.
[0004] In recent years, breakthroughs in deep learning in the fields of natural language processing and semantic understanding have provided new ways to automatically extract knowledge from unstructured text. At the same time, knowledge graphs, due to their excellent knowledge organization and reasoning capabilities, can effectively describe entities and their relational networks and support complex reasoning analysis. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a method and system for evaluating the reliability of hydrogen fuel cells based on multi-source heterogeneous data analysis. It combines knowledge graphs with deep learning semantic analysis models to analyze multi-source heterogeneous data of hydrogen fuel cells under complex environments.
[0006] To solve the above-mentioned technical problems, the technical solution provided by the present invention is as follows: This invention provides a method for evaluating the reliability of hydrogen fuel cells based on multi-source heterogeneous data analysis, comprising the following steps: S1. Collect multi-source heterogeneous data from the hydrogen fuel cell system and perform preprocessing. Among them, multi-source heterogeneous data includes unstructured text data and structured time-series data; S2. Utilize deep learning semantic analysis models to perform entity recognition and relation extraction on entities and entity relationships in unstructured text data; where: entities include fault modes, fault causes, fault effects, component units, and operating conditions, and entity relationships include "cause," "affect," "originate from," and "related to"; Entity disambiguation and entity linking are performed on the entities and entity relationships obtained from entity identification and relation extraction. Triple normalization is used to construct a knowledge graph of hydrogen fuel cell failure and fault propagation of the entities and entity relationships after entity disambiguation and entity linking. S3. Calculate the node centrality index of the knowledge graph of hydrogen fuel cell failure and fault propagation by combining the degree centrality, betweenness centrality and eigenvector centrality of the nodes. Calculate the cascade failure risk index using the node centrality index; S4. Use the Sigmoid normalization function to normalize the cascade failure risk index. Mapping to the [0,1] interval yields the overall reliability score of the hydrogen fuel cell system. : In the formula, For steepness parameter, The threshold value is used.
[0007] Furthermore, in step S1, the preprocessing of multi-source heterogeneous data includes: cleaning, time alignment, noise reduction, missing value imputation and labeling; Among them, unstructured text data in multi-source heterogeneous data is segmented into sentences, words, and standardized, and preliminary entity candidate extraction is carried out using domain dictionaries and regular expressions; Structured time-series data from multi-source heterogeneous data is aligned with text timestamps to generate event sequences; an active learning-based human-computer interaction annotation mechanism is used to label low-confidence automatically extracted results, and a confidence threshold is set. When the model outputs confidence level Manual confirmation or correction may be triggered at any time.
[0008] Furthermore, in step S2, the deep learning semantic analysis model is a pre-trained transformer structure, in which: Entity recognition employs a BERT encoder with a conditional random field or a token-classifier head, along with cross-entropy. As a loss function; Cross-entropy for: In the formula, For sequence position index, For entity categories, For tags, To predict probabilities; Relation extraction employs a combination architecture of sentence-level encoder or cross-sentence encoder and classifier, while also utilizing cross-entropy. As a loss function; Semantic similarity calculation uses cosine similarity. ,in, It is represented as a text vector.
[0009] Furthermore, in step S2, the specific method for constructing the knowledge graph of hydrogen fuel cell failure and fault propagation through entity disambiguation, entity linking, and triple normalization is as follows: The extracted entities are mapped to the domain ontology or vocabulary, and the string similarity and vector similarity are used together to determine whether the entities are the same concept in order to achieve entity disambiguation. If the similarity calculated using a combination of string similarity and vector similarity exceeds the set threshold, then automatic entity linking is executed. If the similarity does not reach the set threshold, manual review will be triggered, and the entity link will be manually implemented. The entities and their relationships after entity disambiguation and linking are assembled into a triplet data structure to construct a knowledge graph G for hydrogen fuel cell failure and fault propagation: G=(V,E,W); where, For a collection of entities, Let be a set of directed edges. Let be the set of edge weights.
[0010] Furthermore, in step S3, the node centrality index : ; In the formula, For coefficients; For degree centrality, , The total number of nodes; For betweenness centrality, ,in, for arrive The number of shortest paths, To pass The number of shortest paths; For eigenvector centrality, satisfy , It is an adjacency matrix. For eigenvalues; Cascade Failure Risk Indicators for: ; In the formula, For nodes The inherent vulnerability score, For nodes The probability of unit failure.
[0011] Furthermore, step S3 is replaced with: A knowledge graph embedding algorithm is used to map entities and relations in the knowledge graph of hydrogen fuel cell failure and fault propagation into entity vectors and relation vectors, respectively. Step S4 is replaced with: Build a regression / classification model : ; in, These are the entity vectors and relation vectors extracted from the knowledge graph and time series analysis of hydrogen fuel cell failure and fault propagation. For hyperparameters, The overall reliability score for prediction; Meanwhile, mean squared error (MSE) or cross-entropy is used. As a regression / classification model The loss function is determined, and the hyperparameters of the regression / classification model are determined through cross-validation or Bayesian optimization. ; Cross-entropy for: ; Entity vectors and relation vectors extracted from the knowledge graph of hydrogen fuel cell failure and fault propagation Input regression / classification model The overall reliability score of the hydrogen fuel cell system is output. .
[0012] Furthermore, in step S3, the knowledge graph embedding algorithm is the TransE algorithm; TransE scoring function of the TransE algorithm for: In the formula, , These are the embedding vectors for the head and tail entities, respectively. It is a relation vector; Training is performed by minimizing the energy of the edges and using negative sampling loss.
[0013] A reliability evaluation system for hydrogen fuel cells based on multi-source heterogeneous data analysis includes a data acquisition module, a data preprocessing module, a semantic analysis module, a knowledge graph construction module, a knowledge graph storage module, a reasoning and quantification module, and a reliability fusion evaluation module, which are connected sequentially via a data bus or API. The semantic analysis module includes a deep learning semantic analysis model. The data acquisition module sends the collected multi-source heterogeneous data to the data preprocessing module for preprocessing, and the semantic analysis module performs entity recognition and relation extraction. The knowledge graph construction module combines the recognized entities and extracted relations to construct a knowledge graph of hydrogen fuel cell failure and fault propagation and stores it in the knowledge graph storage module. The reasoning and quantification module calculates the cascade failure risk index of the hydrogen fuel cell failure and fault propagation knowledge graph, and the reliability fusion assessment module combines the cascade failure risk index to output a comprehensive reliability score of the hydrogen fuel cell system.
[0014] Furthermore, the knowledge graph storage module adopts triple storage, and the triple storage is equipped with a retrieval interface based on SPARQL or graph query language.
[0015] A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described hydrogen fuel cell reliability evaluation method.
[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The hydrogen fuel cell reliability evaluation method and system based on multi-source heterogeneous data analysis provided by this invention combines a deep learning semantic analysis model, a knowledge graph of hydrogen fuel cell failure and fault propagation, and a Sigmoid normalization function. The deep learning semantic analysis model performs entity recognition and relation extraction on multi-source heterogeneous data (a combination of unstructured text data and structured time-series data), facilitating the construction of a knowledge graph of hydrogen fuel cell failure and fault propagation. Using this knowledge graph, node centrality, path propagation probability, and system cascade failure risk indicators are calculated. Then, the Sigmoid normalization function maps these cascade failure risk indicators to the [0,1] interval, yielding a comprehensive reliability score for the hydrogen fuel cell system. This enables dynamic reliability evaluation of hydrogen fuel cell systems; compared to traditional analysis methods that rely on human experience, the data parsing efficiency is improved by approximately 60%.
[0017] 2. The hydrogen fuel cell reliability evaluation method and system based on multi-source heterogeneous data analysis provided by this invention can perform entity recognition and relation extraction (i.e., identify fault modes, failure mechanisms and extract influencing factors) from unstructured text data such as test records, failure cases and operation and maintenance logs through a deep semantic learning model. This breaks through the limitation of traditional methods that can only rely on structured time series data or human experience, and realizes the application of multi-source heterogeneous data analysis. Attached Figure Description
[0018] Figure 1 This is a schematic diagram of the process structure of Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the system structure of Embodiment 1 of the present invention. Detailed Implementation
[0019] The technical solutions of the present invention will now be clearly and completely described 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 are within the scope of protection of the present invention.
[0020] Example 1 like Figure 1 As shown, a method for evaluating the reliability of hydrogen fuel cells based on multi-source heterogeneous data analysis includes the following steps: S1. Collect multi-source heterogeneous data from the hydrogen fuel cell system and perform preprocessing. Among them, the multi-source heterogeneous data includes unstructured text data and structured time-series data. The unstructured text data includes real-time test records, fault reports, maintenance logs and typical failure cases. These data cover multi-dimensional operating parameters such as voltage, current and temperature, as well as failure information such as fuel cell material aging, membrane rupture and cooling system leakage.
[0021] Preprocessing of multi-source heterogeneous data includes: cleaning, time alignment, noise reduction, missing value imputation and labeling; Among them, unstructured text data in multi-source heterogeneous data is segmented into sentences, words, and standardized, and preliminary entity candidate extraction is carried out using domain dictionaries and regular expressions; Structured time-series data from multi-source heterogeneous data is aligned with text timestamps to generate event sequences; an active learning-based human-computer interaction annotation mechanism is used to label low-confidence automatically extracted results, and a confidence threshold is set. When the model outputs confidence level Manual confirmation or correction may be triggered at any time.
[0022] S2. Utilize deep learning semantic analysis models to perform entity recognition and relation extraction on entities and entity relationships in unstructured text data; where: entities include fault modes, fault causes, fault effects, component units, and operating conditions, and entity relationships include "cause," "affect," "originate from," and "related to"; The deep learning semantic analysis model is a pre-trained transformer structure, in which: Entity recognition employs a BERT encoder with a conditional random field or a token-classifier head, along with cross-entropy. As a loss function; Cross-entropy for: In the formula, For sequence position index, For entity categories, For tags, To predict probabilities; Relation extraction employs a combination architecture of sentence-level encoder or cross-sentence encoder and classifier, while also utilizing cross-entropy. As a loss function; Semantic similarity calculation uses cosine similarity. ,in, It is represented as a text vector.
[0023] The extracted entities are mapped to the domain ontology or vocabulary, and the string similarity and vector similarity are used together to determine whether the entities are the same concept in order to achieve entity disambiguation. If the similarity calculated using a combination of string similarity and vector similarity is higher than the set threshold, automatic entity linking is executed; if the similarity does not reach the set threshold, manual review is triggered, and entity linking is implemented manually. The entities and their relationships after entity disambiguation and linking are assembled into a triplet data structure to construct a knowledge graph G for hydrogen fuel cell failure and fault propagation: G=(V,E,W); where, For a collection of entities, Let be a set of directed edges. Let be the set of edge weights.
[0024] S3. Calculate the node centrality index of the hydrogen fuel cell failure and fault propagation knowledge graph by combining the degree centrality, betweenness centrality, and eigenvector centrality of nodes in the graph. : In the formula, For coefficients; For degree centrality, , The total number of nodes; For betweenness centrality, ,in, for arrive The number of shortest paths, To pass The number of shortest paths; For eigenvector centrality, satisfy , It is an adjacency matrix. For eigenvalues; Cascade failure risk index calculated using node centrality index. : In the formula, For nodes The inherent vulnerability score, For nodes The probability of unit failure.
[0025] S4. Use the Sigmoid normalization function to normalize the cascade failure risk index. Mapping to the [0,1] interval yields the overall reliability score of the hydrogen fuel cell system. : In the formula, For steepness parameter, The threshold value is used.
[0026] Overall reliability score The closer the value is to 1, the higher the reliability of the hydrogen fuel system; the closer it is to 0, the lower the reliability of the hydrogen fuel system.
[0027] This embodiment takes a hydrogen fuel cell carried by a low-altitude unmanned aerial vehicle (UAV) as an example to further illustrate the technical solution of the present invention; the low-altitude unmanned aerial vehicle (UAV) is equipped with key components such as a fuel cell stack with high power density, a hydrogen supply system, and an air compression and cooling module, and operates for a long time in multiple working conditions, including take-off, cruise, hovering and landing phases under different temperature, humidity and load conditions.
[0028] During actual operation, hydrogen fuel cell systems generate a large amount of unstructured operational data, such as real-time test records, fault reports, maintenance logs, and typical failure cases. This data covers multi-dimensional operating parameters such as voltage, current, and temperature, as well as failure information such as aging of stack materials, membrane rupture, and cooling system leakage.
[0029] Due to the stringent requirements on weight, power density, and endurance of drones, their hydrogen fuel cell systems demand extremely high reliability and safety. A localized failure can propagate and lead to system-wide shutdown, impacting drone mission execution. Therefore, it is necessary to systematically model the failure modes, causes, and impact paths of hydrogen fuel cell systems and conduct reliability assessments to provide a scientific basis for operation and maintenance decisions, preventative maintenance, and system optimization.
[0030] The specific process of evaluating the reliability of a hydrogen fuel cell carried by a low-altitude unmanned aerial vehicle (UAV) according to the reliability evaluation method of this embodiment is as follows: Step 1, Data Acquisition and Preprocessing: Collect multi-source heterogeneous data from the UAV hydrogen fuel cell system at different flight phases (operating parameters such as stack output voltage, current, and temperature; hydrogen supply system pressure and flow rate records, and cooling system temperature records; maintenance and failure report texts, covering cases such as membrane rupture, plate corrosion, and stack short circuits). Clean and standardize the collected unstructured text data, and annotate key entities such as failure modes, failure causes, and impact ranges to ensure the quality of input data.
[0031] Step 2: Knowledge Extraction and Graph Construction Entity recognition and relation extraction are performed on the processed text using a deep learning semantic analysis model. Entity categories: Failure mode (membrane rupture, sudden drop in stack voltage), Failure cause (high temperature, insufficient hydrogen purity), Failure impact (performance degradation, mission interruption), Component unit (fuel stack, cooling module).
[0032] Relationship categories: leading to, affecting, originating from. Constructing a knowledge graph of hydrogen fuel cell failure and fault propagation. Visualizing the knowledge graph facilitates the analysis of key failure nodes and their propagation paths.
[0033] Step 3: Knowledge Graph Reasoning and Analysis: Node centrality calculation It is used to identify the most critical entity nodes in the fault propagation.
[0034] Path propagation probability calculation ,in, The relation weight is calculated based on semantic similarity or the frequency of relation occurrence.
[0035] System Cascade Failure Risk Index In the formula, For nodes The inherent vulnerability score, For nodes The unit failure probability is used to measure the risk of local faults spreading to the system level.
[0036] Step 4: Comprehensive reliability evaluation.
[0037] Using the Sigmoid normalization function to measure the risk index of cascading failure Mapping to the [0,1] interval yields the overall reliability score of the hydrogen fuel cell system. ; ,in, For steepness parameter, The threshold value is used.
[0038] The obtained comprehensive reliability score The closer the value is to 1, the higher the reliability of the hydrogen fuel system; the closer it is to 0, the lower the reliability of the hydrogen fuel system.
[0039] In this application scenario, the method and system of the present invention can intelligently analyze the operating data of the UAV hydrogen fuel cell system, construct a knowledge graph of failure and fault propagation, quantify key fault propagation paths and system vulnerabilities, and output reliability scores and risk levels, thus providing support for the safe and reliable operation of UAVs.
[0040] This embodiment utilizes a deep learning semantic analysis model to extract entities and relationships from unstructured operational data, enabling automatic identification and modeling of fault modes, failure causes, and their impact paths. Compared to traditional analysis methods relying on human experience, data parsing efficiency is improved by approximately 60%, and the risk of missed detections and misjudgments is significantly reduced. The constructed knowledge graph of hydrogen fuel cell failure and fault propagation fully characterizes the complex coupling relationships across components and mechanisms within the system. Through quantitative calculations of node centrality, path propagation probability, and the system cascade failure risk index, key fault nodes and potentially high-risk paths can be clearly identified. For example, during testing, the correlation propagation probability between membrane rupture and cooling system anomalies was found to be as high as 0.78, and the system-level risk index reached 0.65 during high-load, long-endurance missions, an improvement of approximately 30% compared to traditional methods without knowledge graph modeling, effectively revealing potential system failures in advance.
[0041] By employing weighted fusion and machine learning models to comprehensively analyze multi-dimensional indicators, this invention outputs reliability scores and risk levels, enabling dynamic reliability evaluation of UAV hydrogen fuel cell systems. Experimental results demonstrate that the reliability prediction accuracy exceeds 92% under different operating conditions. The evaluation results intuitively reflect the contribution of key components to the overall system reliability, providing a scientific basis for operation and maintenance strategy formulation. Furthermore, the visualization of the knowledge graph and the display of heatmaps make complex fault propagation relationships readily apparent, enhancing the interpretability and operability of the evaluation results and providing reliable support for UAV mission scheduling, preventative maintenance, and system optimization. Overall, this invention not only significantly improves the accuracy and efficiency of reliability evaluation for hydrogen fuel cell systems but also achieves an intelligent closed loop from data analysis and fault modeling to risk quantification, providing a scalable technical solution for multi-scenario applications.
[0042] Example 2 like Figure 2 As shown, a hydrogen fuel cell reliability evaluation system based on multi-source heterogeneous data analysis includes a data acquisition module, a data preprocessing module, a semantic analysis module, a knowledge graph construction module, a knowledge graph storage module, a reasoning and quantification module, and a reliability fusion evaluation module, which are connected sequentially via a data bus or API. The semantic analysis module includes a deep learning semantic analysis model. The data acquisition module sends the collected multi-source heterogeneous data to the data preprocessing module for preprocessing, and the semantic analysis module performs entity recognition and relation extraction. The knowledge graph construction module combines the recognized entities and extracted relations to construct a knowledge graph of hydrogen fuel cell failure and fault propagation and stores it in the knowledge graph storage module. The reasoning and quantification module calculates the cascade failure risk index of the hydrogen fuel cell failure and fault propagation knowledge graph, and the reliability fusion assessment module combines the cascade failure risk index to output a comprehensive reliability score of the hydrogen fuel cell system.
[0043] The knowledge graph storage module uses triplet storage, and the triplet storage is equipped with a retrieval interface based on SPARQL or graph query language.
[0044] Example 3 The difference between Example 3 and Example 1 is that Example 3 uses a readable storage medium on which a computer program is stored, which, when executed by a processor, implements the above-described hydrogen fuel cell reliability evaluation method.
[0045] Example 4 The difference between Example 4 and Example 1 is that step S3 is: A knowledge graph embedding algorithm is used to map entities and relations in the knowledge graph of hydrogen fuel cell failure and fault propagation into entity vectors and relation vectors, respectively. The knowledge graph embedding algorithm is the TransE algorithm; TransE scoring function of the TransE algorithm for: In the formula, , These are the embedding vectors for the head and tail entities, respectively. It is a relation vector; Training is performed by minimizing the edge energy and negative sampling loss; Step S4 is: Build a regression / classification model : ; in, These are the entity vectors and relation vectors extracted from the knowledge graph and time series analysis of hydrogen fuel cell failure and fault propagation. For hyperparameters, The overall reliability score for prediction; Meanwhile, mean squared error (MSE) or cross-entropy is used. As a regression / classification model The loss function is determined, and the hyperparameters of the regression / classification model are determined through cross-validation or Bayesian optimization. ; Cross-entropy for: ; Entity vectors and relation vectors extracted from the knowledge graph of hydrogen fuel cell failure and fault propagation Input regression / classification model The overall reliability score of the hydrogen fuel cell system is output. .
[0046] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present invention should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A reliability evaluation method for hydrogen fuel cells based on multi-source heterogeneous data analysis, characterized in that, Includes the following steps: S1. Collect multi-source heterogeneous data from the hydrogen fuel cell system and perform preprocessing. Among them, multi-source heterogeneous data includes unstructured text data and structured time-series data; S2. Entity identification and relation extraction are performed on entities and entity relations in unstructured text data using a deep learning semantic analysis model. Then, entity disambiguation and entity linking are performed, and triple normalization is used to construct a knowledge graph of hydrogen fuel cell failure and fault propagation based on the entities and entity relations. Among them: entities include failure modes, failure causes, failure effects, component units, and operating conditions; entity relationships include "cause", "affect", "originate from", and "related to". S3. Calculate the node centrality index of the knowledge graph of hydrogen fuel cell failure and fault propagation by combining the degree centrality, betweenness centrality and eigenvector centrality of the nodes. Calculate the cascade failure risk index using the node centrality index; S4. Use the Sigmoid normalization function to normalize the cascade failure risk index. Mapping to the [0,1] interval yields the overall reliability score of the hydrogen fuel cell system. : In the formula, For steepness parameter, The threshold value is used.
2. The hydrogen fuel cell reliability evaluation method based on multi-source heterogeneous data analysis according to claim 1, characterized in that, In step S1, the preprocessing of multi-source heterogeneous data includes: cleaning, time alignment, noise reduction, missing value imputation and labeling; Among them, unstructured text data in multi-source heterogeneous data is segmented into sentences, words, and standardized, and preliminary entity candidate extraction is carried out using domain dictionaries and regular expressions; Structured time-series data from multi-source heterogeneous data is aligned with text timestamps to generate event sequences; an active learning-based human-computer interaction annotation mechanism is used to label low-confidence automatically extracted results, and a confidence threshold is set. When the model outputs confidence level Manual confirmation or correction may be triggered at any time.
3. The hydrogen fuel cell reliability evaluation method based on multi-source heterogeneous data analysis according to claim 1, characterized in that, In step S2, the deep learning semantic analysis model is a pre-trained transformer structure, in which: Entity recognition employs a BERT encoder with a conditional random field or a token-classifier head, along with cross-entropy. As a loss function; Cross-entropy for: In the formula, For sequence position index, For entity categories, For tags, To predict probabilities; Relation extraction employs a combination architecture of sentence-level encoder or cross-sentence encoder and classifier, while also utilizing cross-entropy. As a loss function; Semantic similarity analysis uses cosine similarity. ,in, It is represented as a text vector.
4. The hydrogen fuel cell reliability evaluation method based on multi-source heterogeneous data analysis according to claim 3, characterized in that, In step S2, the specific method for constructing a knowledge graph of hydrogen fuel cell failure and fault propagation through entity disambiguation, entity linking, and triplet normalization is as follows: The extracted entities are mapped to the domain ontology or vocabulary, and the string similarity and vector similarity are used together to determine whether the entities are the same concept in order to achieve entity disambiguation. If the similarity calculated using a combination of string similarity and vector similarity exceeds the set threshold, then automatic entity linking is executed. If the similarity does not reach the set threshold, manual review will be triggered, and the entity link will be manually implemented. The entities and their relationships after entity disambiguation and linking are assembled into a triplet data structure to construct a knowledge graph G for hydrogen fuel cell failure and fault propagation: G=(V,E,W); where, For a collection of entities, Let be a set of directed edges. Let be the set of edge weights. The edge weight is denoted as .
5. The hydrogen fuel cell reliability evaluation method based on multi-source heterogeneous data analysis according to claim 1, characterized in that, In step S3, the node centrality index : ; In the formula, These are the degree centrality coefficient, betweenness centrality coefficient, and eigenvector centrality coefficient, respectively. For degree centrality, , The total number of nodes; For betweenness centrality, ,in, for arrive The number of shortest paths, To pass The number of shortest paths; For eigenvector centrality, satisfy , It is an adjacency matrix. For eigenvalues; Cascade Failure Risk Indicators for: ; In the formula, For nodes The inherent vulnerability score, For nodes The probability of unit failure.
6. The hydrogen fuel cell reliability evaluation method based on multi-source heterogeneous data analysis according to claim 1, characterized in that, Step S3 is replaced with: A knowledge graph embedding algorithm is used to map entities and relations in the knowledge graph of hydrogen fuel cell failure and fault propagation into entity vectors and relation vectors, respectively. Step S4 is replaced with: Build a regression / classification model : ; in, These are the entity vectors and relation vectors extracted from the knowledge graph and time series analysis of hydrogen fuel cell failure and fault propagation. For hyperparameters, The overall reliability score for prediction; Meanwhile, mean squared error (MSE) or cross-entropy is used. As a regression / classification model The loss function is determined, and the hyperparameters of the regression / classification model are determined through cross-validation or Bayesian optimization. ; Cross-entropy for: ; Entity vectors and relation vectors extracted from the knowledge graph of hydrogen fuel cell failure and fault propagation Input regression / classification model The overall reliability score of the hydrogen fuel cell system is output. .
7. The hydrogen fuel cell reliability evaluation method based on multi-source heterogeneous data analysis according to claim 6, characterized in that, In step S3, the knowledge graph embedding algorithm is the TransE algorithm; TransE scoring function of the TransE algorithm for: In the formula, , These are the embedding vectors for the head and tail entities, respectively. It is a relation vector; Training is performed by minimizing the energy of the edges and using negative sampling loss.
8. A reliability evaluation system based on the hydrogen fuel cell reliability evaluation method based on multi-source heterogeneous data analysis as described in claim 1, characterized in that: It includes a data acquisition module, a data preprocessing module, a semantic analysis module, a knowledge graph construction module, a knowledge graph storage module, a reasoning and quantification module, and a reliability fusion evaluation module, which are connected sequentially via a data bus or API. The semantic analysis module includes a deep learning semantic analysis model. The data acquisition module sends the collected multi-source heterogeneous data to the data preprocessing module for preprocessing, and the semantic analysis module performs entity recognition and relation extraction. The knowledge graph construction module combines the identified entities and extracted relationships to construct a knowledge graph of hydrogen fuel cell failure and fault propagation and stores it in the knowledge graph storage module. The reasoning and quantification module calculates the cascade failure risk index of the knowledge graph of hydrogen fuel cell failure and fault propagation, and the reliability fusion assessment module combines the cascade failure risk index to output the comprehensive reliability score of the hydrogen fuel cell system.
9. The hydrogen fuel cell reliability evaluation system based on multi-source heterogeneous data analysis according to claim 8, characterized in that, The knowledge graph storage module uses triplet storage, and the triplet storage is equipped with a retrieval interface based on SPARQL or graph query language.
