CN118585650A - An equipment fault information retrieval system based on big data storage technology - Google Patents

Abstract

The invention discloses a device fault information retrieval system based on big data storage technology, which comprises device fault diagnosis based on a knowledge graph, expert knowledge base fusion and intelligent question-answering based on a big model, system upgrading and optimizing construction, the equipment fault diagnosis method based on the knowledge graph comprises the steps of building entity relation based on multi-source heterogeneous equipment fault diagnosis information, building the knowledge graph facing to equipment fault diagnosis, reasoning the knowledge graph facing to equipment fault diagnosis, studying and judging equipment fault types based on a classification decision tree algorithm, studying and judging alarm information based on semantic rule matching, and studying knowledge-driven fault cause auxiliary analysis. The invention belongs to the technical field of equipment fault diagnosis, and particularly provides an equipment fault information retrieval system based on a big data storage technology.

Description

An equipment fault information retrieval system based on big data storage technology

Technical Field

The present invention belongs to the technical field of equipment fault diagnosis, and specifically refers to an equipment fault information retrieval system based on big data storage technology.

Background Art

The Guodian Electric Power Knowledge Base and Expert Knowledge System Research and Application Project integrates multi-source heterogeneous data, accesses, processes, cleans, annotates and establishes a knowledge base based on the Guodian Big Data Storage Platform. With the artificial intelligence big model as the base, it realizes the functions of intelligent processing, retrieval, question and answer based on the big model. The research content includes: knowledge discovery technology of multimodal business data, knowledge fusion technology of cross-domain expert knowledge base, new generation of intelligent knowledge search engine technology, knowledge graph visualization application service platform research and development and application, fault diagnosis technology based on knowledge graph and expert knowledge base fusion and intelligent question and answer technology based on big model. On this basis, the optimization direction of system construction is proposed, including: automatic update mechanism of system data, automatic scoring mechanism of documents based on user behavior, continuous update mechanism of plug-in knowledge base and continuous construction mechanism of enterprise intelligent knowledge management.

The knowledge base and expert knowledge system project integrates multi-source heterogeneous data, accesses, processes, cleans, annotates and establishes a knowledge base based on the Guodian big data storage platform. Based on the artificial intelligence big model, it applies knowledge graphs, natural language processing and other technologies to achieve intelligent retrieval, question-answering, reading and creation of data, and improve the overall knowledge management and application level of Guodian.

Summary of the invention

In order to solve the above-mentioned existing problems, the present invention provides an equipment fault information retrieval system based on big data storage technology.

The technical solution adopted by the present invention is as follows: The equipment fault information retrieval system of the present invention is based on big data storage technology, including equipment fault diagnosis based on knowledge graph, expert knowledge base fusion and intelligent question and answer based on big model, and system upgrade and optimization construction.

The equipment fault diagnosis based on knowledge graph includes entity relationship construction based on multi-source heterogeneous equipment fault diagnosis information, knowledge graph construction for equipment fault diagnosis, knowledge graph reasoning for equipment fault diagnosis, equipment fault type analysis based on classification decision tree algorithm, alarm information analysis based on semantic rule matching, and research on knowledge-driven auxiliary analysis of fault causes.

The large-model-based expert knowledge base fusion and intelligent question and answering include data collection and preprocessing, large-model question and answering construction, small sample data fine-tuning and optimization, context awareness, semantic matching and answer generation.

The system upgrade and optimization construction includes an automatic update mechanism for institutional data, an automatic document scoring mechanism based on user behavior, a continuous update mechanism for the plug-in vector knowledge base, and a continuous construction mechanism for the enterprise intelligent knowledge management ecosystem.

Furthermore, the entity relationship construction based on multi-source heterogeneous equipment fault diagnosis information includes studying the construction of power plant equipment fault diagnosis knowledge ontology, studying the entity extraction and linking of equipment fault diagnosis information, and studying the entity relationship extraction in the field of power plant equipment fault diagnosis.

The construction of the knowledge graph for equipment fault diagnosis includes studying the knowledge fusion of equipment fault diagnosis information and studying large-scale knowledge graph hybrid storage management technology.

The knowledge graph reasoning for equipment fault diagnosis includes research on supporting multi-form natural language queries, research on knowledge reasoning based on equipment fault diagnosis knowledge graph, and research on knowledge path mining of equipment fault diagnosis knowledge graph.

Furthermore, the specific steps of data collection and preprocessing include:

(1) Data collection and organization;

(2) question-answer pair generation;

(3) Cleaning and pretreatment;

(4) Data partitioning and test set construction.

Furthermore, the plug-in vector knowledge base continuous update mechanism includes the following steps:

(1) Data collection and screening: First, data is collected through various channels, including but not limited to the Internet, databases, literature, social media, etc. Then, the collected data is automatically screened and verified to ensure that only high-quality and reliable information is included in the knowledge base.

(2) Automated processing: Utilize natural language processing and machine learning technologies to automatically process the collected data, including entity recognition, relationship extraction, topic modeling and other technologies, so as to transform the raw data into structured knowledge.

(3) Manual review and correction: Although automated processing can efficiently process large amounts of data, manual review and correction are still required. The team reviews the results of automated processing, corrects errors, and supplements missing information to ensure the accuracy and completeness of the knowledge base.

(4) Regular automatic updates: The knowledge base will be updated regularly to reflect the latest information and developments. The frequency of updates can be determined based on demand and the real-time nature of the data source, and can be updated daily, weekly, or monthly.

(5) User feedback: The knowledge base accepts user feedback and makes corresponding adjustments and updates based on user suggestions and needs.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a flowchart of the large model question answering for semantic matching and answer generation in this solution;

FIG2 is a flow chart of the automatic update mechanism of the system data of this scheme.

DETAILED DESCRIPTION

The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

As shown in Figures 1 and 2, the equipment fault information retrieval system based on big data storage technology proposed in this solution includes equipment fault diagnosis based on knowledge graph, expert knowledge base fusion and intelligent question and answer based on large models, and system upgrade and optimization construction.

The equipment fault diagnosis based on knowledge graph includes entity relationship construction based on multi-source heterogeneous equipment fault diagnosis information, knowledge graph construction for equipment fault diagnosis, knowledge graph reasoning for equipment fault diagnosis, equipment fault type analysis based on classification decision tree algorithm, alarm information analysis based on semantic rule matching, and research on knowledge-driven auxiliary analysis of fault causes.

Equipment failure type analysis based on classification decision tree algorithm: Research on equipment failure type analysis technology based on classification decision tree algorithm will analyze data sets of different equipment failure characteristics, such as vibration, temperature, current and other parameters, select the most representative and distinguishing features, and establish a decision tree model for equipment failure classification. The model will judge the equipment status layer by layer, so as to accurately determine the equipment failure type, such as circuit failure, mechanical failure, etc., to help business personnel quickly identify and solve equipment failure problems.

Alarm information analysis based on semantic rule matching: Research on alarm information analysis technology based on semantic rule matching will collect and organize equipment alarm information and its corresponding processing rules, and build a semantic relationship model between alarm information and processing rules. When new alarm information enters the diagnosis engine, the model will match its semantic rules, quickly and accurately identify the fault type, and provide corresponding processing suggestions to help business personnel quickly respond to and handle equipment failures.

Research on knowledge-driven auxiliary analysis of fault causes: Research on knowledge-driven auxiliary analysis of fault causes, using the semantic analysis model of defect fault information to identify faults and key defect information, combining equipment fault type analysis technology with fault knowledge graph reasoning drive technology. Based on the complete equipment fault diagnosis knowledge graph constructed by knowledge such as equipment historical faults, maintenance tests, historical fault handling solutions, relevant technical standards, and original data, the possible causes of equipment failures can be identified. This technology will provide engineers with more in-depth and comprehensive auxiliary analysis of fault diagnosis, and improve the efficiency and accuracy of equipment fault handling.

The large-model-based expert knowledge base fusion and intelligent question and answering include data collection and preprocessing, large-model question and answering construction, small sample data fine-tuning and optimization, context awareness, semantic matching and answer generation.

The system upgrade and optimization construction includes an automatic update mechanism for institutional data, an automatic document scoring mechanism based on user behavior, a continuous update mechanism for the plug-in vector knowledge base, and a continuous construction mechanism for the enterprise intelligent knowledge management ecosystem.

Furthermore, the entity relationship construction based on multi-source heterogeneous equipment fault diagnosis information includes studying the construction of power plant equipment fault diagnosis knowledge ontology, studying the entity extraction and linking of equipment fault diagnosis information, and studying the entity relationship extraction in the field of power plant equipment fault diagnosis.

Research on the construction of knowledge ontology for power plant equipment fault diagnosis: Research on the knowledge ontology construction method of text data source based on correlation mining and hierarchical clustering technology, and capture the abstract description terms and relationships in the field of fault diagnosis of various types of equipment; Research on the semi-automatic domain knowledge ontology construction technology based on structured data, and form a knowledge model that contains knowledge ontology and the logical relationships between them.

Research on entity extraction and linking of equipment fault diagnosis information: Based on the basic equipment fault diagnosis corpus containing professional semantic tags, research on multi-round weighted entity linking technology based on semantic similarity and CNN convolutional neural network to realize the association between entities and knowledge ontology (entity extraction) and the linking of entities with entity descriptions in structured corpus.

Research on entity relationship extraction in the field of power plant equipment fault diagnosis: Based on text similarity matching technology, research on the screening method of entity relationship description statements to form a set of relationship description statements; research on entity relationship extraction technology in the field of equipment fault diagnosis based on syntactic dependency templates and Tree-LSTM recurrent neural networks to realize the mining and extraction of specific relationships between entities, and form an entity relationship library in the field of equipment fault diagnosis from the corpus base.

Furthermore, the construction of the knowledge graph for equipment fault diagnosis includes studying the knowledge fusion of equipment fault diagnosis information and studying large-scale knowledge graph hybrid storage management technology.

Research on knowledge fusion of equipment fault diagnosis information: Based on the differences in business description statements of different business departments, research on knowledge consistency check pattern acquisition methods and knowledge fusion strategies, detect logical conflicts and synonymous problems in different maintenance business description statements; research on knowledge fusion methods based on the combination of pattern matching and instance matching, realize the normalization of different description statements with the same semantics, and describe the correlation between different events of the same object.

Research on large-scale knowledge graph hybrid storage management: Research on multi-database hybrid storage technology for the underlying knowledge points of massive equipment fault diagnosis, and research on query optimization technology based on hybrid graph storage to achieve efficient storage and fast query of knowledge in the field of equipment fault diagnosis.

Furthermore, the knowledge graph reasoning for equipment fault diagnosis includes research on supporting multiple forms of natural language queries, research on knowledge reasoning based on equipment fault diagnosis knowledge graph, and research on knowledge path mining of equipment fault diagnosis knowledge graph.

The research supports multiple forms of natural language queries: research on natural language query error correction and input association methods for the field of equipment fault diagnosis; research on query parsing technology based on intent recognition and slot filling; research on methods for converting natural language queries into structured queries for inference engines; research on query result sorting schemes for different subdivided scenarios in the field of equipment fault diagnosis, thereby achieving fuzzy and fast search and professional push of equipment fault diagnosis knowledge.

Research on knowledge reasoning based on equipment fault diagnosis knowledge graph: Research on visualization configuration technology of implicit attribute reasoning and implicit relationship reasoning based on equipment fault diagnosis business rule definition, including rapid configuration of implicit attributes and relationships such as common formulas, equipment topology structure, and guidelines; Research on conceptual reasoning technology based on ontology concept inheritance relationship to realize the induction and summary of similar phenomena and the inference function of implicit conclusions.

Research on knowledge path mining of equipment fault diagnosis knowledge graph: Research on path discovery technology based on graph traversal to realize the visual discovery of implicit relationships between entities in the field of equipment fault diagnosis; Research on knowledge graph path mining technology based on time series analysis, association analysis and other methods.

Furthermore, the data collection and preprocessing is to collect operator training data of a certain type of pile, including equipment parameters, operator training materials, question banks, etc., and process them into large-scale question-answer pair data, including general knowledge and data in specific fields, to construct training sets and test sets.

The specific steps of data collection and preprocessing include:

(1) Data collection and organization: Collect various data sources related to the reactor type equipment, including equipment parameter manuals, operation manuals, technical specifications, training materials, historical questions and answers, etc. Organize and classify the data obtained from different sources to ensure the consistency of data structure and facilitate subsequent data preprocessing;

(2) Question-answer pair generation: Question-answer pairs are extracted from training materials and historical question and answer records. Questions can be practical problems related to equipment operation, troubleshooting, maintenance, etc., and answers are corresponding correct answers or solutions. In the process of generating question-answer pairs, the context information of the questions is retained to build a context-aware question-answering system;

(3) Cleaning and preprocessing: The data contains noise, errors, and redundant information. Text cleaning techniques are needed, such as removing HTML tags, stop words, and processing abbreviations, to ensure the accuracy and consistency of the data. Mark the type of question, such as whether the question is factual, principle explanation, or operation guidance, so that different strategies can be adopted in subsequent answer generation. Perform entity recognition to identify equipment-related entities from the question, such as equipment name, part number, etc., so as to answer the question more accurately;

(4) Data division and test set construction: The cleaned and preprocessed data is divided into training set, validation set and test set to ensure the reasonable division of the data set. The test set covers different types of problems to evaluate the performance and generalization ability of the model.

Fine-tuning and optimization of small sample data: In order to fine-tune the pre-trained model to adapt to the intelligent question-answering task, prepare a corresponding question-answer pair dataset. The dataset should contain questions and corresponding correct answers. Next, use this dataset to fine-tune the model and introduce appropriate loss functions and evaluation indicators.

Loss function: In intelligent question answering tasks, the commonly used loss function is the cross entropy loss function. Because the intelligent question answering task is a classification task, the model needs to select the correct answer from possible answers. The cross entropy loss function can measure the difference between the probability distribution of the answer generated by the model and the true answer, helping the model optimize parameters to better approximate the distribution of the correct answer.

Evaluation metrics: In order to optimize the performance of the model, some common evaluation metrics are used, such as accuracy and BLEU score (Bilingual Evaluation Understudy). Accuracy measures the proportion of correct answers generated by the model on the entire dataset. The BLEU score can be used to evaluate the similarity between the generated answer and the real answer.

Context-awareness: To support context-aware mechanisms for multi-round conversations, we can introduce memory networks or attention mechanisms to help the model track and understand the conversation history to effectively process long-term conversation information.

Memory networks are a mechanism that allows models to dynamically store and retrieve information. In multi-round conversations, models can use memory networks to store conversation histories as memories, and then retrieve these memories as needed in subsequent conversations. In this way, the model can obtain useful contextual information from previous conversations to help better understand the content of the current conversation. Memory networks include operations such as read, write, and forget to update and select information based on the importance of the conversation history.

The attention mechanism is also a powerful tool that can automatically focus on important information in a long sequence. In a multi-round conversation, the model can use the attention mechanism to evaluate the importance of different parts of the conversation history, and then select and integrate information based on these evaluation weights. Through the attention mechanism, the model can more accurately capture the contextual information of the conversation and use relevant contextual information in a targeted manner when generating responses.

Combine the memory network and the attention mechanism to build a context-aware multi-round dialogue model. In each round of dialogue, the model can maintain a memory unit, continuously input the dialogue history into the memory network, and update the representation of the dialogue history in real time. When generating a reply, the model can select appropriate context information through the attention mechanism and weight it according to importance, thereby effectively processing long-term dialogue information.

Semantic matching and answer generation: In order to understand the user's questions more accurately, a semantic matching module is introduced to identify the intent and type of the question. This module can semantically match and classify the questions entered by the user to determine the type of question, such as factual questions, principle explanation questions, or open questions.

For factual questions, the semantic matching module can directly retrieve the answer by matching with a predefined knowledge base. The knowledge base can be a structured database or graph containing facts and information in various fields. By matching with the knowledge base, the model can quickly find the answer to the question and return it to the user.

Automatic update mechanism for system data: In order to meet the regular update requirements of Guodian system data, an automatic update mechanism for system data has been established.

(1) The enterprise system text is transformed from unstructured data to structured data through a series of methods such as fragmentation processing and intelligent processing, and then stored in the database.

(2) The stored institutional structured data is used to establish an external vector knowledge base through the LLM big model to cooperate with the construction of intelligent application of Guodian institutional data.

(3) After the new system is revised, it will be compared with the old system. The system will provide similarity calculation results and display the comparison results.

(4) The comparison results need to be manually reviewed and confirmed. Once confirmed, the old system will be replaced with the new one, and the version information of the system will be saved within the system to facilitate data traceability.

(5) After the system data is updated, the vector will be rebuilt and integrated into the system intelligent application.

Automatic document scoring mechanism based on user behavior:

The usage of documents is affected by aspects such as content quality and topic popularity. Therefore, an automatic document scoring mechanism is established. When users retrieve documents, the scoring affects the document sorting, thereby improving the accuracy of retrieval and the usability of the system. The document scoring mechanism mainly uses dynamic feedback from user behavior.

Continuous update mechanism of plug-in vector knowledge base: The continuous update mechanism of large model plug-in vector knowledge base is a dynamic process designed to ensure that the information in the knowledge base is always up-to-date, accurate and comprehensive. It includes the following key steps:

Data collection and screening: First, data is collected through various channels, including but not limited to the Internet, databases, literature, social media, etc. Then, the collected data is automatically screened and verified to ensure that only high-quality and reliable information is included in the knowledge base.

Automated processing: Use natural language processing and machine learning technologies to automatically process the collected data, including entity recognition, relationship extraction, topic modeling and other technologies, in order to transform raw data into structured knowledge.

Manual review and correction: Although automated processing can efficiently process large amounts of data, manual review and correction are still required. The team reviews the results of automated processing, corrects errors, and supplements missing information to ensure the accuracy and completeness of the knowledge base.

Regular automatic updates: The knowledge base will be updated regularly to reflect the latest information and developments. The frequency of updates can be determined based on demand and the real-time nature of the data source, and can be updated daily, weekly or monthly.

User Feedback: The knowledge base accepts user feedback and makes corresponding adjustments and updates based on user suggestions and needs.

Continuous construction mechanism of enterprise intelligent knowledge management ecology: This project has initially built an intelligent knowledge management mechanism within the enterprise by combining personal network disk with big model. In the later planning, more intelligent knowledge management and service model can be achieved through deep processing of internal enterprise documents. It is mainly reflected in the following aspects:

Through natural processing, data mining and other technologies, deep mining of enterprise data can be achieved and data associations can be established.

A synchronization mechanism between the local knowledge base and the online knowledge base is established through the network disk, and the knowledge base data is automatically learned through a large model to form an enterprise intelligent brain.

Using the knowledge management platform as the enterprise data center, it provides knowledge base embedding services for other enterprise platforms. At the same time, it continuously accumulates other personalized business and user behavior data, improves knowledge data and provides more accurate knowledge services for various business platforms.

Through the continuous accumulation of large models, we establish an enterprise knowledge ecosystem and personal knowledge space, realize the interconnection between enterprise knowledge and personal knowledge, automatically establish a personal knowledge system for each user, and realize precise knowledge services for personalized users.

It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims (9)

1. An equipment failure information retrieval system based on big data storage technology is characterized in that: the method comprises the steps of equipment fault diagnosis based on a knowledge graph, expert knowledge base fusion and intelligent question-answering based on a large model, system upgrading and optimizing construction.

2. The equipment failure information retrieval system based on big data storage technology of claim 1, wherein: the equipment fault diagnosis based on the knowledge graph comprises the establishment of an entity relation based on multi-source heterogeneous equipment fault diagnosis information, the establishment of the knowledge graph facing the equipment fault diagnosis, the knowledge graph reasoning facing the equipment fault diagnosis, the equipment fault type research and judgment based on a classification decision tree algorithm, the alarm information research and judgment based on semantic rule matching and the research of auxiliary analysis based on knowledge-driven fault reasons.

3. The equipment failure information retrieval system based on big data storage technology of claim 1, wherein: the expert knowledge base fusion and intelligent question and answer based on the big model comprises data collection and preprocessing, question and answer big model construction, small sample data fine adjustment and optimization, context sensing, semantic matching and answer generation.

4. The equipment failure information retrieval system based on big data storage technology of claim 1, wherein: the system upgrading and optimizing construction comprises an automatic updating mechanism of system data, an automatic document scoring mechanism based on user behaviors, a continuous updating mechanism of an externally hung vector knowledge base and an ecological continuous construction mechanism of enterprise intelligent knowledge management.

5. A device failure information retrieval system based on big data storage technology as claimed in claim 2, wherein: the entity relation construction based on the multi-source heterogeneous equipment fault diagnosis information comprises construction of a research power plant equipment fault diagnosis ontology, entity extraction and link of research equipment fault diagnosis information and entity relation extraction in the field of research power plant equipment fault diagnosis.

6. A device failure information retrieval system based on big data storage technology as claimed in claim 2, wherein: the knowledge graph construction for equipment fault diagnosis comprises knowledge fusion of research equipment fault diagnosis information and research on a large-scale knowledge graph hybrid storage management technology.

7. A device failure information retrieval system based on big data storage technology as claimed in claim 2, wherein: the knowledge graph reasoning for equipment fault diagnosis comprises researching and supporting multi-form natural language query, researching knowledge reasoning based on equipment fault diagnosis knowledge graph and researching knowledge path mining of equipment fault diagnosis knowledge graph.

8. A device failure information retrieval system based on big data storage technology as claimed in claim 3, wherein: the specific steps of data collection and preprocessing include: (1) data collection and arrangement; (2) question-answer pairs are generated; (3) cleaning and pretreatment; (4) data partitioning and test set construction.

9. The equipment failure information retrieval system based on big data storage technology of claim 4, wherein: the continuous updating mechanism of the plug-in vector knowledge base comprises the following steps: (1) data collection and screening; (2) automated processing; (3) manual auditing and correction; (4) periodically and automatically updating; (5) user feedback.