A fully-mechanized coal mining equipment large model decision support method and system

Abstract

The application relates to the technical field of knowledge graphs and large models, and discloses a fully mechanized mining equipment large model decision support method and system. The method comprises the following steps: defining entities in data information of fully mechanized mining equipment and relationships between the entities to form triple data; converting the triple data into question and answer pairs based on a question and answer template, and optimizing the question and answer pairs; establishing a large language model, training the large language model based on the question and answer pairs, fine-tuning the large language model based on LoRA technology, quantitatively processing the equipment state by combining fuzzy comprehensive evaluation, and obtaining a quantitative result; the system uses the fine-tuned large language model to perform entity and intent recognition on natural language input by a user, constructs a Cypher query statement to query a graph database, obtains subgraph information, and inputs the subgraph information and the quantitative result as prompt words into the large language model to obtain retrieval information supporting large model decision making. The application improves the operation and maintenance efficiency of fully mechanized mining equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge graph and large model, and particularly relates to a fully mechanized mining equipment large model decision support method and system. BACKGROUND

[0002] With the wide application of fully mechanized mining equipment in the mining industry, the monitoring of its operating state, the prediction and diagnosis of faults, and the development of maintenance strategies become particularly important. Traditional fully mechanized mining equipment operation and maintenance relies on manual experience and periodic inspection, which has problems such as slow response speed, insufficient accuracy, and high maintenance cost. With the development of big data, natural language processing technology, and knowledge graph, it has become possible to use advanced technical means to improve the intelligent level of fully mechanized mining equipment operation and maintenance.

[0003] Existing knowledge graph construction methods, such as Chinese patent application No. CN117851610A, disclose a knowledge graph construction method and device based on a pre-trained large language model, wherein the method includes obtaining processing data; processing the unstructured data in the processing data through the pre-trained large language model to obtain a model output result; and constructing a knowledge graph according to the model output result. This application realizes the processing of multiple tasks in financial text data with the help of a pre-trained large language model, and realizes the construction of a knowledge graph. However, this application may perform well in financial text data, but may need additional adjustments or retraining to adapt to the fully mechanized mining equipment field.

[0004] Chinese patent application No. CN117633252A discloses an auxiliary retrieval method combining knowledge graph and large language model. This invention extracts information about the target field problem through a large language model; then performs semantic accurate alignment between the extracted information and the entities in the pre-constructed target field knowledge graph to obtain hit entities; and finally finds the corresponding answers in the pre-constructed target field knowledge graph according to the hit entities. However, this application relies on the accurate alignment between the information extracted by the large language model and the entities in the pre-constructed knowledge graph. If the semantic alignment is not accurate, it may introduce errors or irrelevant information.

[0005] Therefore, there is a need for a fully mechanized mining equipment large model decision support method to realize real-time monitoring and fault prediction of the operating state of fully mechanized mining equipment, while improving the operation and maintenance efficiency of fully mechanized mining equipment. SUMMARY

[0006] The present application provides a fully mechanized mining equipment large model decision support method and system for improving the operating efficiency of fully mechanized mining equipment.

[0007] In a first aspect, the present application provides a fully mechanized mining equipment large model decision support method, which comprises:

[0008] Step S1: obtaining data information of the fully mechanized mining equipment, classifying the data information to obtain multiple information categories, defining entities in each information category, and relationship information between the entities;

[0009] Step S2: identifying a data format of each of the data information, extracting entities, relationship information, and attribute values of the entities from the data information in different ways based on different data formats, forming a triple data, and storing the triple data in a Neo4j database;

[0010] Step S3: establishing a question and answer template, converting the triple data into a question and answer pair in Alpaca format based on the question and answer template, establishing a natural language processing model, optimizing the question and answer pair based on the natural language processing model, and obtaining an optimized question and answer pair;

[0011] Step S4: establishing a large language model of the fully mechanized mining equipment, using a Transformer model as a basic architecture, training the large language model using the optimized question and answer pair as a training set, fine-tuning the large language model based on LoRA technology during the training, defining multiple evaluation indexes for evaluating the fully mechanized mining equipment, quantifying the evaluation indexes, and obtaining a quantification result;

[0012] Step S5: loading the fine-tuned large language model, a user inputting natural language to a fully mechanized mining equipment decision support system to query information of the fully mechanized mining equipment, the fully mechanized mining equipment decision support system performing entity recognition and intent recognition on the natural language using the fine-tuned large language model, and constructing a Cypher query statement based on the recognized entities and intents;

[0013] Step S6: querying in the Neo4j database using the Cypher query statement, determining whether a query result is obtained, if so, obtaining a local feature subgraph based on the query result, if not, performing multi-level relationship reasoning query to obtain an information redundancy subgraph, constructing a prompt word based on the local feature subgraph, the information redundancy subgraph, and the quantification result, and delivering the prompt word to the large language model to obtain retrieval information supporting decision of the large model.

[0014] In combination with the first aspect, in a first implementation manner of the first aspect of the present application, forming the triple data comprises:

[0015] The data format includes structured data, semi-structured data, and unstructured data. If the data information is in the format of structured data, it is directly converted into triple data. If the data information is in the format of semi-structured data, the semi-structured data is parsed to obtain knowledge information, and entity, relation information, and attribute values ​​of the entity are extracted from the knowledge information to obtain triple data.

[0016] If the data information is in the format of unstructured data, a BiLSTM-CRF model is established. The BiLSTM-CRF model includes a bidirectional long short-term memory network model and a conditional random field discriminant model. The context information in the unstructured data is obtained based on the bidirectional long short-term memory network model. Entity information is identified from the unstructured data based on the context information. The boundaries of the entity information are identified based on the conditional random field discriminant model to obtain complete entity information and the relationship information between entities. The attribute information of the entities is extracted. Triple data is formed based on the entity information, relationship information, and attribute information.

[0017] In conjunction with the first aspect, in the second implementation of the first aspect of this application, converting the triplet data into question-answer pairs in Alpaca format based on the question-answer template includes:

[0018] The natural language processing model is the PEGASUS model. The question in the question-answer pair is defined as the first information, and the answer corresponding to the question is defined as the second information. All the second information related to each first information is collected, and the second information is input into the PEGASUS model. The PEGASUS model fuses all the second information to obtain the third information. The third information is a coherent answer containing all the second information. The optimized question-answer pair is constructed based on the first information and the third information.

[0019] In conjunction with the first aspect, in the third implementation of the first aspect of this application, the large language model is fine-tuned based on LoRA technology, including:

[0020] During the pre-training phase of the large language model, LoRA technology is used to keep the original weight matrix of each Transformer layer in the Transformer model unchanged. During the training phase, low-rank matrices A and B are introduced, and the parameters of these two low-rank matrices are adjusted. The fine-tuned first weight matrix W′ is obtained based on the first formula: W′=W+A·B, where W is the original weight matrix, and W∈R d×k This indicates that the original weight matrix is ​​a d-row, k-column real matrix, A∈R. d×r , B∈R r×k, r represents the rank of matrix A or matrix B, where r << min(d,k), and the first weight matrix is ​​used to train on the optimized question-answer pairs to achieve fine-tuning of the large language model.

[0021] In conjunction with the first aspect, in the fourth implementation of the first aspect of this application, the evaluation index is quantified, including:

[0022] A factor set is constructed based on the evaluation indicators, and a comment set is obtained by classifying the evaluation indicators into levels. A membership function is set, and the membership degree μ(x) of each evaluation indicator to the comment is set based on the membership function. The membership function is: x represents the real-time parameters of each evaluation indicator of the fully mechanized mining equipment, and a, b, c, and d are four preset key parameters. The membership degrees are summarized to obtain a single-factor evaluation matrix. The weight coefficients of each evaluation indicator are set, and fuzzy synthesis is performed based on the weight coefficients and the evaluation matrix to obtain a comprehensive evaluation result vector. The comprehensive evaluation result vector is then normalized, and the normalized comprehensive evaluation result vector is used as the quantification result.

[0023] In conjunction with the first aspect, in the fifth implementation of the first aspect of this application, a Cypher query statement is constructed based on the identified entity and intent, including:

[0024] The input natural language is segmented into words, each segmented word is tagged with part-of-speech tags and input into a fine-tuned large language model. The model outputs words or phrases that are identified as entities or intents. The identified entities are mapped to the corresponding nodes in the Neo4j database. The identified intents are matched with the preset intent list in the system to obtain the intent type. If there are multiple intent types, the multiple intents are merged to obtain a comprehensive intent.

[0025] The decision support system for fully mechanized mining equipment determines the query relationship between entities based on the identified intent. The decision support system for fully mechanized mining equipment constructs a Cypher query statement, retrieves the query relationship in the Neo4j database based on the Cypher query statement, and returns the query result to the decision support system for fully mechanized mining equipment.

[0026] In conjunction with the first aspect, in the sixth implementation of the first aspect of this application, multi-level relational reasoning queries are performed to obtain an information redundancy subgraph, including:

[0027] Based on the identified entities and the relationships between them, a query path is constructed. The relationships between entities include first-degree, second-degree, and third-degree relationships. The decision support system for the fully mechanized mining equipment queries local feature subgraphs based on the query path of first-degree relationships. If no local feature subgraph is obtained, the system then extends the query path to second-degree and third-degree relationships to obtain the corresponding subgraph information. The system scores the relevance of each subgraph and defines the subgraph with the highest score as an information redundancy subgraph for output.

[0028] In conjunction with the first aspect, in the seventh implementation of the first aspect of this application, constructing prompt words based on the local feature sub-graph, the information redundancy sub-graph, and the quantization result includes:

[0029] The decision support system for the fully mechanized mining equipment extracts key information from the local feature subgraph or the information redundancy subgraph. The key information includes entity name, type, attributes, and relationship information between entities. The key information is transformed into structured information, and the user's input natural language, the structured information, and the quantification results are combined to generate prompt words.

[0030] In conjunction with the first aspect, in the eighth implementation of the first aspect of this application, the information categories include equipment information, risk information, fault information, cause information, solution information, and specification information.

[0031] Secondly, this application provides a large-scale decision support system for fully mechanized mining equipment, the system comprising:

[0032] The definition module is used to acquire data information from the fully mechanized mining equipment, classify the data information to obtain multiple information categories, define the entities in each information category, and the relationship information between the entities;

[0033] The triplet data processing module is used to identify the data format of each type of data information, extract entity, relation information and entity attribute values ​​from the data information in different ways based on different data formats, form triplet data, and store the triplet data in the Neo4j database;

[0034] The question-answer pair formation module is used to establish a question-answer template, convert the triplet data into question-answer pairs in Alpaca format based on the question-answer template, establish a natural language processing model, and optimize the question-answer pairs based on the natural language processing model to obtain optimized question-answer pairs.

[0035] The fuzzy comprehensive evaluation module is used to establish a large language model of the fully mechanized mining equipment. The large language model uses the Transformer model as its basic architecture, and the optimized question-answer pairs are used as the training set to train the large language model. During the training process, the large language model is fine-tuned based on LoRA technology, and various evaluation indicators for evaluating the fully mechanized mining equipment are defined, and the evaluation indicators are quantified to obtain the quantification results.

[0036] The query statement construction module is used to load the fine-tuned large language model. When the user inputs natural language into the fully mechanized mining equipment decision support system to query information about the fully mechanized mining equipment, the fully mechanized mining equipment decision support system uses the fine-tuned large language model to perform entity recognition and intent recognition on the natural language, and constructs a Cypher query statement based on the recognized entities and intents.

[0037] The query module is used to perform queries in the Neo4j database using Cypher query statements, determine whether query results are obtained, and if so, obtain a local feature subgraph based on the query results. If no query results are obtained, perform multi-level relational reasoning query to obtain an information redundancy subgraph, construct prompt words based on the local feature subgraph or information redundancy subgraph, and pass the prompt words to the large language model to obtain retrieval information that supports the large model's decision-making.

[0038] The technical solution provided in this application acquires data information from fully mechanized mining equipment, defines entities and relationships within the data information, and extracts entities, relationships, and attribute values ​​from the data information using different methods depending on the data format, forming triplet data. This triplet data is stored in a Neo4j database. A triplet data processing module generates a dataset for fine-tuning a large model, which, combined with a large model fine-tuning module, adapts the model to the decision support needs of the fully mechanized mining equipment domain. Furthermore, a Cypher query is used to perform entity and intent queries on the input, obtaining a knowledge graph subgraph associated with the input, thereby providing enhanced retrieval decision support. This invention achieves real-time monitoring and fault prediction of the operating status of fully mechanized mining equipment, while simultaneously improving the operation and maintenance efficiency of the equipment. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of one embodiment of a large-scale decision support method for fully mechanized mining equipment in this application.

[0041] Figure 2 A diagram illustrating the process of constructing the knowledge graph for fully mechanized mining equipment in this application;

[0042] Figure 3 This is a diagram illustrating the LoRA fine-tuning process in this application;

[0043] Figure 4 This is an example diagram of the fault tree structure of the fully mechanized mining equipment in this application;

[0044] Figure 5 This is a schematic flowchart of the large-scale model module of the fully mechanized mining equipment in this application;

[0045] Figure 6 A graph showing the loss rate during fine-tuning training in this application;

[0046] Figure 7 This is a flowchart illustrating the principle of the comprehensive fuzzy evaluation expert module in this application;

[0047] Figure 8 This is a schematic diagram of one embodiment of a large-scale decision support system for fully mechanized mining equipment in this application. Detailed Implementation

[0048] This application provides a decision support method and system for a large-scale model of fully mechanized mining equipment. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0049] Example 1:

[0050] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the large-scale decision support method for fully mechanized mining equipment in this application includes:

[0051] Step S1: Obtain data information from the fully mechanized mining equipment, classify the data information to obtain multiple information categories, define the entities in each information category, and the relationship information between the entities.

[0052] Specifically, in the construction of knowledge graphs, two common approaches are top-down and bottom-up. The top-down approach starts by building concepts at the top level, then gradually refines branches downwards to ensure the rigor of the hierarchical structure, and then adds the collected entities one by one to the pre-built concepts. The bottom-up approach, on the other hand, starts by forming concepts at the bottom level, then gradually abstracts upwards to form higher-level concepts. This invention selects the top-down approach to construct a knowledge graph for fully mechanized mining equipment. The construction process is as follows: Figure 2 As shown.

[0053] First, the knowledge modeling phase is explained: Information related to the fully mechanized mining equipment is acquired, including operation logs, sensor data, maintenance records, and user feedback data. This collected data is categorized into six main types: "Equipment," "Risk," "Fault," "Cause," "Solution," and "Standards / Specifications." Concepts are defined within each category to describe entities within the domain in more detail. Next, attributes are defined for each concept, constructing a hierarchical knowledge system. Finally, the types of relationships between entities are defined to describe their connections and dependencies; for example, there is a "cause" relationship between risk and fault, and a "correspondence" relationship between fault and solution.

[0054] Step S2: Identify the data format of each type of data information, and extract entity, relation information and entity attribute values ​​from the data information in different ways based on different data formats to form triple data, and store the triple data in the Neo4j database.

[0055] Specifically, also refer to Figure 2 The following is an explanation of knowledge extraction: Knowledge extraction is the process of extracting effective information from data and transforming it into a structured form that can be used in knowledge graphs. Before knowledge extraction, the source of the data information is identified. In the field of fully mechanized mining equipment, the data sources mainly include structured data, semi-structured data, and unstructured data. For each source of data information, different methods are used to extract entities from the data information, such as equipment name, fault type, personnel name, etc. The relationships between entities and the attributes of entities are determined, such as equipment model, fault severity, personnel position, etc. The identified entities, relationships, and attributes are then transformed into triples.

[0056] The resulting triplet data is stored in the Neo4j database. Neo4j is a high-performance, scalable storage engine for graph data with graph database characteristics. It is based on the concept of a graph, where data is stored in the form of nodes and edges. Each node represents an entity, and each edge represents a relationship between entities. The four core elements of labels, nodes, relationships, and attributes together constitute the basic data model of the Neo4j graph database.

[0057] Step S3: Establish a question-and-answer template, convert the triplet data into question-and-answer pairs in Alpaca format based on the question-and-answer template, establish a natural language processing model, optimize the question-and-answer pairs based on the natural language processing model, and obtain optimized question-and-answer pairs.

[0058] Specifically, before fine-tuning the large language model, the knowledge in the knowledge graph must be converted into a format suitable for fine-tuning techniques, namely the Alpaca format. The Alpaca format contains question instructions and corresponding response pairs. Through predefined question-and-answer templates, triples are converted into question-and-answer pairs that conform to the Alpaca format. For example, the triple form (S, P, O) stored in the knowledge graph is ("hydraulic support", "used for", "fully mechanized mining face"). Based on this triple, a question-and-answer pair conforming to the Alpaca format is generated: Question: In what situations are hydraulic supports used? Answer: Hydraulic supports are used in fully mechanized mining faces.

[0059] Because knowledge graphs often contain nodes that connect to many other nodes, the question-answer pairs generated from extracted triples may contain many identical questions with different answers. This degrades dataset quality and significantly negatively impacts model performance. Therefore, data augmentation is necessary during data preprocessing to mitigate this negative effect. This invention uses the pre-trained PEGASUS model to capture deep semantic information from text, obtaining high-dimensional text vector representations. Then, using natural language generation techniques, it merges multiple answers into a coherent and natural response text—the optimized answer pair.

[0060] Step S4: Establish a large language model for the fully mechanized mining equipment. The large language model uses the Transformer model as its basic architecture. Optimized question-answer pairs are used as the training set to train the large language model. During the training process, the large language model is fine-tuned based on LoRA technology. Various evaluation indicators for fully mechanized mining equipment are defined, and the evaluation indicators are quantified to obtain quantitative results.

[0061] Specifically, since large language models are trained on massive amounts of text, directly applying the original trained model to decision support tasks in fully mechanized mining equipment often does not meet expectations. Therefore, this invention introduces efficient fine-tuning techniques to enhance the performance of large models in fully mechanized mining equipment tasks. Mainstream large language model fine-tuning techniques can be divided into three categories: Adapter, Prefix-Tuning, and LoRA. Among these three techniques, LoRA has unique advantages over Adapter and Prefix-Tuning, making it the preferred choice in many scenarios. First, LoRA freezes the weights of the pre-trained model and introduces a trainable low-rank decomposition matrix in each Transformer layer, greatly reducing the number of parameters and GPU memory requirements for fine-tuning. The LoRA fine-tuning process is as follows: Figure 3 As shown.

[0062] The operational status and fault analysis of fully mechanized mining equipment involves multiple complex factors, which are often fuzzy and uncertain. Traditional decision support methods may not be able to fully consider the impact of these fuzzy factors. Fuzzy comprehensive evaluation, through the theory of fuzzy mathematics, can effectively handle complex problems involving multiple factors and levels, making it suitable for comprehensive status assessment and fault analysis of fully mechanized mining equipment. Figure 4 The diagram shown is an example of the fault tree structure of a fully mechanized mining equipment. This invention sets up a comprehensive fuzzy evaluation expert module, which enables a comprehensive evaluation of the operating status of the fully mechanized mining equipment and provides a scientific basis for subsequent decision support.

[0063] Step S5: Load the fine-tuned large language model. The user inputs natural language into the fully mechanized mining equipment decision support system to query fully mechanized mining equipment information. The fully mechanized mining equipment decision support system uses the fine-tuned large language model to perform entity recognition and intent recognition on the natural language, and constructs Cypher query statements based on the recognized entities and intents.

[0064] Specifically, such as Figure 5 The diagram shows the principle flowchart of the large model module for fully mechanized mining equipment. First, a finely tuned large language model is loaded. This model is further trained on data in a specific domain based on pre-training. Users can enter natural language to query through the interface of the fully mechanized mining equipment decision support system. The query content can be such as equipment status, fault diagnosis, and maintenance suggestions. The system uses the finely tuned large language model to perform entity recognition and intent recognition on the natural language input by the user. The intent is the goal or type of information that the user hopes to achieve through the query, such as querying equipment status, solutions, or predicting faults.

[0065] Secondly, it determines whether the entity and intent query was successful. If the query is successful, a Cypher query statement is constructed based on the identified entity and intent. Cypher is the query language of the graph database Neo4j, used to retrieve and manipulate data in the graph database. The constructed Cypher query statement is sent to the graph database for execution to retrieve information related to the user's query.

[0066] Step S6: Use the Cypher query statement to query the Neo4j database and determine whether a query result is obtained. If so, obtain a local feature subgraph based on the query result. If no query result is obtained, perform a multi-level relational reasoning query to obtain an information redundancy subgraph. Construct prompt words based on the local feature subgraph, the information redundancy subgraph, and the quantification results, and pass the prompt words to the large language model to obtain retrieval information that supports the large model's decision-making.

[0067] Specifically, after the graph database returns query results, the system checks whether the Cypher query also returned results. If results were returned, it means that data related to the user's query was found, and the system constructs a local feature subgraph based on these results. This subgraph contains entities and relationships directly related to the user's query and is part of the knowledge graph. Then, useful information, such as device status, fault causes, and solutions, is extracted from the local feature subgraph. If no direct query results are obtained, a multi-level relational reasoning query is used. This method returns a subgraph containing more entities and relationships. This subgraph contains some data that is not directly related or redundant, but it provides broader information.

[0068] The system constructs prompt words based on local feature subgraphs or information redundancy subgraphs. These prompt words are guide words. The prompt primarily generates an input format that can be passed to the large-scale model of the fully mechanized mining equipment by combining user input and subgraph information returned by the query. This invention adopts a hierarchical design, combining the structured information of the knowledge graph with the results of fuzzy comprehensive evaluation. By constructing a hierarchical prompt process (task instruction - subgraph parsing - expert rule injection - prompt word construction), the system can generate text that the large-scale model can process based on the structured information in the graph database and the input, and then pass it to the large-scale model. This allows the large-scale model to fully consider the real and powerful information from the graph database when generating decision support information, reducing the "illusion" phenomenon of the large-scale model and effectively improving the authenticity and reliability of decision information. The system consists of the following steps:

[0069] 1. Task Instructions: Define the task objectives of the model, enabling the model to understand the context and objectives of the input information, and limit the output range.

[0070] 2. Subgraph Parsing: Subgraph construction involves not only direct relationships between entities but also different types of edges (relationships) and their attributes. This step parses the subgraph information returned by the graph database into a structured text format, facilitating subsequent prompt word construction.

[0071] 3. Expert rule injection: The equipment status level (such as "abnormal") and key parameter membership degree (such as "the membership degree of the hydraulic support working pressure is 0.8") output by the fuzzy comprehensive evaluation expert module are used as context supplements.

[0072] 4. Constructing prompts: The constructed prompts are presented in natural language and contain the text content of the subgraph parsing. The constructed prompts are concatenated with the input, giving full consideration to the real and powerful information from the graph database, and guiding the model to generate accurate output.

[0073] By inputting the constructed prompt into the large model of the fully mechanized mining equipment, the large model generates intelligent decisions on the operating status, fault analysis, and maintenance strategies of the fully mechanized mining equipment. The enhanced retrieval method enables the system to not only rely on the model's reasoning, but also provide more comprehensive, detailed, and realistic decision suggestions through the support of historical data and professional knowledge.

[0074] In one specific embodiment, forming triplet data specifically includes the following steps:

[0075] Data formats include structured data, semi-structured data, and unstructured data. If the data information is in the format of structured data, it is directly converted into triple data. If the data information is in the format of semi-structured data, the semi-structured data is parsed to obtain knowledge information, and entity, relation information, and attribute values ​​of the entities are extracted from the knowledge information to obtain triple data.

[0076] If the data is in the format of unstructured data, a BiLSTM-CRF model is established. The BiLSTM-CRF model includes a bidirectional long short-term memory network model and a conditional random field discriminant model. The bidirectional long short-term memory network model is used to obtain contextual information from the unstructured data. Based on the contextual information, entity information is identified from the unstructured data. Based on the conditional random field discriminant model, the boundaries of entity information are identified to obtain complete entity information and the relationship information between entities. The attribute information of the entities is extracted, and triple data is formed based on entity information, relationship information, and attribute information.

[0077] Specifically, structured data is stored in a well-defined format, usually in tabular form, such as technical specifications or equipment maintenance records. This type of data can be directly converted into a knowledge graph, or triples, without extraction. Semi-structured data has a certain structure, such as web page data, and information can be extracted through HTML parsing and selector techniques. Based on the extracted information, entities, relationships, and attribute information can be identified.

[0078] The BiLSTM-CRF model is a deep learning model applied to natural language processing tasks. This model integrates two key components: BiLSTM, or Bidirectional Long Short-Term Memory, unlike traditional unidirectional recurrent neural networks, contains recurrent neural networks in two directions: one from left to right (forward) and the other from right to left (backward). This allows the model to consider both preceding and following contextual information, helping to better capture dependencies and patterns in sequence data. CRF, or Conditional Random Field, is a statistical modeling method for sequence labeling tasks. In the BiLSTM-CRF model, CRF is used to capture dependencies between entities, better capturing entity boundaries and structures.

[0079] In one specific embodiment, converting triplet data into question-answer pairs in Alpaca format based on a question-answer template includes the following steps:

[0080] The natural language processing model is the PEGASUS model. In a question-answer pair, the question is defined as the first information and the answer to the question is defined as the second information. All the second information related to each first information is collected and input into the PEGASUS model. The PEGASUS model integrates all the second information to obtain the third information, which is a coherent answer containing all the second information. Based on the first information and the third information, an optimized question-answer pair is constructed.

[0081] Specifically, since an entity may be associated with multiple other entities, answers extracted directly from a knowledge graph can have many variations. To provide more comprehensive and accurate information, these answers need to be integrated, and the PEGASUS model is a pre-trained model for text summarization. In the context of data augmentation, the PEGASUS model may be used to generate or optimize answers, making them more coherent and comprehensive.

[0082] For example, when a user asks, "What are the components of a hydraulic support?", there might be multiple correct answers, such as "A hydraulic support includes a column" or "A hydraulic support includes an oil pump." While these answers are structurally similar, they provide different information. It's helpful to integrate these disparate answers into a more comprehensive response. For instance, combining "A hydraulic support includes a column" and "A hydraulic support includes an oil pump" into "A hydraulic support includes a column, an oil pump, a boot, a panel, etc."

[0083] In one specific embodiment, fine-tuning a large language model based on LoRA technology includes the following steps:

[0084] In the pre-training phase of the large language model, LoRA technology is used to keep the original weight matrix of each Transformer layer unchanged. During the training phase, low-rank matrices A and B are introduced, and their parameters are adjusted. The fine-tuned first weight matrix W′ is obtained based on the first formula: W′=W+A·B, where W is the original weight matrix, and W∈R d×k This indicates that the original weight matrix is ​​a d-row, k-column real matrix, A∈R. d×r , B∈R r×k , r represents the rank of matrix A or matrix B, where r << min(d,k). The first weight matrix is ​​used to train on the optimized question-answer pairs to achieve fine-tuning of the large language model.

[0085] Specifically, in the pre-training phase of the large language model, the model has already been trained on a large amount of text data, learning rich language features and knowledge. The LoRA technique allows the original weight matrix of each layer in the Transformer model to remain unchanged during fine-tuning. Instead, it fine-tunes the large model by introducing two low-rank matrices. The rank of the low-rank matrices is much smaller than the dimension of the original weight matrix, which means that they contain fewer parameters. The first weight matrix W is used to train on the optimized question-answer pairs. This means that the model will be further trained on a specific task dataset to improve its performance on that task. In this way, the large language model can adapt to new task requirements while maintaining pre-trained knowledge, achieve effective fine-tuning, reduce the number of parameters that need to be updated during fine-tuning, and improve the efficiency of fine-tuning.

[0086] The loss rate of fine-tuning training is as follows Figure 6 As shown in the figure, the training loss tends to stabilize after a certain number of iterations, without significant oscillations or increases. This indicates that the model did not overfit during training, and the training process was relatively stable.

[0087] In one specific embodiment, quantifying the evaluation indicators includes the following steps:

[0088] A factor set is constructed based on evaluation indicators. The evaluation indicators are then categorized into levels to obtain a set of comments. Membership functions are set, and the membership degree μ(x) of each evaluation indicator to the comments is set based on these functions. The membership functions are as follows: x represents the real-time parameters of each evaluation indicator of the fully mechanized mining equipment, and a, b, c, and d are four preset key parameters. The membership degrees are summarized to obtain a single-factor evaluation matrix. The weight coefficients of each evaluation indicator are set, and fuzzy synthesis is performed based on the weight coefficients and the evaluation matrix to obtain a comprehensive evaluation result vector. The comprehensive evaluation result vector is then normalized, and the normalized comprehensive evaluation result vector is used as the quantification result.

[0089] Specifically, such as Figure 7 The diagram shown is a flowchart illustrating the principle of the comprehensive fuzzy evaluation expert module. The first step employs a fuzzy comprehensive evaluation method, based on fuzzy set theory, to transform the qualitative evaluation of fully mechanized mining equipment components into a quantitative evaluation. Membership functions describe the degree of membership of each influencing factor to the equipment's evaluation level. The fuzzy comprehensive evaluation requires constructing a factor set U, a comment set V, and a weight vector A, specifically represented as follows: U = {u1, u2, ..., u...} n}, V={v1,v2,…,v m}, A=(a1,a2,…,a n ), The factor set U contains all evaluation indicators, namely: U = {hydraulic support working pressure, initial support force of support system support, coal mining machine traction speed, coal mining machine cutting power, scraper conveyor chain speed}. The comment set V is divided into 4 levels: V = {fault, abnormal, normal, good}. The membership degree is obtained by assigning a membership function. The membership function of this invention adopts a trapezoidal distribution. The trapezoidal distribution function is defined according to the threshold interval of the equipment parameters. The interval classification in the fault tree structure is used as the boundary value of the trapezoidal distribution. The membership degree function is used to define the degree of membership of a single parameter value to a certain comment.

[0090] According to expert evaluation, the weights of the factors in the above examples in fuzzy evaluation decision-making are: A = (0.25, 0.25, 0.15, 0.15, 0.20);

[0091] Based on real-time data, the membership degree of each parameter to each comment is calculated, forming a single-factor evaluation matrix R. This matrix represents the membership degree of each factor to each comment, summarizing the membership degrees of all factors to each comment. It is a prerequisite for comprehensive evaluation, where r... ij ∈[0,1]. The specific evaluation matrix is ​​as follows: Select fuzzy composition operator Calculate the final comprehensive evaluation result vector B, and then normalize the comprehensive evaluation result vector B to obtain Bi. norm The formula is as follows: Based on the aforementioned expert rule settings and fuzzy comprehensive evaluation, B is obtained from the real-time collected data. norm This section's expert rules module is based on B. norm The results determine the state of specific equipment components and are input into the large language model to support subsequent decision support functions.

[0092] In one specific embodiment, constructing a Cypher query statement based on the identified entities and intents includes the following steps:

[0093] The input natural language is segmented into words, each segmented word is tagged with part-of-speech tags and input into a fine-tuned large language model. The model outputs words or phrases that are identified as entities or intents. The identified entities are mapped to the corresponding nodes in the Neo4j database. The identified intents are matched with the preset intent list in the system to obtain the intent type. If there are multiple intent types, the multiple intents are merged to obtain a comprehensive intent.

[0094] The decision support system for fully mechanized mining equipment determines the query relationships between entities based on the identified intent. The system then constructs a Cypher query statement, retrieves the query relationships from the Neo4j database based on the Cypher query statement, and returns the query results to the decision support system for fully mechanized mining equipment.

[0095] Specifically, the input natural language is segmented into words, and each word is labeled with its part-of-speech and category. Generally, the jieba library in Python or some Chinese language models are used. The language model identifies entities such as device names and fault types in the input. All entities are extracted from the input and mapped to the corresponding nodes in the graph database Neo4j. Entity queries can find the corresponding nodes in the knowledge graph. However, it is meaningless for the graph database to only return node information. It needs to return a subgraph full of nodes and relationships to provide sufficient information for decision support for the large model. Therefore, intent queries are also needed to explore the relationships that need to be used.

[0096] This invention utilizes a predefined list of relationships to efficiently perform intent queries. The steps for intent querying are as follows: 1. Tokenization and tagging: The intent information is processed using the tokenization and tagging steps from entity queries; 2. Predefined intent classification: The extracted keywords are matched with the predefined intent list in the system to determine the query type. For example, "what to do" and "how to handle" can be classified as the intent "find a solution"; 3. Multi-intent processing: For inputs containing multiple intents, the system will merge the multiple intents and generate corresponding multi-level queries.

[0097] Based on the input entities and relationships, the system constructs corresponding Cypher queries. For example, when the input is "What to do if hydraulic oil leaks", the system identifies the two entities "hydraulic oil" and "leak" and recognizes the intent "find a solution", converting it into a Cypher query. The Cypher query statement includes: MATCHpath = (e1) - [:The fault symptom is] -> (e2) - [:The solution is] -> (e3);

[0098] WHERE e1.name = 'hydraulic oil' AND e2.name = 'leak'; RETURNpath;

[0099] Using Cypher queries, the system can retrieve related subgraphs in the knowledge graph. These subgraphs can be called local feature subgraphs, which focus on key information in the input.

[0100] In one specific embodiment, performing multi-level relational reasoning queries to obtain an information redundancy subgraph specifically includes the following steps:

[0101] Based on the identified entities and the relationships between them, a query path is constructed. The relationships between entities include first-degree, second-degree, and third-degree relationships. The decision support system for fully mechanized mining equipment queries local feature subgraphs based on the query path of first-degree relationships. If no local feature subgraph is obtained, the system then extends the query path to second-degree and third-degree relationships to obtain the corresponding subgraph information. The system scores the relevance of each subgraph and defines the subgraph with the highest score as an information redundancy subgraph for output.

[0102] Specifically, multi-level relational reasoning queries involve the following steps: 1. Path construction: Path construction typically starts from entities known in the input and expands along layers of relational paths; 2. Relationship level progression: The system incrementally increases the relational level of the reasoning, expanding from first-degree relations (relations directly related to the target entity) to second-, third-, and even higher levels to find possible related information; 3. Relevance scoring: The system scores the relevance of each acquired subgraph, based on the degree of relevance between the subgraph content and the user's query. This may involve natural language processing techniques to evaluate the relevance of information. From all acquired subgraphs, the subgraph with the highest score is selected as the information redundancy subgraph. Although this subgraph may contain some redundant information, it provides the most comprehensive data to support decision-making.

[0103] Multi-level relational reasoning queries can retrieve information redundancy subgraphs because they may contain some irrelevant or redundant data. While the relevance of the query results may not be as high as the local feature subgraphs obtained from entity queries and intent queries, they provide robust and reliable data support for decision-making in the large-scale model of fully mechanized mining equipment when entity and intent queries fail, thus improving the system's fault tolerance.

[0104] In a specific embodiment, constructing prompt words based on local feature subgraphs, information redundancy subgraphs, and quantization results includes the following steps:

[0105] The decision support system for fully mechanized mining equipment extracts key information from local feature subgraphs or information redundancy subgraphs. The key information includes entity names, types, attributes, and relationships between entities. The key information is transformed into structured information, and the system combines the user's input natural language, structured information, and quantitative results to generate prompt words.

[0106] Specifically, the system extracts key information from local feature subgraphs or information redundancy subgraphs. This key information includes entities and relationships related to the user query, identifying entity names, types, attributes, and relationships between entities in the subgraph. For example, if the subgraph contains the entity "hydraulic support," the system extracts its attributes (such as model and manufacturer) and relationships with other entities (such as "connected to" conveyor). The extracted key information is converted into a structured format, such as JSON or XML. The system combines the user's natural language input with the structured information and quantification results, ensuring that the prompts not only contain information from the graph database but also reflect the user's original query and the quantification results of the evaluation metrics. The combined information is used to construct prompts, which are guided statements presented in natural language and include the text content parsed from the subgraph. By concatenating the constructed prompts with the input, the system fully considers the real and compelling information from the graph database, guiding the model to generate accurate output.

[0107] Example 2:

[0108] The above describes a decision support method for a large-scale fully mechanized mining equipment model in the embodiments of this application. The following describes a decision support system for a large-scale fully mechanized mining equipment model in the embodiments of this application. Please refer to [link / reference]. Figure 8 One embodiment of a large-scale decision support system for fully mechanized mining equipment in this application includes:

[0109] The definition module is used to acquire data information from the fully mechanized mining equipment, classify the data information to obtain multiple information categories, define the entities in each information category, and the relationship information between the entities;

[0110] The triplet data processing module is used to identify the data format of each type of data information, and extract entity, relation information and entity attribute values ​​from the data information in different ways based on different data formats to form triplet data, and store the triplet data in the Neo4j database;

[0111] The question-answer pair formation module is used to create question-answer templates, convert triple data into question-answer pairs in Alpaca format based on the question-answer templates, build a natural language processing model, and optimize the question-answer pairs based on the natural language processing model to obtain optimized question-answer pairs.

[0112] The fuzzy comprehensive evaluation module is used to establish a large language model of fully mechanized mining equipment. The large language model uses the Transformer model as its basic architecture, and the optimized question-answer pairs are used as the training set to train the large language model. During the training process, the large language model is fine-tuned based on LoRA technology. Multiple evaluation indicators for fully mechanized mining equipment are defined, and the evaluation indicators are quantified to obtain quantitative results.

[0113] The query statement construction module is used to load the fine-tuned large language model. Users input natural language into the fully mechanized mining equipment decision support system to query information about fully mechanized mining equipment. The fully mechanized mining equipment decision support system uses the fine-tuned large language model to perform entity recognition and intent recognition on the natural language, and constructs Cypher query statements based on the recognized entities and intents.

[0114] The query module is used to perform queries in the Neo4j database using Cypher query statements, determine whether query results are obtained, and if so, obtain a local feature subgraph based on the query results. If no query results are obtained, perform multi-level relational reasoning queries to obtain an information redundancy subgraph. Based on the local feature subgraph or information redundancy subgraph, construct prompt words and pass the prompt words to the large language model to obtain retrieval information that supports the large model's decision-making.

[0115] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0116] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or 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 this application. 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.

[0117] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A decision support method for a large-scale model of fully mechanized mining equipment, characterized in that, The method includes: Step S1: Obtain data information from the fully mechanized mining equipment, classify the data information to obtain multiple information categories, define the entities in each information category, and the relationship information between the entities; Step S2: Identify the data format of each type of data information, and extract entity, relation information and entity attribute values ​​from the data information in different ways based on different data formats to form triple data, and store the triple data in the Neo4j database; Step S3: Establish a question-and-answer template, convert the triplet data into question-and-answer pairs in Alpaca format based on the question-and-answer template, establish a natural language processing model, optimize the question-and-answer pairs based on the natural language processing model, and obtain optimized question-and-answer pairs; Step S4: Establish a large language model for the fully mechanized mining equipment. The large language model uses the Transformer model as its basic architecture. Optimized question-answer pairs are used as the training set to train the large language model. During the training process, the large language model is fine-tuned based on LoRA technology. Various evaluation indicators for the fully mechanized mining equipment are defined, and the evaluation indicators are quantified to obtain quantitative results. Step S5: Load the fine-tuned large language model. The user inputs natural language into the fully mechanized mining equipment decision support system to query information about the fully mechanized mining equipment. The fully mechanized mining equipment decision support system uses the fine-tuned large language model to perform entity recognition and intent recognition on the natural language. Based on the recognized entities and intents, a Cypher query statement is constructed. Step S6: Use the Cypher query statement to query the Neo4j database and determine whether a query result is obtained. If a query result is obtained, obtain a local feature subgraph based on the query result. If no query result is obtained, perform a multi-level relational reasoning query to obtain an information redundancy subgraph. Construct prompt words based on the local feature subgraph, the information redundancy subgraph, and the quantification results, and pass the prompt words to the large language model to obtain retrieval information that supports the large model's decision-making.

2. The decision support method for a large-scale model of fully mechanized mining equipment according to claim 1, characterized in that, The data that forms triples includes: The data format includes structured data, semi-structured data, and unstructured data. If the data information is in the format of structured data, it is directly converted into triple data. If the data information is in the format of semi-structured data, the semi-structured data is parsed to obtain knowledge information, and entity, relation information, and attribute values ​​of the entity are extracted from the knowledge information to obtain triple data. If the data information is in the format of unstructured data, a BiLSTM-CRF model is established. The BiLSTM-CRF model includes a bidirectional long short-term memory network model and a conditional random field discriminant model. The context information in the unstructured data is obtained based on the bidirectional long short-term memory network model. Entity information is identified from the unstructured data based on the context information. The boundaries of the entity information are identified based on the conditional random field discriminant model to obtain complete entity information and the relationship information between entities. The attribute information of the entities is extracted. Triple data is formed based on the entity information, relationship information, and attribute information.

3. The decision support method for a large-scale model of fully mechanized mining equipment according to claim 2, characterized in that, Based on the question-answering template, the triplet data is converted into question-answer pairs in Alpaca format, including: The natural language processing model is the PEGASUS model. The question in the question-answer pair is defined as the first information, and the answer corresponding to the question is defined as the second information. All the second information related to each first information is collected, and the second information is input into the PEGASUS model. The PEGASUS model fuses all the second information to obtain the third information. The third information is a coherent answer containing all the second information. The optimized question-answer pair is constructed based on the first information and the third information.

4. The decision support method for a large-scale model of fully mechanized mining equipment according to claim 1, characterized in that, Fine-tuning of the large language model based on LoRA technology includes: In the pre-training stage of the large language model, the LoRA technique is used to keep the original weight matrix of each Transformer layer in the Transformer model unchanged, and low-rank matrices A and low-rank matrix B are introduced in the training stage. The parameters of the two low-rank matrices are adjusted, and the fine-tuned first weight matrix W is obtained based on the first formula. ′ , the first formula is: W ′ =W + A·B, where W is the original weight matrix, and W ∈ R d×k , indicating that the original weight matrix is a real matrix with d rows and k columns, A ∈ R d×r , B ∈ R r×k , r represents the rank of matrix A or matrix B, where r << min(d, k). The fine-tuned large language model is trained using the first weight matrix on the optimized question-and-answer pairs.

5. The decision support method for a large-scale model of fully mechanized mining equipment according to claim 1, characterized in that, The evaluation indicators are quantified, including: A factor set is constructed based on the evaluation indicators, and a comment set is obtained by classifying the evaluation indicators into levels. A membership function is set, and the membership degree μ(x) of each evaluation indicator to the comment is set based on the membership function. The membership function is: x represents the real-time parameters of each evaluation indicator of the fully mechanized mining equipment, and a, b, c, and d are four preset key parameters. The membership degrees are summarized to obtain a single-factor evaluation matrix. The weight coefficients of each evaluation indicator are set, and fuzzy synthesis is performed based on the weight coefficients and the evaluation matrix to obtain a comprehensive evaluation result vector. The comprehensive evaluation result vector is then normalized, and the normalized comprehensive evaluation result vector is used as the quantification result.

6. The decision support method for a large-scale fully mechanized mining equipment according to claim 5, comprising constructing a Cypher query statement based on the identified entities and intentions, including: The input natural language is segmented into words, each segmented word is tagged with part-of-speech tags and input into a fine-tuned large language model. The model outputs words or phrases that are identified as entities or intents. The identified entities are mapped to the corresponding nodes in the Neo4j database. The identified intents are matched with the preset intent list in the system to obtain the intent type. If there are multiple intent types, the multiple intents are merged to obtain a comprehensive intent. The decision support system for fully mechanized mining equipment determines the query relationship between entities based on the identified intent. The decision support system for fully mechanized mining equipment constructs a Cypher query statement, retrieves the query relationship in the Neo4j database based on the Cypher query statement, and returns the query result to the decision support system for fully mechanized mining equipment.

7. The decision support method for a large-scale model of fully mechanized mining equipment according to claim 6, characterized in that, Perform multi-level relational reasoning queries to obtain an information redundancy subgraph, including: Based on the identified entities and the relationships between them, a query path is constructed. The relationships between entities include first-degree relationships, second-degree relationships, and third-degree relationships. The decision support system for the fully mechanized mining equipment queries local feature subgraphs based on the query path of first-degree relationships. If no local feature subgraph is obtained, the system then extends the query path to second-degree relationships and third-degree relationships to obtain the corresponding subgraph information. The system scores the relevance of each subgraph and defines the subgraph with the highest score as an information redundancy subgraph for output.

8. The decision support method for a large-scale model of fully mechanized mining equipment according to claim 7, characterized in that, Based on the local feature subgraph, the information redundancy subgraph, and the quantization result, prompt words are constructed, including: The decision support system for the fully mechanized mining equipment extracts key information from the local feature subgraph or the information redundancy subgraph. The key information includes entity name, type, attributes, and relationship information between entities. The key information is transformed into structured information, and the user's input natural language, the structured information, and the quantification results are combined to generate prompt words.

9. The decision support method for a large-scale model of fully mechanized mining equipment according to claim 1, characterized in that, The information categories include equipment information, risk information, fault information, cause information, solution information, and specification information.

10. A large-scale decision support system for fully mechanized mining equipment, used to implement the large-scale decision support method for fully mechanized mining equipment as described in any one of claims 1-9, characterized in that, The system includes: The definition module is used to acquire data information from the fully mechanized mining equipment, classify the data information to obtain multiple information categories, define the entities in each information category, and the relationship information between the entities; The triplet data processing module is used to identify the data format of each type of data information, extract entity, relation information and entity attribute values ​​from the data information in different ways based on different data formats, form triplet data, and store the triplet data in the Neo4j database; The question-answer pair formation module is used to establish a question-answer template, convert the triplet data into question-answer pairs in Alpaca format based on the question-answer template, establish a natural language processing model, and optimize the question-answer pairs based on the natural language processing model to obtain optimized question-answer pairs. The fuzzy comprehensive evaluation module is used to establish a large language model of the fully mechanized mining equipment. The large language model uses the Transformer model as its basic architecture, and the optimized question-answer pairs are used as the training set to train the large language model. During the training process, the large language model is fine-tuned based on LoRA technology, and various evaluation indicators for evaluating the fully mechanized mining equipment are defined, and the evaluation indicators are quantified to obtain the quantification results. The query statement construction module is used to load the fine-tuned large language model. When the user inputs natural language into the fully mechanized mining equipment decision support system to query information about the fully mechanized mining equipment, the fully mechanized mining equipment decision support system uses the fine-tuned large language model to perform entity recognition and intent recognition on the natural language, and constructs a Cypher query statement based on the recognized entities and intents. The query module is used to perform queries in the Neo4j database using Cypher query statements, determine whether query results are obtained, and if so, obtain a local feature subgraph based on the query results. If no query results are obtained, perform multi-level relational reasoning queries to obtain an information redundancy subgraph, construct prompt words based on the local feature subgraph or information redundancy subgraph, and pass the prompt words to the large language model to obtain retrieval information that supports the large model's decision-making.

Images