Risk pre-control decision support system for coal mine fire and gas disasters

Abstract

The invention relates to the field of prediction and early warning of coal mine disasters, and discloses a risk pre-control decision support system for coal mine fire and gas disasters, which adopts hierarchical and micro-service design and integrates a data layer, a model analysis layer and a decision application layer: designing a multi-dimensional sensor network in the data layer to comprehensively monitor coal mine environmental parameters; deploying a local large model in the model analysis layer to perform disaster-causing factor feature extraction and risk prediction, and constructing a local pre-control measure knowledge base to realize protective application of privatized data; and a visual display and man-machine interaction function is provided in the decision application layer, and a user is supported to adjust a decision according to an actual situation to generate a customized emergency response scheme. According to the system, the problems of multi-modal data fusion and privatized data protection are solved, the accuracy of fire and gas risk prediction and the pertinence of decision support are improved, and support is provided for accurate pre-control of coal mine fire and gas disasters.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine safety monitoring, disaster prediction and early warning, and relates to but is not limited to a risk pre-control decision support system for coal mine fire and gas disasters. Background Art

[0002] Coal mine fire and gas disasters are one of the most serious safety hazards in coal mine production, often leading to major economic losses and casualties. Most of my country's coal seams are spontaneously combustible or prone to spontaneous combustion, and all coal seams contain gas. These factors seriously threaten the safe production of coal mines. According to statistics, fire and gas disaster accidents account for a large proportion of major coal mine accidents, showing the urgency and necessity of accurate prevention of these disasters.

[0003] Traditional coal mine safety management mainly relies on manual inspections and basic monitoring equipment, which cannot meet the needs of real-time and comprehensive monitoring of complex and changeable underground environments. Many coal mines install fixed sensors at predetermined locations to monitor parameters such as methane concentration and temperature. However, due to the lack of optimized deployment strategies, key areas cannot be fully covered, resulting in an incomplete monitoring range, key areas may be overlooked, and there are safety hazards. Existing technologies cannot achieve early detection and early warning of coal mine fires and gas disasters, resulting in potential dangers that cannot be discovered and handled in a timely manner.

[0004] Some risk prediction and early warning systems provide basic decision support functions, but lack a comprehensive knowledge base and are unable to integrate historical data, expert insights and regulatory guidelines to provide accurate decision support. Moreover, independent solutions proposed for disaster-causing risks cannot be integrated with industry models or public safety management systems, reducing the overall disaster prevention effect. The single system design limits the integration of new technologies and system upgrades, and cannot meet changing safety needs. In addition, data security is also a key issue. Coal mine operation data is sensitive and proprietary, and relying on cloud services for data processing and analysis poses the risk of data leakage. Therefore, it is necessary to deploy advanced models locally to give full play to the advantages of artificial intelligence (AI) while ensuring data security.

[0005] Therefore, although there are some monitoring and data collection components in the field of coal mine safety, there is a lack of a system that can comprehensively monitor and assist in decision-making, especially under the premise of ensuring data security. Summary of the invention

[0006] In view of this, an embodiment of the present invention provides a risk prevention and control decision support system for coal mine fire and gas disasters, aiming to propose a comprehensive system that integrates optimized multi-dimensional sensor networks, local prevention measures knowledge base, data processing and analysis, and intelligent decision support methods to achieve early warning and precise prevention of coal mine fire and gas disasters.

[0007] The technical solution of the embodiment of the present invention is achieved as follows: The embodiment of the present invention provides a risk pre-control decision support system for coal mine fire and gas disasters, which adopts a layered and microservice architecture design, including: The data layer is responsible for the real-time collection, preprocessing and management of multimodal data in the well, including: a data source module, which is used to design and deploy a multidimensional sensor network for comprehensive three-dimensional disaster source data monitoring, and assign different weights to different types of sensors, and achieve accurate optimization of the sensor network through multi-objective optimization; a data access module, which is used to transmit the acquired sensor data to the well aggregation node through the multidimensional sensor network, and adopt a wired + wireless communication method to transmit the data of the well aggregation node to the well data base station through the well communication network, and use Python scripts and object-oriented programming mode at the well data base station to perform data preprocessing; The model analysis layer includes: a risk prediction module, which is used to extract disaster-causing factor features and predict risks based on the multimodal data obtained and preprocessed by the data layer using the locally deployed language model; a knowledge base construction module, which is used to construct a local preventive measures knowledge base based on the language model and the vector database, and conduct disaster risk assessment and classification through vector database retrieval combined with historical cases and accident experience; The decision application layer includes: a decision response module, which is used to generate customized real-time decision and emergency response plans for different risk levels based on the disaster factor characteristics and risk assessment results obtained by the model analysis layer, combined with the local preventive measures knowledge base; a human-computer interaction module, which is used to display risk status, diagnosis results and response strategies in real time through a visual interface, so that on-site personnel can quickly understand the disaster situation and make decision adjustments.

[0008] Furthermore, the data layer and the model analysis layer realize real-time transmission and update of business processing through data flow linkage; when disaster risks occur, the model analysis layer triggers the early warning and response of the decision application layer through an event-driven mechanism.

[0009] Furthermore, the multidimensional sensor network includes gas sensors, temperature sensors, pressure sensors, microseismic sensors, photoelectric smoke sensors, acoustic emission sensors, and distributed optical fiber temperature sensors deployed in preset areas in combination with the underground type and disaster scenario; the data source module obtains structured data, unstructured data, semi-structured data, and time series data through the multidimensional sensor network; wherein the structured data includes gas concentration, temperature, humidity, pressure, and displacement; the unstructured data includes video surveillance, audio recordings, and image data; the semi-structured data includes log files, equipment operation records, and alarm information; and the time series data includes time series data of sensors.

[0010] Furthermore, in the data source module, each sensor type s is assigned a corresponding weight according to its importance in disaster warning. , satisfying 0< ≤1, and construct the following optimization model to optimize the multidimensional sensor network: ; The constraints are ; In the formula, J is an objective function, representing the overall optimization goal of the multidimensional sensor network; S is the number of sensor types; is the weight of the s-th sensor; C s is the coverage of the s-th type of sensor; N s is the number of sensors of the sth type; t delay,s is the data transmission delay of the s-th sensor; α 1. α 2. α 3 are coverage weight coefficient, quantity weight coefficient, and delay weight coefficient respectively; C min,s is the minimum coverage requirement for category s sensors; N max,s is the maximum allowed number of sensors of type s; t max,s is the maximum delay threshold allowed for the s-th type of sensor.

[0011] Furthermore, the improved Transformer model utilizes the self-attention mechanism to process the multimodal time series data of coal mine fire and gas, and designs a multimodal input embedding layer to better capture the complex correlation between disaster-causing factors and achieve accurate extraction of disaster-causing factors.

[0012] Furthermore, the deployment process of the large language model is: install necessary software and dependent libraries on the local server; download weight files of Qwen-14B and Nomic-Embed-Text models through ollama; use anythingLLM to load Qwen-14B and Nomic-Embed-Text models and configure running parameters; run test cases to ensure that the model is deployed successfully and the performance meets the requirements; use the collected coal mine fire and gas accident cases and preventive measures texts to build a fine-tuning dataset; according to predefined tasks, perform supervised fine-tuning of Qwen-14B and Nomic-Embed-Text models; optimize the following hyperparameters: learning rate, batch size, remove redundant neurons and connections, and perform parameter compression and knowledge distillation.

[0013] Furthermore, the local preventive measures knowledge base is constructed through the following process: collecting the multimodal monitoring data collected by the multidimensional sensor network, as well as relevant historical accident data and preventive measures text, and performing cleaning, standardization and structured processing; using the locally deployed and optimized language large model to vectorize the preprocessed text to obtain a semantic vector; loading the semantic vector into the FAISS vector database, and setting the vector index HNSW to accelerate similarity retrieval. Appropriate indexes are set to speed up the retrieval speed, and the semantic vector is stored together with the corresponding original text and metadata; the stored knowledge is classified according to the dimensions of disaster factors, preventive measures types, and accident types, and keywords and tags are added to each piece of knowledge; designing and implementing the interface between the local preventive measures knowledge base and the language large model.

[0014] Furthermore, the decision-making application layer is also designed to protect private data of typical coal mine fire and gas disasters from three aspects: data classification and hierarchical protection, access control and authority management, and data security transmission and storage, to prevent the leakage of sensitive information and ensure the security of the company's data assets.

[0015] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least: In an embodiment of the present invention, based on the generative intelligent cognitive model of the risk origin of typical coal mine fire and gas disasters and the risk origin preventive measures, combined with field data, an auxiliary decision-making generation system with functions such as multimodal data convergence, hierarchical prediction and human-computer interaction is constructed to achieve early warning and precise prevention of disaster risks during mining. A multidimensional sensor network deployment method for multiple scenes underground is proposed to ensure comprehensive monitoring of key environmental parameters; at the same time, a large model is deployed locally for data processing and analysis to protect sensitive coal mine operation data; a local preventive measures knowledge base is constructed, and customized preventive suggestions are generated in combination with historical accident data and expert analysis; finally, a layered and microservice system architecture is designed, which is flexible and scalable, convenient for integrating new technologies and modules, and enables the system to be connected with industry-level models and public safety management systems. The risk prevention decision support system provided by the present invention greatly improves the early detection and prevention capabilities of coal mine fire and gas disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which: Figure 1A schematic diagram of the system framework of the risk pre-control decision support system provided by an embodiment of the present invention; Figure 2 A schematic diagram of the deployment and optimization process of a local language large model provided by an embodiment of the present invention; Figure 3 A block diagram of a preventive control measures knowledge base system provided by an embodiment of the present invention; Figure 4 A design framework diagram of a coal mine fire and gas data protection solution provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme 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 part of the embodiments of the present invention, rather than all the embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments in 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.

[0018] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0019] Those skilled in the art will appreciate that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the field to which the embodiments of the present invention belong. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0020] The embodiment of the present invention provides a risk pre-control decision support system for coal mine fire and gas disasters, which adopts a layered and microservice architecture design, including: The data layer is responsible for the real-time collection, preprocessing and management of multimodal data in the well, including: a data source module, which is used to design and deploy a multidimensional sensor network for comprehensive three-dimensional disaster source data monitoring, and assign different weights to different types of sensors, and achieve accurate optimization of the sensor network through multi-objective optimization; a data access module, which is used to transmit the acquired sensor data to the well aggregation node through the multidimensional sensor network, and adopt a wired + wireless communication method to transmit the data of the well aggregation node to the well data base station through the well communication network, and use Python scripts and object-oriented programming mode at the well data base station to perform data preprocessing; The model analysis layer includes: a risk prediction module, which is used to extract disaster-causing factor features and predict risks based on the multimodal data obtained and preprocessed by the data layer using the locally deployed language model; a knowledge base construction module, which is used to construct a local preventive measures knowledge base based on the language model and the vector database, and conduct disaster risk assessment and classification through vector database retrieval combined with historical cases and accident experience; The decision application layer includes: a decision response module, which is used to generate customized real-time decision and emergency response plans for different risk levels based on the disaster factor characteristics and risk assessment results obtained by the model analysis layer, combined with the local preventive measures knowledge base; a human-computer interaction module, which is used to display risk status, diagnosis results and response strategies in real time through a visual interface, so that on-site personnel can quickly understand the disaster situation and make decision adjustments.

[0021] like Figure 1 The system framework diagram of the risk pre-control decision support system provided by the embodiment of the present invention is shown. The system adopts a layered architecture design and is mainly divided into three levels: data layer (data source and data access), model analysis layer and decision application layer. Each layer has clear functions and interfaces, and has flexibility and scalability, which facilitates the integration of new technologies and modules, and enables the system to be connected with industry-level models and public safety management systems. The system adopts the microservice architecture design principle, and each functional module is independently deployed as a microservice unit. Different microservice units communicate through standardized interfaces and protocols, supporting efficient integration and expansion of heterogeneous systems.

[0022] At the data level, we first clarify the characteristics and types of typical coal mine fire and gas disasters. In view of the complex characteristics of typical coal mine fire and gas disasters, we design multiple sensor layout locations in combination with underground types and disaster scenarios to form a multi-dimensional sensor network. At the same time, we consider the role of different types of sensors, assign different weights, and achieve more accurate sensor network design through multi-objective optimization.

[0023] Then the data is acquired and transmitted to the underground aggregation node through the multi-dimensional sensor network. The node data is transmitted to the uphole data base station, i.e., the data center, through the downhole communication network in a "wired + wireless" manner. The collected data is preprocessed through the uphole data base station, and the response slot function of the prediction module in the model analysis layer is constructed through Python programming and object-oriented programming mode.

[0024] The data preprocessing process includes: taking different types of data such as video, image, gas, temperature as input, the data access module will synchronize and preprocess the data to ensure that the data of the input model analysis layer meets the requirements of structure and timing. For data missing caused by sensor failure or signal interruption, interpolation, mean filling or deletion are used. Abnormal data is detected by machine learning algorithm and corrected or eliminated. After cleaning, the sensor data is filtered to eliminate high-frequency noise; for signal data such as audio and vibration, wavelet transform and Fourier transform are used to extract effective features. Multimodal data of different sensors and different types are converted into a unified format and unit, and multi-source data are time-aligned to ensure data consistency in the time dimension. NoSQL database is used to store data according to predefined patterns, and metadata and tags are added to the data for fast retrieval and association analysis. In the process of data transmission, encryption protocols such as SSL / TLS are used to ensure data security. Finally, a unified data interface and real-time data stream processing are set to ensure the accuracy of data during transmission.

[0025] The Python program is called in the system background to quickly preprocess multimodal data such as video image data, gas concentration, temperature, microseismic signals, smoke concentration, etc., and generate a unified standardized data format to meet the input requirements of the risk prediction module in the subsequent model analysis layer.

[0026] In the model analysis layer, typical disasters such as coal mine fire and gas are predicted and assessed, and the current disaster risk level is predicted to provide a basis for the next step of decision support. When disaster risks occur, the model analysis layer triggers the early warning and response of the decision application layer through an event-driven mechanism. Through data linkage and event response, the system's ability to quickly identify potential disasters and respond to emergencies is ensured. At the same time, since the private data of typical disasters such as coal mine fire and gas is huge and contains sensitive information, a large model is deployed locally for data processing and analysis to protect sensitive coal mine operation data, prevent data leakage, and improve the real-time and security of data processing.

[0027] The decision application layer is the core output module of the entire system, supporting intelligent decision generation and emergency response for typical disasters of coal mine fire and gas. Combining the risk assessment results and the local pre-control knowledge base, the system generates personalized risk prevention suggestions. On the one hand, based on the pre-control measures provided by the knowledge base, the system generates customized pre-control and emergency response strategies for different risk levels. For example, for high-risk events caused by increased gas concentration, the system will recommend immediate start-up of the gas emission system and on-site detection; for signs of fire, the system will provide isolation, fire extinguishing and other measures and evacuate personnel in advance. On the other hand, the system allows on-site operators to adjust decision suggestions according to actual conditions through the human-computer interaction interface. On-site personnel can input more information and data, and the system will automatically readjust the response plan based on the input information to ensure the flexibility and adaptability of decision-making.

[0028] When a user or system module needs to retrieve a specific type of preventive measures or similar cases, the user enters a query text such as "preventive measures for gas outbursts in goaf areas". The system embeds the query text into a vector through the big model, performs similarity matching with the vector representation in the vector database, and retrieves the corresponding preventive measures and decision suggestions from the most relevant text entries. In actual decision-making scenarios, the search results are combined with the understanding and reasoning capabilities of the local big model to form customized emergency response strategies or auxiliary decision-making suggestions, providing accurate and actionable guidance for on-site personnel.

[0029] In the embodiment of the present invention, a multi-dimensional sensor network for multiple scenarios underground is deployed to ensure comprehensive monitoring of key environmental parameters; a large model is deployed locally for data processing and analysis to protect sensitive coal mine operation data, prevent data leakage, and improve the real-time and security of data processing; a local preventive measures knowledge base is built, and combined with historical accident data and expert analysis, customized preventive suggestions are generated to provide accurate decision support; a layered and microservice system architecture is designed with flexibility and scalability to facilitate the integration of new technologies and modules, and enable the system to be connected with industry-level models and public safety management systems. Through improvements, the present invention aims to create a comprehensive, safe, and intelligent decision support system to significantly improve the early detection and prevention capabilities of coal mine fires and gas disasters.

[0030] Furthermore, the data layer and the model analysis layer realize real-time transmission and update of business processing through data flow linkage; when disaster risks occur, the model analysis layer triggers the early warning and response of the decision application layer through an event-driven mechanism.

[0031] Here, the system adopts the microservice architecture design principle and deploys each functional module independently as a microservice unit. These microservice units communicate through standardized interfaces and protocols to support efficient integration and expansion of heterogeneous systems. In this way, through data linkage and event response, the risk prevention and control decision support system can ensure rapid identification and emergency response capabilities for potential disasters.

[0032] Furthermore, the multidimensional sensor network includes gas sensors, temperature sensors, pressure sensors, microseismic sensors, photoelectric smoke sensors, acoustic emission sensors, and distributed optical fiber temperature sensors deployed in preset areas in combination with the underground type and disaster scenario; the data source module obtains structured data, unstructured data, semi-structured data, and time series data through the multidimensional sensor network; wherein the structured data includes gas concentration, temperature, humidity, pressure, and displacement; the unstructured data includes video surveillance, audio recordings, and image data; the semi-structured data includes log files, equipment operation records, and alarm information; and the time series data includes time series data of sensors.

[0033] Here, the selection of sensor types and key deployment points mainly include: Gas sensors are used to detect the concentration of key gases such as methane (CH4), carbon monoxide (CO), carbon dioxide (CO2), and oxygen (O2). Temperature sensors monitor temperature changes and identify early signs of coal spontaneous combustion. Stress sensors monitor changes in rock stress to prevent gas outbursts and collapse accidents. Microseismic sensors capture microseismic activities and warn of collapse and gas dynamic phenomena. Photoelectric smoke detectors detect smoke concentrations and quickly identify fires. Acoustic emission sensors monitor the acoustic signals generated by the development of cracks in the coal body. Distributed fiber optic temperature sensors continuously monitor temperature changes along the fiber optic line. Each sensor node transmits data to the underground aggregation node through the sensor network. Through the underground communication network, the data is transmitted to the data base station on the well using a "wired + wireless" method. Therefore, by planning the deployment locations of different types of sensors and considering the different functions of sensors in the multi-objective optimization model and assigning different weights, the accurate optimization of the sensor network and comprehensive three-dimensional disaster source data monitoring can be achieved.

[0034] Furthermore, in the data source module, each sensor type s is assigned a corresponding weight according to its importance in disaster warning. , satisfying 0< ≤1, and construct the following optimization model to optimize the multidimensional sensor network: ; The constraints are ; In the formula, J is an objective function, representing the overall optimization goal of the multidimensional sensor network; Sis the number of sensor types; is the weight of the s-th sensor; C s is the coverage of the s-th type of sensor; N s is the number of sensors of the sth type; t delay,s is the data transmission delay of the s-th sensor; α 1. α 2. α 3 are coverage weight coefficient, quantity weight coefficient, and delay weight coefficient respectively; C min,s is the minimum coverage requirement for category s sensors; N max,s is the maximum allowed number of sensors of type s; t max,s is the maximum delay threshold allowed for the s-th type of sensor.

[0035] In this way, the role of different types of sensors is considered, different weights are assigned, and a more accurate sensor network design is achieved through multi-objective optimization. In the implementation, the initial sensor weights and parameters are first determined, and different types of sensors and their initial weights are listed. The minimum value of the objective function is calculated through the above optimization model to determine whether the constraints are met. If not, the weighted particle swarm optimization algorithm (Weighted Particle Swarm Optimization, WPSO) is used to initialize the particle swarm for each sensor type, and set the initial position, speed and weight. The fitness value of each particle is calculated, which is the objective function J. The weight is dynamically adjusted according to the importance of the sensor and the current optimization status. Until the above constraints are met, the optimized sensor layout plan and parameters are output.

[0036] In this way, by considering the different functions of sensors in the mathematical optimization model and assigning different weights, accurate optimization of the sensor network and comprehensive three-dimensional disaster source data monitoring can be achieved, thereby providing high-quality data support for subsequent disaster factor feature extraction and risk prediction.

[0037] Furthermore, the deployment process of the large language model is: install necessary software and dependent libraries on the local server; download weight files of Qwen-14B and Nomic-Embed-Text models through ollama; use anythingLLM to load Qwen-14B and Nomic-Embed-Text models and configure running parameters; run test cases to ensure that the model is deployed successfully and the performance meets the requirements; use the collected coal mine fire and gas accident cases and preventive measures texts to build a fine-tuning dataset; according to predefined tasks, perform supervised fine-tuning of Qwen-14B and Nomic-Embed-Text models; optimize the following hyperparameters: learning rate, batch size, remove redundant neurons and connections, and perform parameter compression and knowledge distillation.

[0038] Here, if Figure 2 As shown in the figure, first prepare the environment, including hardware requirements, software dependencies, development environment, network connection, configuration files, and system parameter adjustment. Then download and load the model: in terms of model selection, select the large language model Qwen-14B (capable of processing complex professional field texts, suitable for applications in the field of coal mine safety) and Nomic-Embed-Text (capable of converting text into high-quality semantic vectors, suitable for building vector databases, and supporting similarity retrieval). The tools selected are ollama (used to deploy and manage large models on local servers, supporting model downloads, loading, and running) and anythingLLM (used to build a localized LLM application framework, providing an API interface for easy interaction with the model). After the model is successfully deployed, fine-tune the data set to make it more suitable for the field of coal mine fire and gas disaster prevention: use the collected coal mine fire and gas accident cases and preventive control measures to build a fine-tuning data set. According to the predefined tasks (question-answering, classification), supervised fine-tuning of the model is performed to enable it to better understand coal mine safety-related terms, situations, and response plans. Optimize hyperparameters such as learning rate and batch size to improve the performance of the model in the field of coal mine fire and gas disasters. Finally, optimize and compress the model: remove redundant neurons and connections in the model to reduce the size of the model. Prune, quantize or knowledge distill the model to reduce runtime resource usage and improve the operating efficiency and response speed of local deployment. In terms of parameter compression, compress the model parameters from 32-bit floating point numbers to 8 or 16 bits to reduce computing and storage requirements. In terms of knowledge distillation, train a smaller student model to learn the knowledge of the large model and obtain similar performance. The optimized local large model can provide rich language understanding and feature extraction for the input text, and provide useful intermediate representation and feature information for subsequent vector representation and retrieval.

[0039] In this way, the embodiment of the present invention uses the method of locally deploying a large model to realize the protective application of privatized data. The large model is deployed locally for data processing and analysis, protecting sensitive coal mine operation data, preventing data leakage, and improving the real-time and security of data processing.

[0040] Furthermore, the local preventive measures knowledge base is constructed through the following process: collecting the multimodal monitoring data collected by the multidimensional sensor network, as well as relevant historical accident data and preventive measures text, and performing cleaning, standardization and structured processing; using the locally deployed and optimized language large model to vectorize the preprocessed text to obtain a semantic vector; loading the semantic vector into the FAISS vector database, and setting the vector index HNSW to accelerate similarity retrieval. Appropriate indexes are set to speed up the retrieval speed, and the semantic vector is stored together with the corresponding original text and metadata; the stored knowledge is classified according to the dimensions of disaster factors, preventive measures types, and accident types, and keywords and tags are added to each piece of knowledge; designing and implementing the interface between the local preventive measures knowledge base and the language large model.

[0041] Here, the block diagram of the preventive measures knowledge base system is as follows Figure 3 As shown, it includes: a text processing module, which converts texts such as preventive measures, accident cases, and regulatory standards into semantic vectors through the Nomic-Embed-Text model. The vector dimension is as high as 768 or more to ensure the richness of semantic information; a database module, which uses the high-performance vector database FAISS (Facebook AI Similarity Search) and uses vector indexes (HNSW) to accelerate similarity retrieval, and finally stores the vectors together with the corresponding original text and metadata for easy retrieval and display. The metadata includes data sources, timestamps, category labels, and related case identifiers. Metadata helps filter, sort, and interpret retrieval results; a classification and storage module, which classifies knowledge according to dimensions such as disaster factors, preventive measures, and accident types, and adds keywords and labels to each piece of knowledge to enhance the accuracy of retrieval; a data retrieval module, which uses cosine similarity to measure the similarity between the query vector and the vector in the database. And a fast approximate nearest neighbor search algorithm is used to achieve millisecond-level retrieval response; an update mechanism module, which regularly adds new accident cases and preventive measures to maintain the timeliness of the knowledge base.

[0042] At the logical level, various preventive measures, historical cases, industry regulations and other information are integrated into a structured knowledge base system: classification and labeling, classification and labeling according to the type of disaster-causing factors (coal spontaneous combustion, gas outburst, gas explosion, etc.), scene (goaf, tunnel, working face), preventive measures attributes (prevention, detection, emergency response) and other dimensions. Association and hierarchy, establish the association relationship between knowledge items, such as associating the corresponding historical cases with relevant preventive measures and regulations through metadata. Update and maintenance, regularly update the knowledge base, add new cases, expert advice and changes in regulations and policies, and ensure the timeliness and dynamic adaptability of the knowledge base.

[0043] Therefore, by using the embedded model processing (Nomic-Embed-Text) to vectorize the data, a local preventive measures knowledge base based on the vector database was constructed, achieving targeted and accurate decision support.

[0044] Furthermore, the decision-making application layer is also designed to protect private data of typical coal mine fire and gas disasters from three aspects: data classification and hierarchical protection, access control and authority management, and data security transmission and storage, to prevent the leakage of sensitive information and ensure the security of the company's data assets.

[0045] Here, the coal mine fire and gas data protection framework is as follows Figure 4 As shown in the figure, in terms of data classification and hierarchical protection, a data classification module is designed to classify data into different levels such as public, internal, sensitive, and confidential according to the sensitivity of the data. Corresponding protection measures are formulated for data of different levels.

[0046] In terms of access control and permission management, we designed access control module and audit module. The access control module adopts a combination of role-based access control (RBAC) and attribute-based access control (ABAC) to strictly limit user access rights to data. The audit module audits the access and operation of data throughout the process, and records information such as user, time, and operation type for easy traceability afterwards.

[0047] In terms of secure data transmission and storage, a network security module and an encryption storage module are designed. The network security module adopts the SSL / TLS protocol to ensure the confidentiality and integrity of data during network transmission; the encryption storage module performs high-intensity encryption storage on sensitive data to prevent data leakage caused by loss or theft of physical media.

[0048] Based on the protective scheme design knowledge base expansion method, it is convenient to later connect the local preventive measures knowledge base and large model to the mine safety management model, coal industry large model, urban public safety management system and other industry / government large-scale comprehensive analysis and management systems.

[0049] The knowledge base and large model functions are modularized and divided into data layer, model layer, interface layer, etc. to facilitate independent deployment of functions. The RESTful API standard protocol is adopted to design a unified data access and model call interface to ensure compatibility between different systems. The knowledge base and model are encapsulated to provide a controllable access method, which not only protects internal data and models, but also facilitates external system calls. In terms of compatibility, the design supports common data formats such as JSON, XML, and CSV to facilitate data exchange with different systems. At the same time, a distributed architecture design is adopted to support horizontal expansion to meet the needs of large-scale data processing and concurrent access.

[0050] Example application scenarios include: 1) Mine comprehensive large model integration: Use the local knowledge base and large model as sub-modules of the mine comprehensive large model to provide professional support for coal mine fire and gas prevention and control measures. 2) Coal industry large model sharing: Share the data and model capabilities of the knowledge base with the large model of the coal industry through standard interfaces to improve the overall safety production level of the industry. 3) Urban public safety management system application: In the urban public safety management system, introduce coal mine fire and gas prevention and control knowledge to provide reference for the energy industry and support urban emergency management and disaster prevention and control.

[0051] The risk prevention and control decision support system for coal mine fire and gas disasters provided by the embodiment of the present invention realizes comprehensive monitoring of multiple scenes underground through optimized multi-dimensional sensor network design, ensuring real-time acquisition and analysis of key environmental parameters. The present invention protects sensitive coal mine operation data and prevents data leakage by locally deploying large models for data processing and analysis. At the same time, a local prevention and control measures knowledge base is constructed, and customized prevention and control recommendations are generated by combining historical accident data and expert analysis. The present invention designs a layered and microservice system architecture, which is flexible and scalable, and is convenient for integrating new technologies and modules. The system provided by the present invention can be connected with industry-level models and public safety management systems, and contribute to a comprehensive disaster prevention strategy.

[0052] The present invention has broad application prospects in the coal mining industry and related fields: 1) By providing accurate early warning and customized preventive measures, the occurrence of fires and gas explosions can be significantly reduced, and the lives of miners and corporate assets can be protected. 2) This system can serve as a benchmark for coal mine safety management and promote the standardization and normalization of disaster prevention measures in the industry. 3) The present invention complies with the national policy of improving industrial safety and national security, and can be integrated into the national security management framework to enhance regulatory compliance. 4) Due to the modular design of the system, it can adapt to different mining environments, including metal mines, etc., to expand the scope of application. 5) By introducing advanced AI models and data analysis technologies, it helps mines to transform digitally and intelligently, and improve operational efficiency and safety. 6) The local deployment of AI models sets an example for data security in industrial applications, emphasizing the importance of protecting sensitive information. 7) Integration with city-level public safety systems can enhance emergency response capabilities and contribute to a broader disaster management strategy. 8) Disaster prevention can reduce downtime and accident-related financial losses, and improve the profitability and sustainability of coal mine operations.

[0053] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present invention, the size of the serial number of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention. The serial numbers of the above-mentioned embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0054] It should be noted that, in this article, the terms "include", "comprises" or any other variations 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. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0055] The above is only an embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A risk pre-control decision support system for coal mine fire and gas disasters, characterized in that: Adopt layered and microservices architecture design, including: The data layer is responsible for the real-time collection, preprocessing and management of multimodal data in the well, including: a data source module, which is used to design and deploy a multidimensional sensor network for comprehensive three-dimensional disaster source data monitoring, and assign different weights to different types of sensors, and achieve accurate optimization of the sensor network through multi-objective optimization; a data access module, which is used to transmit the acquired sensor data to the well aggregation node through the multidimensional sensor network, and adopt a wired + wireless communication method to transmit the data of the well aggregation node to the well data base station through the well communication network, and use Python scripts and object-oriented programming mode at the well data base station to perform data preprocessing; The model analysis layer includes: a risk prediction module, which is used to extract disaster-causing factor features and predict risks based on the multimodal data obtained and preprocessed by the data layer using the locally deployed language model; a knowledge base construction module, which is used to construct a local preventive measures knowledge base based on the language model and the vector database, and conduct disaster risk assessment and classification through vector database retrieval combined with historical cases and accident experience; The decision application layer includes: a decision response module, which is used to generate customized real-time decision and emergency response plans for different risk levels based on the disaster factor characteristics and risk assessment results obtained by the model analysis layer, combined with the local preventive measures knowledge base; a human-computer interaction module, which is used to display risk status, diagnosis results and response strategies in real time through a visual interface, so that on-site personnel can quickly understand the disaster situation and make decision adjustments.

2. The risk pre-control decision support system according to claim 1 is characterized in that: The data layer and the model analysis layer realize real-time transmission and update of business processing through data flow linkage; when disaster risks occur, the model analysis layer triggers the early warning and response of the decision application layer through an event-driven mechanism.

3. The risk pre-control decision support system according to claim 1 is characterized in that: The data source module acquires structured data, unstructured data, semi-structured data and time series data through the multi-dimensional sensor network; wherein the structured data includes gas concentration, temperature, humidity, pressure, and displacement; the unstructured data includes video monitoring, audio recording, and image data; the semi-structured data includes log files, equipment operation records, and alarm information; and the time series data includes time series data of sensors.

4. The risk pre-control decision support system according to claim 3 is characterized in that: In the data source module, each sensor type s is assigned a corresponding weight according to its importance in disaster warning. , satisfying 0< ≤1, and construct the following optimization model to optimize the multidimensional sensor network: ； The constraints are ; In the formula, J is an objective function, representing the overall optimization goal of the multidimensional sensor network; S is the number of sensor types; is the weight of the s-th sensor; C s is the coverage of the s-th type of sensor; N s is the number of sensors of the sth type; t delay,s is the data transmission delay of the s-th sensor; α 1. α 2. α 3 are coverage weight coefficient, quantity weight coefficient, and delay weight coefficient respectively; C min,s is the minimum coverage requirement for category s sensors; N max,s is the maximum allowed number of sensors of type s; t max,s is the maximum delay threshold allowed for the s-th type of sensor.

5. The risk pre-control decision support system according to claim 1 is characterized in that: The deployment process of the large language model is: install the necessary software and dependent libraries on the local server; download the weight files of the Qwen-14B and Nomic-Embed-Text models through ollama; Use anythingLLM to load the Qwen-14B and Nomic-Embed-Text models and configure the running parameters; run the test cases to ensure that the model deployment is successful and the performance meets the requirements; use the collected coal mine fire and gas accident cases and preventive control measures texts to build a fine-tuning dataset; perform supervised fine-tuning on the Qwen-14B and Nomic-Embed-Text models according to the predefined tasks; Optimize the following hyperparameters: learning rate, batch size, remove redundant neurons and connections, and perform parameter compression and knowledge distillation.

6. The risk pre-control decision support system according to claim 1, characterized in that: The local preventive measures knowledge base is constructed through the following process: Collecting the multimodal monitoring data collected by the multidimensional sensor network, as well as relevant historical accident data and preventive measures text, and performing cleaning, standardization and structured processing; The pre-processed text can be vectorized using the locally deployed and optimized language model to obtain a semantic vector. Load the semantic vector into the FAISS vector database, set the vector index HNSW to accelerate similarity retrieval and an appropriate index to speed up the retrieval, and store the semantic vector together with the corresponding original text and metadata; Classify the stored knowledge according to the dimensions of disaster-causing factors, types of preventive measures, and types of accidents, and add keywords and tags to each piece of knowledge; Design and implement the interface between the local preventive measures knowledge base and the language large model.

7. The risk pre-control decision support system according to claim 1, characterized in that: The decision-making application layer also includes a data protection application module, which is used to design and protect private data of typical coal mine fire and gas disasters from three aspects: data classification and hierarchical protection, access control and authority management, and data security transmission and storage, to prevent the leakage of sensitive information and ensure the security of the company's data assets.

Images