A mine management system and method based on a mine management and control platform
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GANSU JINGTIESHAN MINING CO LTD
- Filing Date
- 2026-02-13
- Publication Date
- 2026-06-05
Smart Images
Figure CN122155457A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent mine management technology, specifically a mine management system and method based on a mine control platform. Background Technology
[0002] With the accelerated intelligent transformation of the mining industry, mine management and control platforms have become the core carrier for realizing automated mine production and visualized safety. Currently, most mainstream mine management and control platforms integrate basic technologies such as the Internet of Things, big data analysis, and video surveillance, and have made some progress in data collection and status monitoring. However, in practical applications, there are still many technical defects that are difficult to overcome, which seriously restrict the level of intelligence in mine management and the ability to ensure safety. First, multi-source data fusion is inefficient and prone to privacy breaches. Data generated during mining operations, including equipment operation data (such as excavator speed and conveyor load), environmental monitoring data (such as methane concentration and dust content), and personnel management data (such as location information and work qualifications), suffers from heterogeneous formats and inconsistent standards. Existing management platforms often employ a centralized data fusion architecture, requiring all raw data to be uploaded to a cloud server for unified processing. This not only leads to high data transmission latency (especially in underground mines with weak network signals, latency can reach several minutes) but also poses risks of commercial secret and personnel privacy breaches when data is shared across departments. For example, equipment operation data in an underground coal mine contains core production process parameters, making centralized storage vulnerable to unauthorized access; personnel location data involves work trajectory privacy and lacks effective protection mechanisms. While some existing technologies employ encrypted data transmission, they do not address the conflict between privacy protection and low latency at the fusion architecture level, and centralized processing still cannot fundamentally prevent the exposure of raw data. Secondly, risk warnings are delayed and predictive capabilities are insufficient. Existing mine management platforms rely heavily on fixed threshold triggering mechanisms for safety warnings, meaning alarms are only issued when monitored data (such as gas concentration and equipment temperature) exceed preset thresholds. This approach can only identify explicit risks that have already occurred and cannot predict potential risks such as hidden equipment failures or minor geological deformations. For example, in the early stages of bearing wear in mining equipment, temperature and vibration data may not reach thresholds, but long-term operation can lead to sudden failures, which existing platforms cannot warn of in advance. Similarly, minor deformations in the surrounding rock of underground mine roadways (daily deformation less than 1 mm) are difficult to detect through conventional threshold monitoring but may gradually develop into collapse accidents. Furthermore, existing warning systems lack the ability to analyze risks associated with multiple factors, such as the correlation between increased environmental humidity and electrical equipment failures, or the coupling relationship between geological structural changes and gas emissions. This results in low warning accuracy and high false alarm rates, failing to provide effective support for safety decision-making. Secondly, the coordination between equipment and personnel scheduling is rigid. Mining production involves multiple stages such as mining, transportation, ventilation, and drainage, and the coordinated scheduling of equipment and personnel directly affects operational efficiency. Existing management platforms often rely on pre-set fixed processes or manual ad-hoc planning, lacking the ability to adapt to real-time changes in operating conditions. For example, when transportation routes become blocked due to geological subsidence, the scheduling plan for mining equipment and transport vehicles cannot be optimized in real time, leading to mining stagnation and transportation backlogs. Furthermore, the allocation of personnel work areas does not consider equipment operating status and environmental risk distribution, potentially resulting in safety hazards from personnel working alongside high-risk equipment. In addition, existing scheduling systems do not adequately consider factors such as equipment load balancing and personnel skill matching, leading to unreasonable resource allocation and low operational efficiency. For instance, some equipment may operate at full capacity for extended periods while others remain idle, or unqualified personnel may be assigned high-difficulty tasks. Finally, the emergency response is passive and lacks coordination. Mine accidents (such as water inrush, fire, and gas explosion) are characterized by their suddenness and wide-ranging hazards, requiring rapid coordinated responses from multiple departments and nodes. Existing management platforms often rely on independent responses from single modules, lacking a unified information synchronization and coordination mechanism. For example, in the event of a water inrush accident, the personnel evacuation module can only issue evacuation instructions but cannot simultaneously notify the equipment shutdown module to cut off power, the transportation module to adjust rescue channels, or the rescue center to plan the optimal route. Emergency information transmission uses traditional communication methods, which are prone to information tampering, delays, or loss, leading to flawed emergency decision-making. Furthermore, existing emergency systems lack the ability to simulate accident scenarios and cannot develop dynamic response plans based on real-time accident development trends. They rely heavily on pre-set emergency plans, resulting in poor adaptability and difficulty in handling complex and ever-changing accident scenarios. In summary, existing mine management platforms have significant technical deficiencies in core areas such as data fusion, risk warning, scheduling and coordination, and emergency response. Upgrading a single technology can no longer meet the needs of intelligent and safe mine management. There is an urgent need to build an innovative solution that integrates multiple technologies to achieve efficient, safe, and intelligent operation of the entire mine management process. Summary of the Invention
[0003] To address the problems in existing technologies, such as low efficiency of multi-source heterogeneous data fusion, privacy risks associated with sharing raw data, reliance on fixed thresholds for risk warning which fails to predict potential associated risks, resulting in delayed and inaccurate warnings, rigid equipment and personnel scheduling schemes which cannot adapt to real-time changes in operating conditions and lead to unreasonable resource allocation, and the lack of multi-node linkage mechanisms in emergency response, untimely information synchronization, and the inability to simulate accident scenarios, this invention provides a mine management system and method based on a mine management platform.
[0004] The technical solution adopted by this invention to solve its technical problem is: a mine management system based on a mine management platform, comprising: The data acquisition layer includes a micro-vibration / stress sensing network, equipment operation sensors, environmental monitoring sensors, and a UWB high-precision positioning module, used to collect multi-source heterogeneous data on mining equipment, geology, environment, and personnel. The edge computing layer contains several edge computing gateways deployed in the mining operation area, which are used for preprocessing of collected data, including format conversion, outlier removal, and feature extraction. The Federated Learning-Digital Twin Fusion Layer includes a Federated Learning Server and a Digital Twin. The Federated Learning Server receives preprocessed feature data from each edge node, trains a global fusion model through horizontal federated learning, and distributes updates. The Digital Twin is built based on the actual mining scenario and maps the unified data output by the global fusion model in real time. The intelligent decision-making layer includes a deep learning time series prediction module and a reinforcement learning scheduling module. The time series prediction module predicts potential risks such as equipment failure and geological deformation based on fused data, while the reinforcement learning scheduling module dynamically adjusts the equipment-personnel scheduling scheme with the goal of improving work efficiency and safety. The blockchain emergency response layer contains multiple nodes (operation area, dispatch center, and rescue center), adopts a practical Byzantine fault-tolerant consensus mechanism, synchronizes emergency status information, and links digital twins to complete emergency simulations.
[0005] Specifically, the federated learning server supports dynamic node access. Each edge node only uploads model parameters and does not upload raw data. The raw data is stored in a local encrypted database.
[0006] Specifically, the digital twin includes a terrain mirroring module, an equipment twin module, a personnel twin module, and an environment twin module, with a mapping update frequency of no less than 10Hz.
[0007] Specifically, the deep learning time series prediction module adopts an LSTM+attention mechanism architecture, with input features including historical running data, real-time sensing data and environmental correlation data, and outputs risk probability and trend curves.
[0008] Specifically, the state space of the reinforcement learning scheduling module includes equipment location / load, personnel location / qualification, and task progress; the action space includes equipment path planning and personnel area allocation; and the reward function is α×work efficiency + β×safety coefficient (α and β are weighting coefficients, and α+β=1).
[0009] Specifically, the positioning accuracy of the UWB high-precision positioning module is no less than 10cm, and the positioning signal penetration depth is adapted to underground mining scenarios up to 500m.
[0010] Specifically, the node data of the blockchain emergency response layer is stored using asymmetric encryption, and emergency commands are automatically triggered and executed through smart contracts.
[0011] Specifically, the sensor deployment density of the micro-vibration / stress sensing network is per 100m 2 At least one, with a detection accuracy of 0.01mm micro-deformation and 0.1MPa stress change.
[0012] A mine management method based on a mine management platform, implemented using the aforementioned mine management system, includes the following steps: S1: Data Acquisition, which collects equipment operating parameters, geological micro-vibration / stress data, environmental monitoring data and personnel UWB positioning data in real time through the data acquisition layer; S2: Edge preprocessing, the edge computing gateway performs format standardization, outlier removal and feature extraction on the collected data; S3: Data fusion, each edge node trains a local model based on local preprocessed features, uploads the model parameters to the federated learning server to aggregate and obtain a global model, and the digital twin synchronously maps the unified data output by the global model; S4: Risk prediction, the deep learning time series prediction module takes fused data as input and outputs potential risk warning information and handling suggestions; S5: Dynamic scheduling, the reinforcement learning scheduling module dynamically generates equipment-personnel scheduling plans based on real-time mapping data and early warning information, and issues them for execution; S6: Emergency Response Linkage. When an emergency threshold is triggered, the blockchain emergency response linkage layer synchronizes the status of each node, the digital twin simulates the optimal handling plan, and triggers a multi-node linkage response through smart contracts.
[0013] Specifically, in step S2, outlier removal uses an algorithm based on isolated forests, and feature extraction uses a combination of principal component analysis and local linear embedding.
[0014] The beneficial effects of this invention are: Improving the efficiency of multi-source data fusion while ensuring data privacy and security: Edge computing layers perform format standardization, outlier removal, and feature extraction preprocessing on collected heterogeneous multi-source data, significantly reducing the pressure on subsequent data transmission and centralized processing. Combined with a federated learning architecture, each edge node only uploads trained model parameters, not the original data, which remains in a local encrypted database, ensuring data stays within its domain while models are jointly trained. This model overcomes the efficiency bottlenecks of traditional centralized fusion architectures and avoids the risks of trade secret leakage (such as core equipment process parameters) and personnel privacy leakage (such as work trajectories) during cross-departmental data sharing at the underlying architecture level, balancing fusion efficiency and data security.
[0015] Early identification of potential risks and enhanced safety early warning capabilities: The deep learning time-series prediction module of the intelligent decision-making layer adopts an LSTM plus attention mechanism architecture, which can integrate historical operating data, real-time sensor data, and environmental correlation data. This overcomes the limitations of traditional fixed threshold-triggered warnings, not only identifying explicit risks that have reached the threshold, but also accurately capturing potential risks such as hidden equipment faults (e.g., initial bearing wear) and geological micro-deformations (e.g., slow deformation of surrounding rock in roadways), and outputting risk trends and targeted handling suggestions. Simultaneously, by strengthening the weight of key risk factors through the attention mechanism, the false alarm rate of early warnings is effectively reduced, providing more forward-looking and reliable support for mine safety decision-making.
[0016] Achieving dynamic collaborative scheduling of equipment and personnel to optimize resource allocation: The reinforcement learning scheduling module focuses on balancing operational efficiency and safety. It incorporates equipment location / load, personnel location / qualification, and task progress into the state space, and equipment path planning and personnel area allocation into the action space. By dynamically adjusting the scheduling scheme, it adapts to real-time changes in operating conditions. Compared to traditional preset fixed processes or manual ad-hoc planning, this module avoids problems such as equipment operating at full capacity and idle for extended periods, and personnel working on overlapping tasks with high-risk equipment. It achieves resource load balancing and personnel qualification matching, improving operational collaboration efficiency while reducing safety hazards.
[0017] Enhancing emergency response coordination and improving accident handling capabilities: The blockchain emergency response coordination layer relies on a practical Byzantine fault-tolerant consensus mechanism to ensure real-time synchronization and immutability of emergency status information across multiple nodes, including the work site, dispatch center, and rescue center. Combined with the accident scenario simulation capabilities of digital twins, it can quickly simulate accident development trends (such as gas diffusion and water seepage) and generate optimal handling solutions. Emergency commands are automatically triggered and executed through smart contracts, achieving a closed loop of information synchronization, scenario simulation, and multi-node coordinated response. This changes the passive situation of traditional emergency response, where a single module operates independently and information transmission is delayed or tampered with, significantly improving the timeliness and coordination of accident handling and maximizing personnel safety and equipment protection.
[0018] Promoting visualization and refinement of mine management and enhancing overall intelligence: The digital twin constructs a full-scene mirror image covering terrain, equipment, personnel, and environment, which is updated in real time with high frequency to reflect the actual operating status of the mine. This allows managers to intuitively grasp the dynamics of each link (such as equipment operating status, personnel distribution, and environmental parameter distribution) through a visual interface. From data collection, preprocessing, and fusion to risk prediction, dynamic scheduling, and emergency response, the system forms a full-process collaboration at all levels. This breaks through the limitations of data fragmentation, delayed decision-making, and disconnected links in traditional mine management, and promotes the intelligent transformation of mine management from extensive to refined and from passive response to proactive prediction, adapting to the intelligent transformation needs of the mining industry. Attached Figure Description
[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0020] Appendix Figure 1 This is a system architecture diagram of the present invention, which includes the hierarchical relationship and module composition of the data acquisition layer, edge computing layer, federated learning-digital twin fusion layer, intelligent decision-making layer, and blockchain emergency response layer; Appendix Figure 2 This is a schematic diagram of the digital twin structure, showing the relationship between the four major modules: terrain mirroring, equipment twin, personnel twin, and environment twin. Appendix Figure 3 The management method flowchart sequentially illustrates the steps of data acquisition, edge preprocessing, data fusion, risk prediction, dynamic scheduling, and emergency response. Appendix Figure 4 The emergency response logic diagram in Example 3 illustrates the collaborative relationship between blockchain node synchronization, digital twin simulation, and smart contract triggering. Detailed Implementation
[0021] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.
[0022] The specific implementation of this invention requires adjustment of system parameters based on actual conditions such as mine type (open-pit / underground), production capacity, and operational layout. Sensor network deployment density: every 150m in open-pit mines 2 Deploy one micro-vibration / stress sensor every 80m in underground mines 2 Deploy 1; Number of federated learning nodes: determined by the number of job partitions, with at least one edge node per partition; Deep learning model parameters: training data volume of no less than 6 months of historical data, number of iterations ≥ 300 rounds; Blockchain nodes: covering all key operational areas and management departments, with a total of ≥10 nodes.
[0023] Recommended system hardware configuration: The edge computing gateway uses an industrial-grade CPU (clock speed ≥ 2.0 GHz), the federated learning server GPU has ≥ 24 GB of video memory, and the UWB positioning base station has a transmission distance ≥ 50 m.
[0024] The software uses the Linux operating system, supports Python / C++ development languages, and uses an encrypted database (AES-256 encryption) for data storage.
[0025] like Figures 1-2 As shown, the mine management system based on a mine management platform according to the present invention includes: Data Acquisition Layer: Constructing a comprehensive data acquisition network combining sensing and high-precision positioning. This includes a micro-vibration / stress sensing network deployed at key locations such as the mining face, surrounding rock of roadways, and equipment frames to detect geological micro-deformations and changes in equipment structural stress; equipment operation sensors collecting parameters such as rotational speed, temperature, and load from equipment like mining machines, conveyors, and ventilation fans; environmental monitoring sensors monitoring environmental indicators such as methane concentration, dust content, humidity, and temperature; and a UWB high-precision positioning module enabling personnel to wear positioning tags and equipment to install positioning terminals, achieving 10cm-level high-precision positioning to adapt to complex underground environments. Edge computing layer: Edge computing gateways are deployed in operational zones such as mining areas, transportation areas, and ventilation areas. Each gateway connects to the data acquisition equipment in its corresponding zone. The edge computing gateways have data preprocessing capabilities, converting heterogeneous data into a unified format through format standardization, using the isolated forest algorithm to remove outlier data, and extracting key features through a combination of principal component analysis and local linear embedding, reducing data transmission volume and alleviating the pressure of subsequent data fusion. Federated Learning-Digital Twin Fusion Layer: This layer is the core of data fusion, employing a dual-core design of privacy protection and real-time mapping. The federated learning server is deployed in the mine control center. Each edge node trains a local model based on pre-processed feature data, uploading only the model parameters to the server. The server then uses weighted average aggregation to obtain the global fusion model, which is then distributed to each edge node for updates. This ensures data remains within its domain and models are trained together, preventing the leakage of original data. The digital twin constructs a full-scene mirror based on the mine's topography, equipment layout, and personnel distribution. This includes a terrain mirror module (restoring the mine's geological structure), an equipment twin module (mapping equipment operating status), a personnel twin module (displaying personnel location and work status in real time), and an environment twin module (simulating environmental parameter distribution). By receiving unified data output from the global fusion model, it updates the mirror state at a frequency of no less than 10Hz, achieving data visualization mapping.
[0026] The intelligent decision-making layer comprises two core modules: risk prediction and dynamic scheduling. The deep learning time-series prediction module employs an LSTM architecture with an attention mechanism. Input features include three months of historical equipment operation data, geological monitoring data, environmental data, and real-time fused data. The attention mechanism strengthens the weight of key risk factors (such as changes in equipment bearing temperature and sudden changes in surrounding rock stress), outputting potential risk information such as equipment failure probability and geological deformation trends for the next 24 hours. When the risk probability exceeds 30%, an early warning is triggered, and targeted handling suggestions (such as equipment maintenance priorities and roadway reinforcement plans) are pushed. The reinforcement learning scheduling module aims for optimal operational efficiency and the highest safety factor. Its state space includes 12 dimensions of features such as equipment location, load rate, personnel location, qualification level, and task progress. The action space includes eight types of actions such as equipment path planning, personnel work area allocation, and task priority adjustment. The reward function is set as α × (tons of coal / hour) + β × (1 - violation probability) (α = 0.6, β = 0.4). Adaptive scheduling schemes are generated through continuous iterative training. The blockchain emergency response layer consists of a distributed network comprising work area nodes, dispatch center nodes, rescue center nodes, equipment control nodes, and personnel management nodes, employing a practical Byzantine fault-tolerant consensus mechanism (fault tolerance of 1 / 3 node failure). Data from each node is stored using asymmetric encryption, and emergency status information (such as accident type, personnel location, and equipment status) is synchronized in real-time and cannot be tampered with. When a risk warning reaches an emergency threshold (such as gas concentration exceeding 1.0% or water permeability exceeding 5m³ / h), the system will respond accordingly. 3 When / h), the smart contract automatically triggers the emergency response process. The digital twin builds an accident scenario simulation model based on real-time synchronized data, simulates the accident development trend (such as the gas diffusion range and the rate of water inundation), deduces the optimal disposal plan (such as personnel evacuation routes, equipment shutdown sequence, and rescue channel planning), and issues execution instructions to each node through the blockchain network to achieve multi-node linkage response.
[0027] like Figure 3 As shown, the mine management method based on a mine management platform according to the present invention includes the following steps: Data Acquisition (S1): Various sensors and UWB positioning modules in the data acquisition layer collect data in real time. The acquisition frequency is adjusted according to the data type: equipment operating parameters are collected once every 1 second, environmental data is collected once every 5 seconds, geological micro-vibration / stress data is collected once every 0.1 seconds, and personnel positioning data is collected once every 0.5 seconds. The collected data is transmitted to the edge computing gateway through a wireless sensor network (leakage coaxial cable is used for auxiliary transmission in underground mines). Edge Preprocessing (S2): After receiving the collected data, the edge computing gateway first converts heterogeneous data (such as JSON data from devices and CSV data from sensors) into a unified format through XML format standardization; it then uses the isolated forest algorithm to identify and remove abnormal data (such as extreme data caused by sensor failures), with an abnormal data removal accuracy of ≥98%; it extracts the top 5 principal components (cumulative variance contribution rate ≥90%) from the device operation data through principal component analysis, and extracts low-dimensional features of geological data through local linear embedding, reducing the data dimension to 30% of the original dimension. Data Fusion (S3): Each edge node trains a local fusion model using a random forest algorithm based on its preprocessed feature data, with 100 training iterations. The edge nodes encrypt and upload the trained model parameters (weight matrix, bias vector) to the federated learning server. The server uses a weighted average method (weights are positively correlated with the amount of data at the edge nodes) to aggregate all local model parameters and generate a global fusion model. The global model is distributed to each edge node to update the local model and output fusion data in a unified format. The digital twin receives the fusion data in real time, updates the mirrored status of terrain, equipment, personnel, and environment, and generates a visual operation interface. Risk Prediction (S4): The deep learning time series prediction module loads the trained LSTM plus attention mechanism model, inputs historical data for 3 months and real-time data from the fused data, and calculates and outputs equipment failure probability curve, geological deformation trend curve and environmental risk level through the model; when the equipment failure probability is ≥30%, the geological deformation rate is ≥0.5mm / h or the environmental risk level reaches orange or above, the system triggers an audible and visual warning, marks the risk location on the digital twin interface, and pushes disposal suggestions (such as "High risk of wear on the mining machine bearing, it is recommended to stop the machine for inspection within 2 hours" and "Deformation of the surrounding rock in No. 3 roadway has intensified, it is recommended to reinforce it immediately"). Dynamic Scheduling (S5): The reinforcement learning scheduling module reads the mirror data and risk warning information of the digital twin in real time. Based on the current state space (e.g., "Mining machine A is located at working face 2, with a load rate of 80%; transport vehicle B is located in roadway 3, which is idle; personnel C has high-risk operation qualifications and is located in the dispatch room"), it generates a scheduling plan through reinforcement learning algorithm (e.g., "Instruct transport vehicle B to go to working face 2 for connection, and personnel C to go to roadway 3 to assist in reinforcement"). The scheduling plan is sent to the equipment control terminal and personnel positioning tags, and the execution objects are informed through voice prompts and screen displays. The system monitors the execution status of the plan in real time. If the working conditions change (e.g., the transport road is blocked), the scheduling plan is recalculated and updated, with an update response time of ≤10 seconds. Emergency Response Coordination (S6): When the risk warning reaches the emergency threshold, the smart contract of the blockchain emergency response layer is automatically triggered, and each node synchronously uploads real-time status data (such as personnel location, equipment operating status, and accident area environmental data). The digital twin builds an accident simulation model based on the synchronous data, sets the simulation time step to 1 minute, and simulates the accident development trend within the next 30 minutes. The optimal handling plan is generated based on the simulation results. For example, in the case of a water inrush accident, the power supply to the accident area is cut off first, the shortest evacuation route is planned (avoiding the waterlogged area), and rescue vehicles are dispatched to the nearest entrance. The handling plan is distributed to each node through the blockchain network, the equipment automatically executes shutdown and power-off commands, the personnel positioning tags issue voice evacuation prompts, and the rescue center receives personnel location and accident scene information and activates the rescue plan. During the emergency, the system updates the simulation model every 5 minutes and dynamically adjusts the handling plan until the accident is under control.
[0028] Example 1: Application of Multi-Source Data Fusion and Privacy Protection in Open-Pit Coal Mines This embodiment is applied to a large open-pit coal mine (annual production capacity of 15 million tons). The existing management and control platform of this coal mine suffers from problems such as high data fusion latency (average latency of 3 minutes) and high risk of privacy leakage during cross-departmental data sharing. The system and method of this invention are used for modification, and the specific implementation details are as follows: System Deployment: The data acquisition layer deploys 300 micro-vibration sensors (distributed at the mining face and spoil heap slopes), 500 equipment operation sensors (installed on 10 mining machines, 20 transport vehicles, and 5 loaders), 200 environmental monitoring sensors (monitoring gas, dust, and temperature), and 300 UWB positioning tags (distributed to workers); the edge computing layer deploys 10 edge computing gateways, corresponding to 5 mining areas, 3 transport areas, and 2 spoil heap areas; the federated learning server is deployed at the coal mine control center, configured with a GPU server (32GB of video memory); the digital twin is constructed based on coal mine terrain data (accuracy 1:500), equipment 3D models, and personnel work area planning, with a mapping update frequency set to 15Hz; the blockchain emergency linkage layer includes 15 nodes: mining area nodes, transport area nodes, dispatch center nodes, rescue center nodes, and equipment control nodes, adopting a practical Byzantine fault-tolerant consensus mechanism. Data Acquisition and Preprocessing: Equipment operating parameters (such as excavator speed and transport vehicle load) are collected every 1 second, environmental data every 5 seconds, geological micro-vibration data every 0.1 seconds, and personnel positioning data every 0.5 seconds. The edge computing gateway standardizes the format of the collected data (converts it to JSON format) and uses the isolated forest algorithm to remove abnormal data (such as abnormal data where the transport vehicle speed is 0 but the load rate is 100%), achieving an abnormal data removal accuracy of 99.2%. Principal component analysis is used to extract the top 5 principal components of the equipment operating data (cumulative variance contribution rate of 91.5%), and local linear embedding is used to extract low-dimensional features of the geological micro-vibration data, reducing the data dimension from the original 12 dimensions to 4 dimensions. Federated Learning Fusion and Digital Twin Mapping: Each edge node trains a local random forest model based on local preprocessed features, undergoing 100 training iterations with an accuracy of 88%. Edge nodes upload model parameters (weight matrix size 500KB / node) to the federated learning server via an HTTPS encrypted channel. The server aggregates the parameters using a weighted average method (mining area node weight 0.4, transportation area node weight 0.3, spoil heap node weight 0.3), generating a global fusion model with an aggregation time ≤3 seconds. The global model is distributed to each edge node, outputting fusion data in a unified format (including equipment operating status scores, geological stability levels, and personnel work status). The digital twin receives the fusion data in real time, updating the working position of the mining machine, the driving trajectory of the transportation vehicle, the deformation of the spoil heap slope, and the personnel distribution area. A visual mirror image is displayed on the large screen in the control center, allowing managers to view detailed data for any equipment / personnel by clicking with the mouse. Application Results: After three months of operation, the data fusion latency of this embodiment was reduced from 3 minutes to 0.8 seconds, and the fusion efficiency was improved by 99.6%. During cross-departmental data sharing, no leakage of original data occurred (verified by a third-party privacy testing agency). The mapping error of the digital twin was ≤2cm, which can accurately reflect the real-time status of the mine. The matching degree between the equipment operation status score and the actual maintenance results reached 95%, providing effective support for the formulation of maintenance plans. Example 2: Application of Prediction and Early Warning of Potential Risks in Underground Metal Mines This embodiment is applied to an underground copper mine (mining depth 500m). The existing management platform of this mine can only provide early warnings of explicit risks based on fixed thresholds. Two production stoppages caused by latent equipment failures and one near-miss incident of roadway rock collapse have occurred. The system and method of this invention enhance the risk prediction and early warning capabilities. Specific implementation details are as follows: System Deployment: The data acquisition layer focuses on deploying 400 stress sensors (installed in the surrounding rock of roadways and on the frames of mining equipment), 250 micro-vibration sensors (distributed in the mining face and around the goaf), 150 environmental monitoring sensors (monitoring gas, humidity, and ground temperature), and 200 sets of UWB positioning modules (100 sets for personnel and 100 sets for equipment); the edge computing layer deploys 8 edge computing gateways, covering 4 mining areas, 2 transport roadways, and 2 ventilation systems; the deep learning time series prediction module of the intelligent decision-making layer adopts an LSTM plus attention mechanism architecture, trained based on the TensorFlow framework, with training data consisting of equipment operation data (1 million records), geological monitoring data (800,000 records), and environmental data (500,000 records) from the past year; the digital twin focuses on constructing roadway surrounding rock mirrors and equipment twin modules, with a mapping update frequency of 20Hz.
[0029] Model Training and Parameter Settings: The input features of the deep learning time series prediction model include seven categories of features: equipment bearing temperature, rotational speed, load, surrounding rock stress, micro-vibration frequency, ambient humidity, and ground temperature. The outputs are the equipment failure probability (0-100%) and the surrounding rock deformation trend (mm / 24h). During model training, the ratio of training set, validation set, and test set is 7:2:1, with 500 iterations and a learning rate of 0.001. The final test set accuracy reached 93%. The attention mechanism showed the highest attention to changes in equipment bearing temperature (weight 0.3) and sudden changes in surrounding rock stress (weight 0.25), which can enhance the identification of key risk factors.
[0030] Risk prediction and early warning implementation: The system collects data in real time and, after preprocessing and fusion, inputs it into a deep learning time series prediction model; the model outputs risk prediction results for the next 24 hours. When the probability of equipment failure is ≥30% or the trend of surrounding rock deformation is ≥0.5mm / 24h, an audible and visual early warning is triggered, and the risk location is marked in red on the digital twin interface; for equipment risks, maintenance suggestions are pushed (e.g., "The probability of bearing failure of the mining machine in mining area No. 3 is 42%, it is recommended to stop the machine for inspection within 1.5 hours"); for geological risks, reinforcement plans are pushed (e.g., "The trend of surrounding rock deformation in roadway No. 5 is 0.8mm / 24h, it is recommended to use anchor bolt reinforcement with a reinforcement range of 10m"). Application Results: This implementation model has been running for 6 months and has successfully predicted 12 latent equipment failures, with an average early warning time of 9.2 hours, avoiding production stoppage losses of approximately 8 million yuan; it has predicted the risk of micro-deformation of the surrounding rock in the roadway 8 times, and through timely reinforcement, no collapse accidents have occurred, reducing the risk incidence rate by 100% compared to before implementation; the false alarm rate of early warnings has been reduced from 25% to 8%, and the decision-making efficiency of management personnel has been improved by 60%. Example 3: Application of Dynamic Dispatch and Emergency Response in Underground Coal Mines (as attached) Figure 4 (As shown) This embodiment is applied to an underground coal mine (annual production capacity of 10 million tons, total length of underground roadways of 80km). The existing management and control platform of this coal mine adopts a static scheduling scheme, which results in low operational efficiency and poor coordination among departments during emergency response. The system and method of this invention are used to achieve dynamic scheduling and emergency linkage. Specific implementation details are as follows: System Deployment: The UWB high-precision positioning module of the data acquisition layer is deployed in various underground roadways and working faces, with a total of 50 positioning base stations and a positioning accuracy of 8cm. The reinforcement learning scheduling module is deployed on the control center server and is developed using the PyTorch framework. The state space includes 12-dimensional features such as equipment location, load rate, personnel location, qualification level, work task progress, and roadway congestion. The action space includes 8 types of actions such as equipment path planning, personnel area allocation, and task priority adjustment. The blockchain emergency linkage layer includes 20 nodes (8 working face nodes, 4 transportation nodes, 3 ventilation nodes, 3 dispatch center nodes, and 2 rescue center nodes), adopts a practical Byzantine fault-tolerant consensus mechanism, and the smart contract pre-sets emergency procedures for three types of accidents: water inrush, fire, and gas explosion. Dynamic scheduling implementation: The reward function of the reinforcement learning scheduling module is set to 0.6 × (tons of coal / hour) + 0.4 × (1 - probability of violation). The model is optimized by combining offline training (based on scheduling data from the past 6 months) with online iteration (real-time operating data). During real-time scheduling, the module reads the mirror data of the digital twin (equipment operating status, personnel location, and roadway conditions) every 10 seconds to generate the optimal scheduling plan. For example, when the load rate of the mining machine on the No. 2 working face reaches 90%, while the load rate of the mining machine on the No. 3 working face is only 60%, and the transport vehicles are congested near the No. 2 working face, the system automatically adjusts the scheduling plan: instructs the mining machine on the No. 3 working face to increase the mining intensity, schedules 2 idle transport vehicles to transport from the No. 3 working face, and optimizes the transport route to avoid congested roadways. Emergency response test: Simulating a water inrush accident in underground roadway No. 3 (water inrush flow rate 6m³ / h) 3After triggering the emergency threshold, each node in the blockchain emergency response layer synchronizes data in real time (personnel location in Lane 3: 12 people; equipment status: 3 transport vehicles in operation; environmental data: methane concentration 0.9%, water level rise rate 0.5 m / min); a digital twin is constructed to create a permeability simulation model, simulating the water level rise trend over the next 30 minutes (the lower half of Lane 3 will be submerged in 15 minutes, and the entire lane will be submerged in 25 minutes); based on the simulation results, a response plan is generated: immediately cut off the power supply to Lane 3, and plan an evacuation route (evacuate from the safety exit on the east side of Lane 3, avoiding...). (Open the flooded area), dispatch two rescue vehicles to the east exit, and notify the ventilation system to increase airflow to dilute the gas; the smart contract automatically triggers the execution instructions of each node, the personnel positioning tags issue voice evacuation prompts, the transport vehicles automatically stop and avoid the evacuation channel, and the rescue center receives personnel location and accident scene information in real time; the emergency response time (from the triggering of the accident to the start of personnel evacuation) is only 8 seconds, which is 80% shorter than the original emergency system (response time 40 seconds); through the evacuation route optimized by digital twin simulation, all 12 personnel were safely evacuated, and the evacuation time was shortened by 30% compared with the original plan. Application Results: After four months of operation, this embodiment improved downhole operation efficiency from 85 tons / hour to 106 tons / hour, an increase of 24.7%; equipment load balancing rate increased from 72% to 91%; three simulated emergency drills (water inrush, fire, and gas explosion) were conducted, with an average emergency response time reduction of 55%, a 100% personnel evacuation success rate, and a 3-fold increase in multi-department collaborative response efficiency. Comparison Example Compare with Example 1: Existing centralized data fusion solutions The centralized data fusion architecture of a mainstream mine management platform is adopted. All raw data is directly uploaded to the cloud server, and fusion is achieved through data cleaning and format conversion. Applied to the open-pit coal mine in Example 1, the test results show that the average data fusion latency is 3.2 minutes, 240 times higher than that of the present invention (0.8 seconds); when sharing data across departments, the probability of raw data being obtained by third parties reaches 40%, indicating insufficient privacy protection; the data loss rate due to network fluctuations during the data fusion process reaches 5%, while that of the present invention is only 0.1%. Compare with Example 2: Existing threshold early warning scheme The existing fixed threshold early warning system only monitors conventional parameters such as equipment temperature and gas concentration. An early warning is triggered when the equipment temperature exceeds 80℃ or the gas concentration exceeds 1.0%. Applied to the underground metal mine in Example 2, the test results showed that it could not identify potential risks such as early-stage bearing wear (temperature 65-75℃) and micro-deformation of the surrounding rock (0.1-0.4mm / 24h). Within 6 months, there were 4 equipment failures and 2 near-collapse incidents. The false alarm rate was 28%, 250% higher than that of this invention (8%). It lacked risk trend prediction capabilities, preventing managers from developing proactive response plans. Compare with Example 3: Existing static scheduling and a single emergency response plan A pre-set fixed scheduling scheme is adopted (such as the mining machine and transport vehicle coordinating along a fixed route). In case of emergency response, all departments are notified by telephone for coordination. Applied to the underground coal mine of Example 3, the test results are as follows: the average operating efficiency is 82 tons / hour, which is 22.6% lower than that of the present invention (106 tons / hour); the equipment load balance rate is 68%, and the probability of personnel violating operating rules is 15%; in the case of simulated water inrush accident, the emergency response time is 45 seconds, and the personnel evacuation time is 35% longer than that of the present invention. Due to the information transmission delay, two people were simulated to be trapped, and the efficiency of coordinated handling was low.
[0031] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A mine management system based on a mine management platform, characterized in that, include: The data acquisition layer includes a micro-vibration / stress sensing network, equipment operation sensors, environmental monitoring sensors, and a UWB high-precision positioning module, used to collect multi-source heterogeneous data on mining equipment, geology, environment, and personnel. The edge computing layer contains several edge computing gateways deployed in the mining operation area, which are used for preprocessing of collected data, including format conversion, outlier removal, and feature extraction. The Federated Learning-Digital Twin Fusion Layer includes a Federated Learning Server and a Digital Twin. The Federated Learning Server receives preprocessed feature data from each edge node, trains a global fusion model through horizontal federated learning, and distributes updates. The Digital Twin is built based on the actual mining scenario and maps the unified data output by the global fusion model in real time. The intelligent decision-making layer includes a deep learning time series prediction module and a reinforcement learning scheduling module. The time series prediction module predicts potential risks such as equipment failure and geological deformation based on fused data, while the reinforcement learning scheduling module dynamically adjusts the equipment-personnel scheduling scheme with the goal of improving work efficiency and safety. The blockchain emergency response layer includes multiple nodes such as the work site, dispatch center, and rescue center. It adopts a practical Byzantine fault-tolerant consensus mechanism to synchronize emergency status information and link digital twins to complete emergency simulations.
2. The mine management system based on a mine management platform according to claim 1, characterized in that: The federated learning server supports dynamic node access. Each edge node only uploads model parameters and does not upload raw data. The raw data is stored in a local encrypted database.
3. A mine management system based on a mine management platform according to claim 1, characterized in that: The digital twin includes a terrain mirroring module, an equipment twin module, a personnel twin module, and an environment twin module, with a mapping update frequency of no less than 10Hz.
4. A mine management system based on a mine management platform according to claim 1, characterized in that: The deep learning time series prediction module adopts an LSTM plus attention mechanism architecture. The input features include historical running data, real-time sensing data and environmental correlation data, and the output is risk probability and trend curve.
5. A mine management system based on a mine management platform according to claim 1, characterized in that: The state space of the reinforcement learning scheduling module includes equipment location / load, personnel location / qualification, and task progress; the action space includes equipment path planning and personnel area allocation; and the reward function is α×work efficiency + β×safety coefficient, where α and β are weight coefficients, and α+β=1.
6. A mine management system based on a mine management platform according to claim 1, characterized in that: The positioning accuracy of the UWB high-precision positioning module is no less than 10cm, and the positioning signal penetration depth is adapted to underground mining scenarios up to 500m.
7. A mine management system based on a mine management platform according to claim 1, characterized in that: The node data in the blockchain emergency response layer is stored using asymmetric encryption, and emergency commands are automatically triggered and executed through smart contracts.
8. A mine management system based on a mine management platform according to claim 1, characterized in that: The sensor deployment density of the micro-vibration / stress sensing network is per 100m 2 At least one, with a detection accuracy of 0.01mm micro-deformation and 0.1MPa stress change.
9. A mine management method based on a mine management platform, wherein the method is implemented using the mine management system described in any one of claims 1-8, characterized in that, Includes the following steps: S1: Data Acquisition, which collects equipment operating parameters, geological micro-vibration / stress data, environmental monitoring data and personnel UWB positioning data in real time through the data acquisition layer; S2: Edge preprocessing, the edge computing gateway performs format standardization, outlier removal and feature extraction on the collected data; S3: Data fusion, each edge node trains a local model based on local preprocessed features, uploads the model parameters to the federated learning server to aggregate and obtain a global model, and the digital twin synchronously maps the unified data output by the global model; S4: Risk prediction, the deep learning time series prediction module takes fused data as input and outputs potential risk warning information and handling suggestions; S5: Dynamic scheduling, the reinforcement learning scheduling module dynamically generates equipment-personnel scheduling plans based on real-time mapping data and early warning information, and issues them for execution; S6: Emergency Response Linkage. When an emergency threshold is triggered, the blockchain emergency response linkage layer synchronizes the status of each node, the digital twin simulates the optimal handling plan, and triggers a multi-node linkage response through smart contracts.
10. A mine management method based on a mine management platform according to claim 9, characterized in that: In step S2, outlier removal uses an algorithm based on isolated forest, and feature extraction uses a combination of principal component analysis and local linear embedding.