Mine production and operation intelligent decision-making system based on multi-source heterogeneous data fusion

Abstract

The invention relates to the technical field of mine production operation management, in particular to a mine production operation intelligent decision-making system based on multi-source heterogeneous data fusion. A data preprocessing module; a data fusion modeling module; an intelligent decision module; and a decision execution feedback module. In the scheme, an edge computing terminal is deployed through a data acquisition unit, a multimode communication technology is integrated, data of an underground sensor, electromechanical equipment and the like are accessed in real time, and the data transmission reliability is ensured through a communication quality evaluation formula; the business system interconnection unit maps heterogeneous fields of systems such as ERP, WMS and the like into global unified business keys by means of semantic middleware, and semantic ambiguity is eliminated. The data preprocessing module integrates the dispersed data into a standardized data set through intelligent cleaning, format conversion and three-dimensional association, and provides data support for decision making.

Description

Technical Field

[0001] This invention relates to the field of mine production and operation management technology, and in particular to an intelligent decision-making system for mine production and operation based on the fusion of multi-source heterogeneous data. Background Technology

[0002] Mine production and operation management involves a complex underground working environment and multi-stage collaboration. With the application of intelligent technologies, a data acquisition system has gradually been formed, centered on IoT devices, business management systems, and external environmental monitoring. Among them, IoT devices collect real-time operating data such as gas concentration and equipment vibration, while ERP and MES systems record business information such as production plans and material inventory. External data covers environmental factors such as market prices and weather warnings. These data together constitute the basic information source for mine management.

[0003] However, in existing management models, multi-source data exhibits significant heterogeneity and dispersion: sensor data is mostly time-series signals, business system data is primarily structured tables, and external data often contains unstructured text; these format differences hinder data interoperability. Furthermore, data is stored in independent systems, creating "information silos" that prevent the linkage analysis of equipment status and production plans. In addition, decision-making relies on human experience or single models, resulting in delayed responses to real-time risks and a lack of a closed-loop mechanism from data acquisition to execution feedback, leading to low production efficiency and insufficient safety risk management. Therefore, we propose an intelligent decision-making system for mine production and operation based on the fusion of multi-source heterogeneous data. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent decision-making system for mine production and operation based on the fusion of multi-source heterogeneous data, thereby solving the technical problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A mine production and operation intelligent decision-making system based on multi-source heterogeneous data fusion includes:

[0007] Data acquisition and access module: used to collect multi-source data from the entire mine area, including underground sensor data, electromechanical equipment data, personnel positioning data, and business management system data;

[0008] Data preprocessing module: Cleans, transforms, and correlates the collected multi-source data to ensure data integrity and format consistency;

[0009] Data fusion modeling module: Utilizes graph neural networks and transfer learning techniques to fuse multimodal data to build intelligent models for safety assessment, equipment prediction, etc., and supports model self-training and dynamic optimization;

[0010] Intelligent Decision Module: Based on the real-time status indicators output by the fusion model, combined with constraint programming, reinforcement learning and cost-sensitive algorithms, it generates multi-objective decision-making schemes covering production scheduling, safety control and cost optimization;

[0011] Decision execution feedback module: Transforms decision-making schemes into equipment control instructions and personnel task lists, and provides feedback to the optimization model.

[0012] In one possible implementation, the data acquisition and access module includes:

[0013] Full-domain data acquisition unit: Through edge computing and protocol adaptation technology, it collects multi-source data from downhole sensors, electromechanical equipment, and personnel positioning in real time, adds spatiotemporal tags, and evaluates communication quality (CQ) to ensure the reliability of data transmission;

[0014] Business System Interconnection Unit: Develops semantic middleware, defines unified business keys, achieves standardized extraction and real-time synchronization of system data, and performs integrity verification of business data; the formula for business data integrity verification is: Among them, F 有效 F represents the number of required fields with no missing fields. 总 This represents the total number of required fields.

[0015] When the data communication quality assessment result CQ collected by the global data acquisition unit is ≥0.6, the business system interconnection unit will perform standardized extraction and real-time synchronization of the collected data;

[0016] When the data communication quality assessment result CQ collected by the global data acquisition unit is <0.4 and lasts for 1 minute, relay transmission is triggered. If it is not restored within 3 minutes, the data is sent to the data preprocessing module for missing value repair.

[0017] In one possible implementation, the data preprocessing module includes:

[0018] Intelligent cleaning unit: Utilizing physical laws and statistical methods, it identifies and repairs missing and outlier values ​​in the data, improving data quality and verifying data integrity; the data integrity verification formula is: Among them, D 有效 For valid data, D 总 Total data;

[0019] Format conversion unit: unifies the time format, unit system and encoding rules of multi-source data, normalizes numerical data, and establishes a data dictionary mapping table;

[0020] Data Association Unit: This unit associates multi-source data using spatiotemporal coordinates and business keys, constructs an association network, and calculates the association success rate. The formula for calculating the data association success rate is: Among them, K 有效 To associate valid data, K 总 This represents the total number of associations to be determined.

[0021] When the data integrity DI is less than 90%, a manual verification process is triggered. The data administrator corrects the data through a visual interface. If "suspicious data" is found, the data is returned to the intelligent cleaning unit for secondary detection; otherwise, it enters the format conversion unit.

[0022] In one possible implementation, the data fusion modeling module includes:

[0023] Feature fusion unit: Converts data into a unified feature vector, extracts spatial features using a graph convolutional network, extracts temporal features using an LSTM, and fuses them through a gating mechanism to generate a fused feature vector and calculate the feature discrimination index; the feature fusion formula is: h = σ(W g h g +W t h t ), where σ is the Sigmoid function, W g W t The weight matrix is ​​a learnable weight matrix;

[0024] Model training unit: Selects the optimal algorithm through automated machine learning, supports self-training of multiple models, and uses Bayesian optimization of hyperparameters to evaluate and optimize the model;

[0025] Real-time assessment unit: Calculates core indicators in real time, outputs early warning signals, and triggers corresponding decision-making units based on indicator values; the formula for calculating the safety status index is: Where w i x was determined by AHP. i ′ represents a standardized feature;

[0026] If SSI < 70 triggers a yellow alert, the safety risk decision-making unit will be activated.

[0027] In one possible implementation, the intelligent decision-making module includes:

[0028] Production scheduling optimization unit: Based on the constraints of comprehensive equipment status, material inventory, and personnel allocation, it dynamically generates the optimal mining operation plan and resource scheduling strategy, constructs a mixed integer programming model, designs a scheduling agent, learns the long-term optimal strategy through a deep Q-network, and continuously optimizes the scheduling scheme every 2 hours to adapt to dynamic events;

[0029] Security Risk Decision Unit: Based on real-time security status assessment results, a multi-level risk decision system is constructed, and handling strategies are automatically generated. The compliance and effectiveness of the strategies are verified through a security rule engine.

[0030] Cost-benefit optimization unit: Identifies cost optimization opportunities in equipment operation and maintenance, material procurement, and production organization; evaluates the input-output ratio of decision-making schemes using a cost-sensitive model; provides optimal economic improvement suggestions; identifies cost drivers using activity-based costing; constructs a cost calculation model; and utilizes Monte Carlo simulation technology to output the expected cost distribution. It compares actual costs with budgeted costs in real time and initiates cost attribution analysis and cost anomaly auditing processes. The cost calculation model for the cost-benefit optimization unit is: C 总 =C 设备 +C 物资 +C 人力 +C 能耗 +C 风险 ,in, R j The risk level is determined by the loss coefficient: Level 1 = 1, Level 2 = 5, Level 3 = 20.

[0031] In one possible implementation, the decision execution feedback module includes:

[0032] Command issuing unit: converts scheduling parameters into control signals that the device can recognize, supports anti-misoperation verification, performs protocol conversion and task push, and corrects commands that exceed the device's range;

[0033] The performance feedback unit compares the decision-making objectives with the actual results, quantifies the deviation and optimizes the model, calculates the achievement rate of the decision-making objectives and the cost saving rate, outputs an evaluation report, and adds the deviation data to the training set to trigger incremental training of the model; the formula for calculating the achievement rate of the decision-making objectives is: Among them, Q 实际 Q represents the actual performance indicator achieved. 计划 For a pre-set goal;

[0034] The formula for calculating the cost savings rate is: Among them, C 预算 For the pre-set cost budget, C 实际 The actual costs incurred.

[0035] In one possible implementation, in the business system interconnection unit, when the business data integrity verification result CI < 95%, a supplementary data entry reminder is pushed through WeChat. If the data entry is not completed within 2 hours, the same type of planned data from the previous 24 hours is automatically used, and the logical error correction secondary verification of the data preprocessing module is triggered.

[0036] In one possible implementation, in the data association unit, when the data association success rate AR < 95%, the time window is automatically extended and the association is re-established; if the failure is caused by an incorrect business number, the data is returned to the business system interconnection unit for correction, and qualified data enters the data fusion modeling module.

[0037] In one possible implementation, in the feature fusion unit, when the feature discriminative degree D I When the value is less than 1.5, the weight of geographical features is automatically increased, and the model enters the training unit after reaching the target.

[0038] Beneficial effects compared to existing technologies:

[0039] 1. In this solution, edge computing terminals are deployed through the data acquisition unit, integrating multi-mode communication technology to access data from downhole sensors, electromechanical equipment, etc. in real time, and ensuring data transmission reliability through a communication quality assessment formula; the business system interconnection unit uses semantic middleware to map heterogeneous fields from systems such as ERP and WMS to globally unified business keys, eliminating semantic ambiguity. The data preprocessing module integrates scattered data into a standardized dataset through intelligent cleaning, format conversion, and three-dimensional correlation, providing data support for decision-making;

[0040] 2. In this solution, the feature fusion unit employs graph convolutional networks and LSTM gating mechanisms to fuse sensor data, geographic information, and business rules into a 256-dimensional feature vector, capturing deep data correlations. The model training unit utilizes automated machine learning technology, combined with Bayesian optimization and cross-algorithm switching, to construct equipment fault prediction and safety status assessment models. In the intelligent decision-making module, the production scheduling optimization unit dynamically balances output and energy consumption through mixed integer programming and reinforcement learning agents, continuously optimizing the plan every 2 hours to adapt to dynamic scenarios such as equipment failures. The safety risk decision-making unit, based on a three-level risk level model and a standardized handling process library, achieves second-level response to risks such as gas over-limit, significantly improving the timeliness and accuracy of decision-making. Attached Figure Description

[0041] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0042] Figure 1 This is a schematic diagram of the intelligent decision-making system framework for mine production and operation of the present invention; Detailed Implementation

[0043] Preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention can also be implemented in various different forms, and therefore the present invention is not limited to the embodiments described below. In addition, for the purpose of more clearly describing the present invention, parts not connected to the invention will be omitted from the drawings.

[0044] The technical solutions in this application are designed to address the problems described in the background, and are generally as follows:

[0045] Example:

[0046] This embodiment provides an intelligent decision-making system for mine production and operation that integrates data acquisition, processing, modeling, decision-making, and feedback, forming a closed-loop management system of "acquisition-processing-modeling-decision-execution". It includes a data acquisition and access module, a data preprocessing module, a data fusion and modeling module, an intelligent decision-making module, and a decision execution feedback module.

[0047] I. Data Acquisition and Access Module

[0048] This module constructs a mine-wide data acquisition network, enabling real-time access and preliminary verification of equipment, business, and environmental data through edge computing and protocol adaptation technologies. It includes a mine-wide data acquisition unit and a business system interconnection unit.

[0049] 1. Global Data Acquisition Unit

[0050] This unit collects multi-source data in real time from downhole sensors, electromechanical equipment, personnel positioning, etc., and solves the signal transmission problem in complex environments through a self-organizing network, enabling high-frequency perception of equipment status.

[0051] Deploy an edge computing terminal (ECU) that integrates a 5G / 4G / LoRa multi-mode communication module to collect device parameters (such as gas concentration C) at 500ms intervals. gas Vibration amplitude A vib The data is collected and labeled with spatiotemporal tags (longitude L, latitude B, and acquisition time t). Communication quality is evaluated using the following formula: Among them, S 阈值 Take -90dBm, N 有效 To correctly receive the number of data packets, N 总 The total number of packets sent is CQ. Communication is considered stable if CQ ≥ 0.6.

[0052] Generate raw data with verification flags. For example, device E01 collected a gas concentration of 0.8% LEL at 09:00, with CQ = 0.75 (communication is good).

[0053] If CQ < 0.4 and persists for 1 minute, trigger Mesh network relay transmission; if it does not recover after 3 minutes, send critical data (device ID, fault code) and mark it as "weak connection data" to enter the missing value repair of the data preprocessing module.

[0054] 2. Business System Interconnection Unit

[0055] This unit breaks down data barriers between management systems such as ERP and MES, and uses semantic mapping technology to achieve standardized extraction and real-time synchronization of business data.

[0056] Develop semantic middleware to define unified business keys such as "material ID" and "work order ID". For example, map the "material code" of ERP and the "inventory code" of WMS to a globally unique MID, and extract them periodically through ETL tools (full extraction at 2 o'clock every day, and real-time capture of incremental data).

[0057] The formula for verifying the integrity of business data is as follows: Among them, F 有效 F represents the number of required fields with no missing fields. 总 The total number of required fields is [number]. Required fields include "Plan ID", "Mining Area", and "Production Target".

[0058] Next, the business data will be used to generate a standardized business dataset T = {Plan ID, Region Code, Q}. 目标 For example, P20250703 plans to mine 5000 tons from shaft No. 3, with associated equipment E01-E03.

[0059] If CI < 95%, it is marked as abnormal business data, and a reminder to supplement the data is pushed through WeChat. If the data is not supplemented within 2 hours, the same type of plan data from the previous 24 hours will be used automatically. This data will be marked as temporary data, and the logical error correction secondary verification of the preprocessing module will be triggered.

[0060] II. Data Preprocessing Module

[0061] This module cleans, transforms, and correlates multi-source data, using dynamic thresholds to identify anomalous data and ensure the integrity and consistent format of the input data for modeling. It includes an intelligent cleaning unit, a format conversion unit, and a data correlation unit.

[0062] 1. Intelligent cleaning unit

[0063] This unit uses physical laws to verify and statistical methods to identify and repair missing and outlier values, thereby improving data quality.

[0064] Missing value repair: Linear interpolation is used for continuous data. For discrete data with a missing rate ≤10%, fill with the mode; otherwise, mark as “unknown”.

[0065] Outlier detection: Combining the 3σ principle with business logic, the outlier range is defined as [μ-3σ, μ+3σ]. For example, the current should be ≤5A when the hoist stops. If it exceeds this range, it is marked as "logic error".

[0066] The formula for verifying the data integrity score is as follows: Among them, D 有效 For valid data, D 总 This is the total data.

[0067] Generate and output data with cleaning labels. For example, if E01 temperature 120℃ is marked as an outlier (outside the 3σ range), it is recommended to calibrate the sensor.

[0068] If DI < 90%, a manual verification process is triggered, and the data administrator corrects it through a visual interface; if there is "suspicious data" (i.e., model confidence < 95%), it returns to the intelligent cleaning unit for secondary detection; otherwise, it enters the format conversion unit.

[0069] 2. Format Conversion Unit

[0070] This unit unifies the time format, unit system, and encoding rules for multi-source data, resolving the issue of field heterogeneity.

[0071] Establish a data dictionary mapping table, define the time format as ISO 8601, the unit conversion formula as T(K)=T(℃)+273.15, and the text status code mapping as [0: Stop, 1: Run, 2: Fault].

[0072] The formula for normalizing numerical data is as follows: Where, x min x is the minimum value of the data. max x is the maximum value of the data, and x′ is the normalized data.

[0073] Generate uniform format data D 标 ={t ISO ,x′,unit,equipment type},for example, if the vibration amplitude of E01 is 15.2mm / s, normalized to 0.68,unit remains mm / s.

[0074] If an undefined field (such as a new device parameter) is found, it is marked as "to be defined" and the dictionary update process is triggered; if it is not processed within 24 hours, the field will not be included in the modeling, and qualified data will enter the data association unit.

[0075] 3. Data Association Unit

[0076] This unit associates multi-source data with spatiotemporal coordinates and business keys to construct a three-dimensional association network of "device-work order-region".

[0077] Spatiotemporal correlation: Using UWB positioning accuracy to match the GIS map grid (10m×10m), the area code AREA(L,B) is generated.

[0078] Business association: Using the work order ID as the key, link equipment operation data with personnel attendance to form work order W20250702. Equipment E01 The relationships of personnel Zhang San.

[0079] The success rate of spatiotemporal correlation and business correlation is calculated using the following formula: Among them, K 有效 To associate valid data, K 总 This represents the total number of connections to be made.

[0080] Generate an associated dataset R = {work order ID, equipment list, personnel ID, AREA code}, for example, W20250702 associated with equipment E01 and area A01 of shaft 3.

[0081] If AR < 95%, automatically expand the time window (from 10 minutes to 15 minutes) and re-associate; if the failure is due to an incorrect business number, return to the business system interconnection unit for correction, and qualified data enters the data fusion modeling module.

[0082] III. Data Fusion Modeling Module

[0083] This module utilizes graph neural networks and transfer learning techniques to integrate multimodal data to construct intelligent models for safety assessment and device prediction, supporting model self-training and dynamic optimization. It includes a feature fusion unit, a model training unit, and a real-time evaluation unit.

[0084] 1. Feature fusion unit

[0085] This unit converts sensor values, business rules, geographic information, etc., into a unified feature vector to overcome the limitations of single-modal information.

[0086] Construct a device association graph (nodes represent devices, edges represent relationships such as "same region" and "same work order"), and use a graph convolutional network (GCN) to extract spatial features h. g LSTM extracts temporal features h t Fusion is achieved through a gating mechanism: h = σ(W) g h g +W t h t ), where σ is the Sigmoid function, W g W t This is a learnable weight matrix.

[0087] Generate a 256-dimensional fused feature vector H, with SHAP value sorting (e.g., "bearing temperature" weight 0.3, "vibration frequency" weight 0.25).

[0088] Calculate feature discrimination If D I If the value is less than 1.5, the weight of geographic features will be automatically increased (from 40% to 60%), and the model will be put into the training unit after the target is met.

[0089] 2. Model Training Unit

[0090] This unit uses automated machine learning to select the optimal algorithm and supports the self-training of models such as equipment failure prediction and production prediction.

[0091] Data partitioning: 70% training set, 20% validation set, and 10% test set. Bayesian optimization of hyperparameters (such as tree depth and learning rate in XGBoost) is employed.

[0092] Fault prediction model output probability: Where W is the weight vector, b is the bias term, and the output is a model evaluation report (accuracy, AUC value), for example, "Equipment failure model accuracy 88%, AUC 0.92".

[0093] If the validation set accuracy is less than 85%, the algorithm will be automatically switched (XGBoost → LightGBM). Manual intervention will be triggered after three consecutive failures, and qualified models will be deployed to the real-time evaluation unit.

[0094] 3. Real-time evaluation unit

[0095] This unit calculates key indicators such as Safety Status Index (SSI) and Equipment Health Index (HI) in real time and outputs early warning signals.

[0096] The formula for calculating the safety status index is: Where w i Determined by AHP, such as a gas concentration weight of 0.3, x i ′ represents a standardized feature.

[0097] The formula for calculating equipment health is: HI = 1 - Pr (fault).

[0098] The system calculates and outputs {SSI=85, HI=0.92, PE=450 tons / hour}, with an example of "SSI of Shaft No. 3 is 85 (safe), HI of Equipment E01 is 0.92 (healthy)".

[0099] If SSI < 70, a yellow alert is triggered, and the safety risk decision-making unit is activated; if HI < 0.8, an equipment maintenance work order is generated, and qualified data enters the intelligent decision-making module.

[0100] IV. Intelligent Decision-Making Module

[0101] This module, based on real-time status indicators output by the fusion model, combines constraint programming, reinforcement learning, and cost-sensitive algorithms to generate multi-objective decision-making schemes covering production scheduling, safety control, and cost optimization. As the core of the system, this module achieves intelligent transformation from data to executable solutions through multi-level decision logic and dynamic optimization algorithms, ensuring a balance between safety, efficiency, and economy in mine production. It includes a production scheduling optimization unit, a safety risk decision-making unit, and a cost-benefit optimization unit.

[0102] 1. Production Scheduling Optimization Unit

[0103] This unit dynamically generates the optimal mining operation plan and resource scheduling strategy based on constraints such as comprehensive equipment status, material inventory, and personnel allocation.

[0104] Construct a mixed-integer programming (MIP) model, defining the objective function and constraints: Where, q i For equipment i output, HI i For device health, e j For equipment j energy consumption, p k For the number of workers, S represents the maximum number of workers in a single area (default 20). 库存 For the inventory of materials, S 安全 This is the safety stock threshold.

[0105] The SchedulerAgent is designed and modeled using a Markov Decision Process (MDP). Its state space includes over 50 dimensions such as equipment health, inventory, and personnel distribution, while its action space covers over 20 operations including equipment start / stop, workspace allocation, and personnel scheduling. The reward function is: By learning long-term optimal strategies through deep Q-networks (DQN), the scheduling scheme is continuously optimized every 2 hours based on real-time data, adapting to dynamic events such as equipment failures (e.g., automatically calling backup equipment E04) and order changes (e.g., temporarily increasing production by 10%).

[0106] The decision generation process is as follows:

[0107] Initial plan generation: Based on the mining plan and equipment health status, prioritize the allocation of equipment with HI≥0.8 to the main mining area (e.g., allocate 3 coal mining machines with HI≥0.85 to Shaft No. 3).

[0108] Constraint verification: Check the number of workers (e.g., the current number of workers in shaft No. 3 is 18 ≤ 20) and the inventory of materials (e.g., the inventory of explosives is 120kg ≥ 100kg safety threshold) and generate a feasible solution.

[0109] Multi-scenario simulation: Five candidate scenarios (such as different equipment combinations and work time allocation) are simulated using a digital twin system to evaluate the Pareto optimal solution for output, energy consumption, and safety risks.

[0110] Generate a multi-objective scheduling scheme S = {main mining area, equipment configuration, personnel allocation}, including the start and stop times of each piece of equipment and the material transportation route. For example, if explosives are transported from warehouse C through main roadway 3, it takes 20 minutes.

[0111] Calculate the feasibility score S of the proposed solution score =0.4Q达成率 +0.3E 节约率 +0.3R 合规度 ,in, The degree of satisfaction of safety constraints is (0-1).

[0112] If S score If the health status is less than 0.8, the equipment allocation strategy will be automatically adjusted (e.g., one low-health device will be removed and one backup device will be added), and the plan will be regenerated. If the plan still fails to meet the standard after three consecutive adjustments, manual intervention will be triggered, and the dispatcher will manually adjust and verify the plan in the visual interface.

[0113] 2. Safety Risk Decision-Making Unit

[0114] Based on real-time safety status assessment results, this unit constructs a multi-level risk decision-making system, automatically generates response strategies such as ventilation adjustment, personnel evacuation, and emergency equipment activation, and verifies the compliance and effectiveness of the strategies through a safety rule engine.

[0115] Establish a risk level quantification model:

[0116] Combining fuzzy logic and decision tree techniques, a three-level risk level is defined:

[0117] Among them, "single parameters" refer to key safety indicators such as gas concentration (exceeding 1% LEL) and roof stress (exceeding the rated value by 15%).

[0118] Develop a standardized handling procedure library containing operation instructions for 100+ risk scenarios, as shown in the example:

[0119] Level 2 risk (gas concentration 1.2% LEL): Automatically send the instruction "Start standby ventilation fan in shaft No. 3, increase ventilation volume by 30%" to the equipment control bus, and at the same time send an early warning text message to the duty leader (with real-time concentration curve and affected area).

[0120] Level 3 risk (SSI<60 and roof stress exceeds threshold by 25%): Trigger the "full area power outage + personnel evacuation" plan, broadcast the evacuation route through the underground broadcast system in a loop, and send real-time location data to the mine emergency command center.

[0121] A rule base is established based on the inner circle procedures, with examples such as "The width of the evacuation route must be ≥1.2m" and "The activation time of emergency equipment must be ≤60 seconds". The logical correctness of the disposal strategy is verified by using a Petri net model.

[0122] It generates a risk handling instruction set CMD = {risk level, trigger parameters, handling measures, execution time, compliance status} in real time, and links with the industrial television system to capture on-site video to verify whether the ventilation fan starts within 10 seconds.

[0123] If the SSI does not increase by 10% (e.g., from 65 to 71) within 15 minutes after the action, it is determined that the "measure has failed", the risk level is automatically upgraded by 1 level, a higher level response is initiated (e.g., from level 2 to level 3, adding "partial power outage" measures), and the maintenance team is notified to arrive at the site with spare ventilation fan parts within 30 minutes.

[0124] If the compliance check fails (e.g., the evacuation route width is only 1.1m), the decision-making process is automatically backtracked, the weight of the path planning algorithm is adjusted (prioritizing routes that meet the width requirement rather than the shortest route), the instructions are regenerated, and a manual confirmation step is added.

[0125] 3. Cost-benefit optimization unit

[0126] This unit identifies cost optimization opportunities in areas such as equipment operation and maintenance, material procurement, and production organization. It evaluates the input-output ratio of decision-making schemes through a cost-sensitive model and provides the most economically optimal improvement suggestions.

[0127] Cost drivers are identified using Activity-Based Costing (ABC), and a cost calculation model is constructed: C 总 =C 设备 +C 物资 +C 人力 +C 能耗 +C 风险 ,in, R j The risk level is determined by the loss coefficient: Level 1 = 1, Level 2 = 5, Level 3 = 20.

[0128] Using Monte Carlo simulation technology, input uncertainties such as market price fluctuations (e.g., coking coal price fluctuations of ±10%) and equipment reliability (e.g., a failure probability of 5% when HI=0.8) output the expected cost distribution of the decision-making scheme. For example, when evaluating the "equipment upgrade scheme," 1000 market price changes are simulated to calculate the payback period with a 75% probability.

[0129] Real-time comparison of actual costs with budgeted costs, defining the cost deviation rate: For any single cost overrun exceeding 15%, or an overall cost overrun exceeding 10%, initiate cost attribution analysis to generate alternative solutions (such as evaluating the cost-effectiveness of domestically produced bearings and recommending three candidate suppliers).

[0130] Next, the cost analysis report CR = {Cheapest Steps, Budget, Actual Costs, Overspending Rate, Main Causes, Optimization Suggestions} is output, along with Pareto charts for each cost item (showing that material procurement accounts for 35% of the total cost, making it the primary target for optimization).

[0131] If a single item's cost is ≤8% or the overall cost is ≤5%, a cost warning will be generated and the business department will be advised to optimize the process (such as centralized procurement to reduce the unit price). If the threshold is exceeded, the cost anomaly audit process will be triggered: purchase orders, supplier contracts, and quality inspection reports will be retrieved to check for issues such as price monopoly or substandard quality. If necessary, a legal compliance review will be initiated, and the procurement strategy will be adjusted accordingly.

[0132] V. Decision Execution Feedback Module

[0133] This module transforms decision-making schemes into equipment control commands and personnel task lists, monitors execution results in real time, and provides feedback to the optimization model, forming a closed loop. It includes a command issuance unit and a result feedback unit.

[0134] 1. Command Issuance Unit

[0135] This unit converts scheduling parameters into control signals that the device can recognize, and supports verification against misoperation.

[0136] Protocol conversion: Converts "lifter speed 1800rpm" into a Modbus register value, with CRC checksum.

[0137] Task push: Send work steps via AR terminal, for example, [T20250702: Repair E01, first, disconnect the power].

[0138] If the command parameters exceed the equipment range (e.g., the maximum speed of the hoist is 2000 rpm, but the command is 2500 rpm), it will be automatically corrected to 1900 rpm. If the correction fails three times in a row, the equipment upgrade assessment will be triggered.

[0139] 2. Effect Feedback Unit

[0140] This unit compares the decision-making objectives with the actual results, quantifies the deviation, and optimizes the model.

[0141] The formula for calculating the achievement rate of decision-making objectives is: Among them, Q 实际 Q represents the actual output, quality, and other performance indicators achieved. 计划 For pre-set targets such as output and quality;

[0142] The formula for calculating the cost savings rate is: Among them, C 预算 For the pre-set cost budget, C 实际 The actual costs incurred.

[0143] Output an evaluation report, such as "Production achievement rate 96%, it is recommended to increase the weight of vibration features", and add the deviation data to the training set to trigger incremental training of the model.

[0144] After calculation, if η < 90% or δ < 5%, in-depth optimization (such as A / B testing to verify new strategies) is initiated, and the effective solution is solidified into the model and rule base.

[0145] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating the present invention and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A mine production and operation intelligent decision-making system based on multi-source heterogeneous data fusion, characterized in that, include: Data acquisition and access module: used to collect multi-source data from the entire mine area, including underground sensor data, electromechanical equipment data, personnel positioning data, and business management system data; Data preprocessing module: Cleans, transforms, and correlates the collected multi-source data to ensure data integrity and format consistency; Data fusion modeling module: Utilizes graph neural networks and transfer learning techniques to fuse multimodal data to build intelligent models for safety assessment, equipment prediction, etc., and supports model self-training and dynamic optimization; Intelligent Decision Module: Based on the real-time status indicators output by the fusion model, combined with constraint programming, reinforcement learning and cost-sensitive algorithms, it generates multi-objective decision-making schemes covering production scheduling, safety control and cost optimization; Decision execution feedback module: Transforms decision-making schemes into equipment control instructions and personnel task lists, and provides feedback to the optimization model.

2. The intelligent decision-making system for mine production and operation based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, The data acquisition and access module includes: Full-domain data acquisition unit: Through edge computing and protocol adaptation technology, it collects multi-source data from downhole sensors, electromechanical equipment, and personnel positioning in real time, adds spatiotemporal tags, and evaluates communication quality (CQ) to ensure the reliability of data transmission; Business System Interconnection Unit: Develops semantic middleware, defines unified business keys, achieves standardized extraction and real-time synchronization of system data, and performs integrity verification of business data; the formula for business data integrity verification is: Among them, F 有效 F represents the number of required fields with no missing fields. 总 This represents the total number of required fields. When the data communication quality assessment result CQ collected by the global data acquisition unit is ≥0.6, the business system interconnection unit will perform standardized extraction and real-time synchronization of the collected data; When the data communication quality assessment result CQ collected by the global data acquisition unit is <0.4 and lasts for 1 minute, relay transmission is triggered. If it is not restored within 3 minutes, the data is sent to the data preprocessing module for missing value repair.

3. The intelligent decision-making system for mine production and operation based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, The data preprocessing module includes: Intelligent cleaning unit: Utilizing physical laws and statistical methods, it identifies and repairs missing and outlier values ​​in the data, improving data quality and verifying data integrity; the data integrity verification formula is: Among them, D 有效 For valid data, D 总 Total data; Format conversion unit: unifies the time format, unit system and encoding rules of multi-source data, normalizes numerical data, and establishes a data dictionary mapping table; Data Association Unit: This unit associates multi-source data using spatiotemporal coordinates and business keys, constructs an association network, and calculates the association success rate. The formula for calculating the data association success rate is: Among them, K 有效 To associate valid data, K 总 This represents the total number of associations to be determined. When the data integrity DI is less than 90%, a manual verification process is triggered. The data administrator corrects the data through a visual interface. If "suspicious data" is found, the data is returned to the intelligent cleaning unit for secondary detection; otherwise, it enters the format conversion unit.

4. The intelligent decision-making system for mine production and operation based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, The data fusion modeling module includes: Feature fusion unit: Converts data into a unified feature vector, extracts spatial features using a graph convolutional network, extracts temporal features using an LSTM, and fuses them through a gating mechanism to generate a fused feature vector and calculate the feature discrimination index; the feature fusion formula is: h = σ(W g h g +W t h t ), where σ is the Sigmoid function, W g W t The weight matrix is ​​a learnable weight matrix; Model training unit: Selects the optimal algorithm through automated machine learning, supports self-training of multiple models, and uses Bayesian optimization of hyperparameters to evaluate and optimize the model; Real-time assessment unit: Calculates core indicators in real time, outputs early warning signals, and triggers corresponding decision-making units based on indicator values; the formula for calculating the safety status index is: Where w i x was determined by AHP. i ′ represents a standardized feature; If SSI < 70 triggers a yellow alert, the safety risk decision-making unit will be activated.

5. The intelligent decision-making system for mine production and operation based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, The intelligent decision-making module includes: Production scheduling optimization unit: Based on the constraints of comprehensive equipment status, material inventory, and personnel allocation, it dynamically generates the optimal mining operation plan and resource scheduling strategy, constructs a mixed integer programming model, designs a scheduling agent, learns the long-term optimal strategy through a deep Q-network, and continuously optimizes the scheduling scheme every 2 hours to adapt to dynamic events; Security Risk Decision Unit: Based on real-time security status assessment results, a multi-level risk decision system is constructed, and handling strategies are automatically generated. The compliance and effectiveness of the strategies are verified through a security rule engine. Cost-benefit optimization unit: Identifies cost optimization opportunities in equipment operation and maintenance, material procurement, and production organization; evaluates the input-output ratio of decision-making schemes using a cost-sensitive model; provides optimal economic improvement suggestions; identifies cost drivers using activity-based costing; constructs a cost calculation model; and utilizes Monte Carlo simulation technology to output the expected cost distribution. It compares actual costs with budgeted costs in real time and initiates cost attribution analysis and cost anomaly auditing processes. The cost calculation model for the cost-benefit optimization unit is: C 总 =C 设备 +C 物资 +C 人力 +C 能耗 +C 风险 ,in, R j The risk level is determined by the loss coefficient: Level 1 = 1, Level 2 = 5, Level 3 = 20.

6. The intelligent decision-making system for mine production and operation based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, The decision execution feedback module includes: Command issuing unit: converts scheduling parameters into control signals that the device can recognize, supports anti-misoperation verification, performs protocol conversion and task push, and corrects commands that exceed the device's range; The performance feedback unit compares the decision-making objectives with the actual results, quantifies the deviation and optimizes the model, calculates the achievement rate of the decision-making objectives and the cost saving rate, outputs an evaluation report, and adds the deviation data to the training set to trigger incremental training of the model; the formula for calculating the achievement rate of the decision-making objectives is: Among them, Q 实际 Q represents the actual performance indicator achieved. 计划 For a pre-set goal; The formula for calculating the cost savings rate is: Among them, C 预算 For the pre-set cost budget, C 实际 The actual costs incurred.

7. The intelligent decision-making system for mine production and operation based on multi-source heterogeneous data fusion as described in claim 2, characterized in that, In the business system interconnection unit, when the business data integrity verification result CI < 95%, a supplementary data entry reminder is pushed through WeChat. If the data entry is not completed within 2 hours, the same type of plan data from the previous 24 hours will be automatically used, and the logical error correction secondary verification of the data preprocessing module will be triggered.

8. The intelligent decision-making system for mine production and operation based on multi-source heterogeneous data fusion as described in claim 3, characterized in that, In the data association unit, when the data association success rate AR < 95%, the time window is automatically extended and the association is re-associated. If the failure is caused by an incorrect business number, the system returns to the business system interconnection unit for correction, and the qualified data enters the data fusion modeling module.

9. The intelligent decision-making system for mine production and operation based on multi-source heterogeneous data fusion as described in claim 4, characterized in that, In the feature fusion unit, when the feature discrimination D I When the value is less than 1.5, the weight of geographical features is automatically increased, and the model enters the training unit after reaching the target.

Images