A mineral energy collection digital twin monitoring method and system driven by quality-safety coupling

Abstract

The application discloses a mineral energy collection digital twin monitoring method and system driven by quality-safety coupling, and belongs to the technical field of intelligent monitoring of mineral energy exploitation. Quality parameters and safety parameters are collected in real time by a multi-source heterogeneous sensor network and arranged into vectors. A nonlinear coupling correlation model is constructed, a deep neural network is trained, and an edge-cloud collaborative architecture is deployed. The edge end performs real-time mapping and early warning at a frequency of no less than 1 Hz, and the cloud end performs multi-scenario situation deduction. A full-life-cycle digital twin is constructed. The quality level is determined based on Mahalanobis distance and pseudo-inverse matrix, the safety level is determined based on a fuzzy membership function, joint early warning grading response is performed through a two-dimensional early warning matrix, and a multi-objective collaborative optimization model is run to generate forward-looking control instructions and perform closed-loop execution. The application realizes mechanism-level coupling monitoring of quality and safety, changes passive alarm to forward-looking control, and improves the quality control and safety guarantee level of the whole process of mineral energy collection.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring of mineral energy mining, and in particular to a quality-safety coupled-driven digital twin monitoring method and system for mineral energy acquisition. Background Technology

[0002] With the deepening of smart mine construction, digital twin technology has been widely applied in the field of mineral and energy extraction. Various mine monitoring and management solutions based on digital twins have emerged in existing technologies.

[0003] For example, Chinese patent application CN119259232A discloses an intelligent safety management and control system for mine crushing based on digital twin technology, which achieves safety early warning by combining ore image analysis with blasting vibration signals. Chinese patent application CN121707339A discloses a risk situation prediction system for non-coal mines based on digital twins and reinforcement learning, which predicts the evolution trend of mine risks by constructing a digital twin.

[0004] In addition, existing research has explored the application of digital twin technology in areas such as mine safety early warning and production management, and some proposals have suggested the construction of a digital platform for the entire process of safety and quality management in open-pit mines. However, overall, existing technologies still have the following shortcomings: (1) Safety monitoring and quality monitoring are isolated from each other. In current digital twin mine systems, safety technologies (such as disaster early warning and risk control) and quality / production technologies (such as ore grade control and dilution rate optimization) belong to different technical branches and develop independently. Although some solutions mention the integration of the safety subsystem and the production subsystem, most of them remain at the level of data integration and display, and have not yet gone deep into the mechanism-level correlation modeling and collaborative reasoning between quality parameters and safety parameters. In existing technologies, quality control and safety monitoring are independent of each other in data acquisition, model building, early warning decision-making and other aspects, and lack a unified coupled analysis framework.

[0005] (2) The monitoring mode is mainly static monitoring, lacking the ability to make forward-looking adjustments. Most existing systems are one-way processes of "collection-monitoring-alarm", lacking the ability to predict and extrapolate future states based on real-time data. Most solutions only trigger alarms after an anomaly occurs, and cannot integrate and analyze the current state with future trends to generate forward-looking adjustment strategies.

[0006] (3) Quality control is passive and reactive. Quality testing of mineral energy (such as ore grade, calorific value, sulfur content, etc.) is mostly done through periodic sampling tests or offline analysis, lacking the ability to trace quality and control it online throughout the entire process. The acquisition frequency of quality parameters is low and the delay is large, making it difficult to reflect quality changes during the mining process in a timely manner.

[0007] (4) There are gaps in the whole-chain monitoring. Existing digital twin systems mostly focus on single links such as mining or transportation, and have not yet achieved integrated quality and safety monitoring of the entire life cycle from exploration to mining to transportation to washing and storage.

[0008] (5) The edge-cloud collaboration mechanism is imperfect. Although some existing solutions involve the collaboration between edge computing and cloud computing, they have failed to establish a knowledge distillation and lifecycle management mechanism from the high-precision model in the cloud to the lightweight model at the edge. It is difficult to balance the accuracy and real-time performance of the edge model.

[0009] In summary, there is currently a lack of a monitoring scheme for the entire mineral energy extraction process that can couple quality parameters and safety parameters at the mechanistic level, and realize collaborative extrapolation and forward-looking regulation throughout the entire life cycle. Summary of the Invention

[0010] The purpose of this invention is to provide a quality-safety coupled-driven digital twin monitoring method and system for mineral energy acquisition. By constructing a nonlinear coupling correlation model between quality parameters and safety parameters, a two-dimensional joint early warning matrix for quality and safety is established. With the help of an edge-cloud collaborative digital twin architecture, multi-scenario situational simulation and forward-looking collaborative regulation throughout the entire life cycle are realized, achieving a leap from "separate monitoring" to "joint decision-making", from "static monitoring" to "forward-looking regulation", and from "single-point control" to "full-chain collaboration".

[0011] To achieve the above objectives, this invention provides a quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting, comprising the following steps: Step 1: Through a multi-source heterogeneous sensor network deployed throughout the entire mineral energy extraction chain, real-time collection of quality parameter sets and safety parameter sets is performed, and these are arranged into quality parameter vectors according to a preset dimensional order. With security parameter vector ; Step 2: Process the obtained mass parameter vector With security parameter vector Perform data cleaning, spatiotemporal alignment, and standardization to form a unified quality-safe joint data pool; Step 3: Based on the joint quality-safety data pool, construct a nonlinear coupling correlation model between quality parameters and safety parameters, and deploy the coupling correlation model based on the edge-cloud collaborative computing architecture. The edge deploys a lightweight reduced-order model generated by knowledge distillation of the full model in the cloud, and performs real-time data mapping and first-level early warning at a frequency of no less than 1Hz. The cloud deploys a high-precision full model to perform multi-scenario situational simulation. Step 4: Based on the real-time acquired quality parameter vector With security parameter vector And a nonlinear coupling correlation model to construct a digital twin of the entire life cycle of mineral energy extraction; Step 5: Based on the quality parameter vector Calculate the quality status level based on the safety parameter vector. Calculate the safety status level, and make a joint early warning level determination in the two-dimensional early warning matrix based on the quality status level and the safety status level to trigger a graded response strategy; Step 6: Based on the current quality-safety coupling situation, use the digital twin to perform multi-scenario future situation simulation, run the quality-safety multi-objective collaborative optimization model, generate forward-looking collaborative control instructions and send them to the field execution equipment, collect execution effect data and feed it back to the digital twin.

[0012] This invention also provides a quality-safety coupled-driven digital twin monitoring system for mineral energy harvesting, used to implement the above method, comprising: The multi-source heterogeneous data acquisition and fusion module is used to collect quality parameter sets and safety parameter sets in real time through a multi-source heterogeneous sensor network deployed throughout the entire mineral energy acquisition chain, and to clean, align and standardize the collected data to form a unified quality-safety joint data pool. The quality-safety coupling correlation modeling module is used to construct a nonlinear coupling correlation model between quality parameters and safety parameters based on the joint quality-safety data pool. The coupling correlation model includes a coupling strength coefficient calculated using normalized mutual information. The edge-cloud collaborative digital twin building module includes a lightweight reduced-order model deployed on the edge computing device at the mining end and a full high-precision model deployed in the cloud. The lightweight reduced-order model is obtained through knowledge distillation from the full high-precision model and performs real-time data mapping and first-level early warning at a frequency of no less than 1Hz. The full high-precision model performs multi-scenario future situation inference. The quality-safety joint early warning and graded response module is used to calculate the quality status level based on the quality parameter set, calculate the safety status level based on the safety parameter set, and perform joint early warning level determination in a two-dimensional early warning matrix based on the quality status level and the safety status level to trigger a graded response strategy. The coupled-driven collaborative decision-making and forward-looking regulation module is used to run a quality-safety multi-objective collaborative optimization model based on the current quality-safety coupled situation and the inference results of the full high-precision model, and generate forward-looking collaborative regulation instructions. The closed-loop execution and feedback module is used to issue forward-looking collaborative control commands to the field execution equipment, collect execution effect data, and feed it back to the digital twin.

[0013] Preferably, in the edge-cloud collaborative digital twin construction module, when the lightweight reduced-order model at the edge detects the comprehensive coupling situation value... When the value exceeds 80% of the preset safety threshold, the cloud emergency acceleration computing mode is automatically triggered, shortening the update cycle of cloud multi-scenario simulation from once every 5 minutes to less than 30 seconds; when the coupling situation value falls back to below 60% of the safety threshold, the normal computing cycle is restored. The forward-looking collaborative control instructions generated by the coupled-driven collaborative decision-making and forward-looking control module include at least one of the following types: mining parameter adjustment instructions, ventilation and environmental control instructions, washing and beneficiation parameter adjustment instructions, and equipment operation and maintenance scheduling instructions.

[0014] Therefore, the quality-safety coupled driven digital twin monitoring method and system for mineral energy harvesting of the present invention has the following beneficial effects: (1) First realization of mechanism-level coupled monitoring of quality and safety: In the prior art, quality control and safety monitoring belong to different technical routes. This invention establishes a nonlinear coupled correlation model between the two at the mechanism level, and quantitatively calculates the coupling strength coefficient through normalized mutual information, filling the technical gap in this field.

[0015] (2) Change passive alarm to forward control: Existing technologies are mostly passive alarm modes triggered by thresholds. This invention realizes the prediction of future states through digital twin multi-scenario simulation, and automatically generates forward control instructions through NSGA-II multi-objective optimization algorithm, thus moving the accident prevention checkpoint forward.

[0016] (3) Achieving precise grading of joint quality and safety early warning: The two-dimensional early warning matrix constructed in this invention overcomes the limitations of existing single-dimensional threshold alarms. It achieves precise quantification of quality and safety status through Mahalanobis distance (including pseudo-inverse matrix processing) and fuzzy membership function, thus realizing precise grading response of quality and safety linkage.

[0017] (4) Breaking through the monitoring gaps in the entire chain: This invention covers the entire life cycle from exploration to storage, and realizes online traceability and control of the quality and safety of mineral energy throughout the entire process.

[0018] (5) Edge-cloud collaborative architecture that balances real-time performance and accuracy: This invention compresses the high-precision model in the cloud into a lightweight model at the edge through knowledge distillation technology and establishes a hot update mechanism for the model. While ensuring a real-time response of no less than 1Hz at the edge, it retains the high-precision inference capability of the cloud.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall architecture of a quality-safety coupled driven digital twin monitoring system for mineral energy harvesting according to the present invention. Figure 2 This is a flowchart of a quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting according to the present invention; Figure 3 This is a flowchart illustrating the construction process of the quality-safety coupled correlation model of this invention. Figure 4 This is a schematic diagram of the edge-cloud collaborative digital twin architecture of the present invention; Figure 5 This is a schematic diagram of the cross-mine knowledge transfer framework based on federated learning of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0023] like Figure 1 As shown, the overall system architecture of this invention includes three vertical layers: a data acquisition layer, a model building and intelligent decision-making layer, and an execution and feedback layer, as well as a horizontal support layer for cross-mine knowledge transfer based on federated learning.

[0024] The data acquisition layer uses a multi-source heterogeneous sensor network deployed across the entire mineral energy acquisition chain (exploration, mining, transportation, washing, and storage) to collect quality and safety parameter sets in real time. The multi-protocol adaptive acquisition submodule collects sensor data from different manufacturers and using different communication protocols in a unified manner. The spatiotemporal alignment and standardization submodule then solves the scale difference problem between static geological exploration data and dynamic production data, forming a unified quality-safety joint data pool.

[0025] Model building and intelligent decision-making layer construct a nonlinear coupling correlation model between quality parameters and safety parameters based on a joint data pool. The coupling correlation model is constructed using a deep neural network, and the coupling strength coefficient is calculated through normalized mutual information. The system constructs a full-lifecycle digital twin based on a coupled correlation model, achieving macro-, meso-, and micro-level twin mapping. It then uses the digital twin to jointly determine the quality status level (Mahathano distance) and safety status level (fuzzy membership function), implementing tiered responses through a two-dimensional early warning matrix. Based on the current quality-safety coupled situation, it performs multi-scenario future situation simulations and runs a quality-safety multi-objective collaborative optimization model to generate forward-looking control instructions. The coupled correlation model and digital twin are deployed using an edge-cloud collaborative computing architecture. At the edge, a lightweight, reduced-order model generated by knowledge distillation is deployed to perform real-time data mapping and first-level early warning at a frequency of at least 1Hz. At the cloud, a high-precision full-scale model is deployed to perform multi-scenario situation simulations.

[0026] The execution and feedback layer issues forward-looking control instructions to the field execution equipment, collects execution effect data, and feeds it back to the digital twin, forming a closed-loop control system.

[0027] The horizontal support layer enables cross-mine knowledge transfer based on federated learning. Under the premise that the original data of each mining area does not leave the mining area, the federated averaging algorithm is used to achieve collaborative training and parameter aggregation and updating of multi-mine end models.

[0028] like Figure 2 As shown, the specific process of a digital twin-based end-to-end monitoring method driven by the quality and safety coupling of mineral energy extraction is as follows: Step 1: Multi-source heterogeneous data acquisition.

[0029] A multi-source heterogeneous sensor network deployed across the entire mineral energy extraction chain is used to collect quality and safety parameter sets in real time. The entire extraction chain includes the exploration, mining, transportation, washing, and storage stages.

[0030] After data collection, the quality parameter set and the safety parameter set are arranged in a preset dimensional order to form a quality parameter vector. With security parameter vector .

[0031] The quality parameter set includes at least one of the following parameters: ore grade, calorific value, sulfur content, moisture, ash content, volatile matter, concentrate grade, recovery rate, tailings grade, and inventory quality degradation rate.

[0032] The safety parameter set includes at least one of the following parameters: ground pressure, ground temperature, rock displacement, roof settlement, gas concentration, dust concentration, toxic and harmful gas concentration, oxygen content, equipment temperature, equipment vibration, and equipment operating current.

[0033] To address the issue of heterogeneous communication protocols among mining equipment, a multi-protocol adaptive acquisition technology is adopted to uniformly acquire and adaptively identify sensor data from different manufacturers and using different communication protocols.

[0034] Step 2: Data cleaning, spatiotemporal alignment, and standardization.

[0035] The collected quality parameter set With safety parameter set Data cleaning, spatiotemporal alignment, and standardization are performed to form a unified quality-safety joint data pool.

[0036] Data cleaning employs filtering algorithms and outlier detection techniques to remove noise and outliers. Spatiotemporal alignment addresses the scale differences between static geological exploration data and dynamic production data, achieving uniformity of multi-source data in both time and space. Standardization converts data of different dimensions and formats into a unified data format and dimension.

[0037] Step 3: Quality-safety coupling correlation modeling.

[0038] like Figure 3 As shown, based on the joint quality-safety data pool, a nonlinear coupling correlation model between quality parameters and safety parameters is constructed according to the process of "sample input → definition of coupled situation expression → calibration of real situation value → solution of coupling strength coefficient → deep neural network training → model distillation deployment".

[0039] The mathematical form of the nonlinear coupling correlation model is: ; in ,for The mass parameter vector at time t, This represents the total number of dimensions for quality parameters. ,for The safety parameter vector at time t, This represents the total number of dimensions for security parameters. Let the vector be the rate of change of the quality parameter. For the rate of change vector of safety parameters, The preset delay time is determined based on the lag characteristics of the mining area's production process, and is preferably 5-30 minutes. This is a nonlinear coupling function learned through a deep neural network. for Quality-security coupled situational value at any given moment.

[0040] Deep neural networks can be fully connected deep neural networks (DNNs) or gated recurrent unit networks (GRUs). The training loss function for deep neural networks is: ; in, For summation operators, These are the weight parameters of the deep neural network. The L2 regularization coefficient is preferably in the range of 0.001 to 0.01, and is used to suppress model overfitting. For the output of deep neural networks Real-time coupled situational prediction values.

[0041] for The true calibration value of the coupled situation at any given time is constructed using the following calibration formula: ; in, This represents the average value of the quality parameter within the same statistical period. The quality benchmark value for the corresponding mining area can be determined based on the average parameter value of the historical high-quality range of the mining area; , , The preset weighting coefficient is preferably set to a value of [value to be filled in]. =0.5、 , =0.2, which can be adjusted according to the actual management focus of the mining area.

[0042] Accident severity scores are used to evaluate the severity of accidents within the same statistical period. The quantitative standard is: ; in, The number of accident fatalities during the statistical period. To count the number of seriously injured in accidents during the statistical period. To calculate the direct economic losses from accidents during the statistical period, the formula uses 30, 100, and... These correspond to the thresholds for determining the number of deaths, the number of serious injuries, and the direct economic losses in particularly serious accidents as defined in the "Regulations on Reporting and Handling Production Safety Accidents," and are used to normalize the three indicators. The preset weighting coefficient is preferably set to a value of [value to be filled in]. =0.5、 =0.3、 =0.2.

[0043] Hidden danger accumulation index The time weight is calculated as the duration for which a safety parameter exceeds the warning threshold within the same time period. The weighting is set to increase in a positive correlation with the magnitude of the parameter exceeding the threshold; the greater the magnitude of the exceedance, the higher the time weight.

[0044] Furthermore, normalized mutual information is used to calculate the first... Quality parameters With the Security parameters Coupling strength coefficient between , is represented as: ; in, The information entropy function; For quality parameters Edge information entropy, For safety parameters Edge information entropy, for and The joint information entropy. Before calculation, the continuous quality parameters and safety parameters are discretized using the equal-frequency binning method, with the preferred number of bins being 10 to 20.

[0045] when When the value of the quality parameter and the safety parameter exceeds the sum of the historical average coupling strength of the mining area and twice the standard deviation, the quality parameter and the safety parameter are marked as a strongly coupled pair and will be closely monitored in subsequent monitoring.

[0046] Step 4: Constructing a digital twin throughout the entire lifecycle.

[0047] Deploying coupled and related models based on an edge-cloud collaborative computing architecture. For example... Figure 4 As shown, the edge deployment uses a lightweight, reduced-order model generated by knowledge distillation from the full model in the cloud to perform real-time data mapping and first-level early warning at a frequency of no less than 1Hz; the cloud deployment uses a high-precision full model to perform multi-scenario situational simulations.

[0048] The lightweight reduced-order model at the edge is obtained and updated in the following way: The full high-precision model in the cloud serves as the teacher network. Knowledge distillation technology is used to generate a student network with fewer parameters and lower computational cost. This student network is the lightweight reduced-order model deployed at the edge. The full cloud model is periodically retrained or fine-tuned (preferably once daily at midnight), followed by knowledge distillation to generate an updated lightweight reduced-order model, which is then distributed to the edge via the Industrial Internet to replace the old model. When the edge detects that the prediction error of the lightweight reduced-order model exceeds a preset threshold, it proactively requests an emergency model update from the cloud. The preset threshold is preferably 5%, meaning that an emergency update request is triggered when the relative error between the lightweight model's prediction value and the output value of the full cloud model exceeds 5% for 10 consecutive minutes.

[0049] Based on real-time acquired quality parameter vector With security parameter vector And a coupled correlation model to construct a digital twin of the entire life cycle of mineral energy extraction.

[0050] The digital twin comprises three scales: a macro-scale, a three-dimensional digital twin model of the entire mining area, used to display the overall quality distribution and safety status; a meso-scale, a refined twin model of the mining face, transport roadways, and washing and beneficiation workshop; and a micro-scale, a local high-fidelity twin model of key equipment and key ore layers. All three scales of the twin models are updated synchronously with real-time acquired data, at an update frequency of no less than 1 Hz.

[0051] Step 5: Joint quality and safety early warning and graded response.

[0052] Based on mass parameter vector Calculate the quality status level based on the safety parameter vector. Calculate the safety status level, and determine the joint early warning level in the two-dimensional early warning matrix based on the quality status level and the safety status level, triggering a graded response strategy.

[0053] The method for determining the quality status level is as follows: calculate the real-time quality parameter vector. The mean vector of quality parameter samples within the historical high-quality interval Mahalanobis distance between them: ; Among them, the historical high-quality period refers to the historical period in which the quality parameters of the mining area are stably within the design index range under normal production conditions, and is determined by the stable production period data in the historical production records of the mining area. This is the mean vector of all quality parameter samples within this historical high-quality interval. This is the covariance matrix of all quality parameter samples within this historical high-quality interval.

[0054] This represents the Mahalanobis distance value corresponding to the real-time quality status. for The real-time quality parameter vector at any given time; Covariance matrix The Moore-Penrose pseudo-inverse matrix; when the covariance matrix When the matrix is ​​singular, a pseudo-inverse matrix is ​​used instead of direct inverse operation to avoid the problem that the determinant of the covariance matrix approaches zero due to multicollinearity between quality parameters (such as a strong negative correlation between ash content and volatile matter), making it impossible to invert the matrix. This is a vector transpose operation.

[0055] when At that time, it was judged as "excellent quality"; when At that time, it was determined to be "quality qualified"; when At that time, it was judged as "quality concern"; when At that time, it was determined to be "quality hazard"; in, The statistical boundary value is dynamically updated based on historical data of the mining area (preferred). (This can be adjusted based on the actual data distribution in the mining area).

[0056] The safety status level is determined using a fuzzy membership function, preferably a triangular membership function. The threshold boundaries for each level of the triangular membership function are determined as follows: Historical safety parameter data for at least six months during normal production in the mine are collected. Using the upper limit of the prescribed threshold for each safety parameter as a benchmark, and the standard deviation of the historical data as a scale, four level intervals are divided according to the degree of deviation from the benchmark. The boundary values ​​of each interval are the threshold parameters of the corresponding membership function. The threshold boundaries for each level are calibrated based on the mine's safety regulations and historical data. Based on the degree of deviation and trend of real-time safety parameters from the preset safety thresholds, four levels are defined: "Normal Safety," "Safety Concern," "Safety Hazard," and "Safety Emergency."

[0057] The two-dimensional early warning matrix uses quality status as the first dimension and safety status as the second dimension. Quality status includes four levels: excellent, acceptable, quality concern, and quality hazard. Safety status includes four levels: safe, safety concern, safety hazard, and safety emergency. Quality status and safety status are cross-located within the two-dimensional early warning matrix to determine the joint early warning level. The specific mapping relationship of the two-dimensional early warning matrix is ​​shown in Table 1 below.

[0058] Table 1: Two-Dimensional Early Warning Matrix

[0059] As shown in Table 1, the joint early warning levels are divided into four levels: Level I is normal operation, where the system automatically records data and does not trigger manual intervention; Level II is early warning and attention, where the system pushes early warning information to the terminal of the person in charge on duty; Level III is joint control, where the system automatically generates control plans and pushes them to the terminal of the mine manager or chief engineer, who must confirm and handle the situation within a preset time; Level IV is emergency response, where the system automatically triggers emergency shutdown or personnel evacuation procedures and simultaneously reports to the superior regulatory department.

[0060] Step 6: Multi-scenario future situation simulation and collaborative optimization.

[0061] Based on the current quality-safety coupling situation, digital twins are used to conduct multi-scenario future situation simulations.

[0062] Specifically, this includes: based on the current quality-safety coupling situation. And a lightweight, reduced-order model at the edge to predict the trajectory of future situation evolution. In the formula The preset situation prediction duration is set according to the mine production control response cycle, with an optimal range of 15 minutes to 24 hours.

[0063] The study simulates the combined impact of different control strategies on the quality-safety coupling situation. The control strategies include at least one of adjusting the mining speed, changing ventilation parameters, and changing the washing and screening parameters. The study outputs the range and confidence level of future changes in quality and safety parameters under each control strategy.

[0064] The edge-cloud collaboration mechanism is as follows: In normal mode, the full model in the cloud performs a simulation every 5 minutes; when the edge detects a comprehensive coupled situation value... When the value exceeds 80% of the preset safety threshold, the cloud emergency acceleration computing mode is automatically triggered, shortening the update cycle of cloud multi-scenario simulation from once every 5 minutes to less than 30 seconds; when the coupling situation value falls back to below 60% of the safety threshold, the normal computing cycle is restored.

[0065] Based on the results of multi-scenario future situation projections, a multi-objective collaborative optimization model for operational quality and safety is used to generate forward-looking collaborative control instructions.

[0066] The quality-safety multi-objective collaborative optimization model is as follows: ; in, The comprehensive target value for quality-safety collaborative optimization; For quality utility function, , For the first Target values ​​for each quality parameter For the first Weighting coefficients for each quality parameter; For safety utility function, , For the first The upper limit of the security threshold for each security parameter For the first The weighting coefficients of each safety parameter; a negative sign indicates that the closer the safety parameter is to the threshold, the lower the safety utility. To couple the risk penalty function, , For the first The change in a quality parameter before and after adjustment For the first The change in each safety parameter before and after adjustment; The weighting coefficient is adjusted in real time using a linear weighting based on the dynamic ratio of the current mineral energy quality premium to the safety compensation cost in the mining area.

[0067] The constraints include: , , In addition to constraints on equipment capacity and personnel allocation.

[0068] The collaborative optimization model uses the NSGA-II multi-objective genetic algorithm to solve the Pareto optimal solution set, and selects the solution with the closest Euclidean distance to the current operating condition as the optimal control command.

[0069] Forward-looking coordinated control instructions include at least one of the following types: mining parameter adjustment instructions, ventilation and environmental control instructions, washing and beneficiation parameter adjustment instructions, and equipment operation and maintenance scheduling instructions.

[0070] Step 7: Closed-loop execution and feedback.

[0071] like Figure 5 As shown, forward-looking collaborative control commands are issued to field execution equipment via the Industrial Internet, and execution effect data (including changes in quality parameters, safety parameters, and coupling status values ​​before and after control) are collected and fed back to the digital twin. The feedback data is used to update the state of the digital twin and correct the parameters of the coupling correlation model, forming a complete closed loop of "perception-inference-decision-execution-feedback".

[0072] Under the premise that the original quality and safety parameters of each mining area do not leave the mining area, the federated averaging algorithm is used to achieve collaborative training and parameter aggregation update of the quality-safety coupling model at multiple mining ends.

[0073] The federated averaging algorithm introduces a dynamic regularization term during the aggregation process to reduce the weight of mining end models whose gradient direction and the global average gradient direction exceed a preset angle threshold. The preset angle threshold is preferably 45° to overcome the heterogeneity (Non-IID) problem of multi-mining end data.

[0074] Furthermore, a model version management mechanism is established to record the version number, training data range, and performance metrics for each model update. When the mean square error of the coupled model on the validation set in a certain mining area exceeds 1.5 times the initial validation error for three consecutive days, the system automatically determines that the geological conditions of the mining area have changed significantly, triggering online fine-tuning of the local model for that mining area.

[0075] This invention also provides a digital twin end-to-end monitoring system driven by the quality and safety coupling of mineral energy extraction, comprising: The multi-source heterogeneous data acquisition and fusion module is used to collect quality parameter sets and safety parameter sets in real time through a multi-source heterogeneous sensor network deployed throughout the entire mineral energy acquisition chain, and to clean, align and standardize the collected data to form a unified quality-safety joint data pool. The quality-safety coupling correlation modeling module is used to construct a nonlinear coupling correlation model between quality parameters and safety parameters based on the joint quality-safety data pool. The coupling correlation model includes a coupling strength coefficient calculated using normalized mutual information. The edge-cloud collaborative digital twin building module includes a lightweight reduced-order model deployed on the edge computing device at the mining end and a full high-precision model deployed in the cloud. The lightweight reduced-order model is obtained through knowledge distillation from the full high-precision model and performs real-time data mapping and first-level early warning at a frequency of no less than 1Hz. The full high-precision model performs multi-scenario future situation inference. The quality-safety joint early warning and graded response module is used to calculate the quality status level based on the quality parameter set, calculate the safety status level based on the safety parameter set, and perform joint early warning level determination in a two-dimensional early warning matrix based on the quality status level and the safety status level to trigger a graded response strategy. The coupled-driven collaborative decision-making and forward-looking regulation module is used to run a quality-safety multi-objective collaborative optimization model based on the current quality-safety coupled situation and the inference results of the full high-precision model, and generate forward-looking collaborative regulation instructions. The closed-loop execution and feedback module is used to issue forward-looking and coordinated control commands to the field execution equipment, collect execution effect data, and feed it back to the digital twin; The federated learning-based cross-mine knowledge transfer module is used to achieve collaborative training and parameter aggregation updates of multi-mine models through a federated averaging algorithm, without requiring the original data from each mining area to leave the mining area. It also includes a model version management submodule. The connections between modules are as follows: the output of the multi-source heterogeneous data acquisition and fusion module is connected to the input of the quality-safety coupled correlation modeling module; the output of the quality-safety coupled correlation modeling module is connected to the inputs of the edge-cloud collaborative digital twin construction module and the quality-safety joint early warning and graded response module; the output of the edge-cloud collaborative digital twin construction module is connected to the input of the coupled-driven collaborative decision-making and forward-looking control module; the output of the coupled-driven collaborative decision-making and forward-looking control module is connected to the input of the closed-loop execution and feedback module; the output of the closed-loop execution and feedback module is fed back to the edge-cloud collaborative digital twin construction module and the quality-safety coupled correlation modeling module; and the federated learning-based cross-mine knowledge transfer module communicates with the quality-safety coupled correlation modeling modules of each mining area.

[0076] Example 1: Application in underground fully mechanized mining faces of coal mines In this embodiment, the system and method of the present invention are deployed in a fully mechanized mining face of a large coal mine.

[0077] 1. Sensor deployment and data acquisition: A mining-grade intrinsically safe near-infrared spectrometer (model: NIR-3000, sampling frequency 10Hz) is installed at the transfer point of the scraper conveyor at the coal mining face to monitor the calorific value, ash content, volatile matter, sulfur content, and other quality parameters of the coal flow in real time. Intrinsically safe mining-grade gas sensors (model: GJ4-100, sampling frequency 1Hz), intrinsically safe mining-grade dust sensors (model: GCG1000, sampling frequency 0.5Hz), intrinsically safe mining-grade ground pressure sensors (model: GPD60, sampling frequency 10Hz), and intrinsically safe mining-grade vibration sensors (model: GBZ-20, sampling frequency 100Hz) are deployed at the working face. All sensors are connected to a mining-grade explosion-proof and intrinsically safe edge computing gateway (configuration: industrial-grade quad-core processor, 16GB memory, 256GB solid-state drive) via a mining industrial Ethernet network.

[0078] After data collection is complete, all quality parameters and safety parameters are arranged in a preset dimensional order to form real-time quality parameter vectors and safety parameter vectors.

[0079] 2. Data Preprocessing and Fusion: After receiving the raw data from each sensor, the edge computing gateway uses a median filtering algorithm (window size 5) to remove noise, and then uses 3... Outliers were removed using established criteria. To address the scale difference between geological exploration data (static, spatial resolution 5m×5m) and production dynamic data (dynamic, temporal resolution 1Hz), bilinear interpolation was used for spatial alignment, and spline interpolation was used for temporal alignment. The aligned data was then stored in a time-series database (InfluxDB) with unified timestamps and spatial coordinates.

[0080] 3. Coupled model training: Using historical data from the mine over the past three years (including approximately 94.6 million records of quality parameters and 94.6 million records of safety parameters), a quality-safety coupling correlation model was trained. The deep neural network employed a three-layer gated recurrent unit (GRU) architecture, with 128, 64, and 32 hidden units per layer, ReLU activation function, and a fully connected output layer (output dimension 1). The training and test sets were split in an 8:2 ratio, and the Adam optimizer (learning rate 0.001) was used. =0.9, =0.999), batch size 256, training epochs 200. Regularization coefficient in the loss function. =0.001.

[0081] After training, the normalized mutual information between each quality parameter and the safety parameter is calculated. Before the calculation, the parameters are discretized by 15-bin equal-frequency binning.

[0082] The results showed that: "Gas emission was strongly positively correlated with the volatile matter content of the coal seam ( =0.82), "The roof displacement rate and the ore grade change rate show a delayed negative correlation (delay time)". (≈2.5 working days)

[0083] 4. Knowledge distillation and edge deployment: A full GRU model in the cloud (approximately 1.5 million parameters) serves as the teacher network. A lightweight student network (using a single-layer GRU architecture, 16 hidden units, and approximately 12,000 parameters) is generated through knowledge distillation. Distillation temperature parameters are also considered. =3, soft tag weight =0.7. After distillation, the student network is deployed to the edge computing gateway, with an inference latency of approximately 15ms, meeting the 1Hz real-time update requirement.

[0084] 5. System performance: After the system is put into operation, the quality-safety coupled situation will be monitored in real time. When the edge terminal predicts that the gas concentration may exceed the limit within the next 30 minutes (based on the current situation), (and trend extrapolation), the system adjusts the coal mining machine's cutting speed in advance (from 5m / min to 3m / min) and adjusts the ventilation volume accordingly (from 800m). 3 / min increased to 1200m 3 / min).

[0085] After 6 months of operation and verification: the response time for safety incidents was reduced from an average of 8 minutes to 2.5 minutes (a reduction of approximately 69%); the range of coal quality fluctuations (standard deviation of calorific value) narrowed from 185 kcal / kg to 112 kcal / kg (a reduction of approximately 39%); and the loss in selling price due to quality fluctuations was reduced by approximately 42%.

[0086] Those skilled in the art can replace the corresponding quality and safety parameter types according to the characteristics of different mining processes for petroleum, natural gas, non-metallic minerals, etc., and reuse the coupled modeling, edge-cloud collaboration and joint early warning methods of this invention, all of which fall within the protection scope of this invention.

[0087] Example 2: Multi-mining area federated learning application In this embodiment, the system of the present invention is deployed in five coal mines under a mining group.

[0088] 1. Local training: Each mine independently deploys its own data acquisition, coupled modeling, and digital twin subsystems. Each mine's local model uses the same GRU architecture (three layers, 128 / 64 / 32 hidden units) and is independently trained for 20 rounds on each mine's local historical data as the initial model.

[0089] 2. Federated Aggregation: The original data from each mine remains within its mining area. Every 24 hours, each mine uploads its encrypted model parameter updates (gradient information) to the group's central server. The central server uses a federated averaging algorithm (FedAvg) to aggregate the model parameters from the five mines. During aggregation, the angle between the gradient of each mine and the global average gradient is calculated. Mine-side models with angles exceeding a preset threshold (45° in this embodiment) are downweighted (weight coefficients are multiplied by 0.5). The aggregated global model is then distributed to each mine, where each mine performs three rounds of fine-tuning and fusion based on its local model with a learning rate of 0.001.

[0090] 3. Model version management: Establish a model version management database (PostgreSQL) to record the version number (v1.0~v12.0), number of participating miners, training data time range, and validation set performance metrics (such as MSE, MAE, R) for each aggregation. 2 ).

[0091] 4. Effect Verification: After 3 months (12 federated aggregations) of iterative training, the prediction accuracy (Ri) of each mine model was achieved. 2 On average, the performance improved by 23% (from 0.71 to 0.87). After loading the global model v12.0 as the initial model in newly commissioned mines, the model convergence time was shortened from 45 days to 7 days.

[0092] Example 3: Specific Implementation of Quality Status Level Determination The quality status grade determination method of this invention was applied in a copper mine.

[0093] 1. Historical high-quality range identified: We collected raw ore quality parameter data (including eight dimensions such as copper grade, moisture content, and sulfur content) from the mine during its normal production period over the past 12 months, totaling approximately 31.536 million records. We then calculated the mean vector for each quality parameter. Covariance matrix of samples within the historical high-quality interval Because there is a negative correlation between copper grade and sulfur content (correlation coefficient -0.73), the covariance matrix... The condition number is 1560, which is close to singular. This embodiment uses the Moore-Penrose pseudo-inverse matrix. Replaces the direct inverse operation.

[0094] 2. Calculation of statistical boundary values: Mahalanobis distance calculated based on historical data The distribution, determine =1.2 (68th percentile) =2.5 (95th percentile) =4.0 (99.7th percentile).

[0095] 3. Real-time determination: The mass parameter vector collected at a certain moment =[18.5%, 8.2%, 0.35%, ...], calculated as follows =1.8. Because It was determined to be "qualified in quality".

[0096] Example 4: Lifecycle Management of Lightweight Edge Models Based on Example 1, an update mechanism for the lightweight edge model is established.

[0097] 1. Regular Updates: During the daily non-production maintenance period in the mine (preferably 2:00 AM), the full cloud model undergoes incremental training (based on the data added the previous day). Subsequently, a knowledge distillation is performed to generate a new lightweight model. After the new model passes the accuracy verification on the validation set, it is distributed to each edge terminal via the mining industrial Ethernet and hot replacement is completed during the maintenance window of the day (preferably 4:00-5:00 AM). During the replacement, the edge terminal uses the old model to cache the results, and automatically switches after the replacement is completed.

[0098] 2. Urgent Update: When the edge device detects that the relative prediction error of the lightweight model (compared to the full model results in the cloud) exceeds 5% for 10 consecutive minutes, it proactively requests an emergency model update from the cloud. The cloud completes model redistillation and verification and distributes the update within 5 minutes.

[0099] After 6 months of operation, the edge model has completed 180 regular updates and 3 emergency updates, and the model accuracy (relative to the full model in the cloud) has remained above 95%.

[0100] Therefore, this invention employs a quality-safety coupled-driven digital twin monitoring method and system for mineral energy mining. This invention can be widely applied to intelligent monitoring of the entire mining process of coal, oil, natural gas, metallic minerals, and non-metallic minerals. The method and system of this invention can be upgraded based on existing mine monitoring systems, exhibiting good compatibility and scalability. Through this invention, the quality control level and safety assurance capabilities in the mineral energy mining process can be significantly improved, the incidence of safety accidents can be reduced, and quality fluctuation losses can be minimized, resulting in significant economic and social benefits.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting, characterized in that, Includes the following steps: Step 1: Through a multi-source heterogeneous sensor network deployed throughout the entire mineral energy extraction chain, real-time collection of quality parameter sets and safety parameter sets is performed, and these are arranged into quality parameter vectors according to a preset dimensional order. With security parameter vector ; Step 2: Process the obtained mass parameter vector With security parameter vector Perform data cleaning, spatiotemporal alignment, and standardization to form a unified quality-safe joint data pool; Step 3: Based on the joint quality-safety data pool, construct a nonlinear coupling correlation model between quality parameters and safety parameters, and deploy the coupling correlation model based on the edge-cloud collaborative computing architecture. The edge deploys a lightweight reduced-order model generated by knowledge distillation of the full model in the cloud, and performs real-time data mapping and first-level early warning at a frequency of no less than 1Hz. The cloud deploys a high-precision full model to perform multi-scenario situational simulation. Step 4: Based on the real-time acquired quality parameter vector With security parameter vector And a nonlinear coupling correlation model to construct a digital twin of the entire life cycle of mineral energy extraction; Step 5: Based on the quality parameter vector Calculate the quality status level based on the safety parameter vector. Calculate the safety status level, and make a joint early warning level determination in the two-dimensional early warning matrix based on the quality status level and the safety status level to trigger a graded response strategy; Step 6: Based on the current quality-safety coupling situation, use the digital twin to perform multi-scenario future situation simulation, run the quality-safety multi-objective collaborative optimization model, generate forward-looking collaborative control instructions and send them to the field execution equipment, collect execution effect data and feed it back to the digital twin.

2. The quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting according to claim 1, characterized in that: In step 3, the nonlinear coupling correlation model is expressed as: ； in, ,for The mass parameter vector at time t, This represents the total number of dimensions for quality parameters. ,for The safety parameter vector at time t, This represents the total number of dimensions for security parameters. Let the vector be the rate of change of the quality parameter. For the rate of change vector of safety parameters, To preset the delay time, This is a nonlinear coupling function learned through a deep neural network. for The quality-security coupled situational value at any given moment; The training loss function for a deep neural network is: ； in, These are the weight parameters of the deep neural network. The regularization coefficient is used. For the output of deep neural networks Predicted values ​​of the coupled situation at any given moment; for The true calibration value of the coupled situation at any given time is constructed using the following calibration formula: ； in, This represents the average value of the quality parameter within the same statistical period. This is the quality benchmark value for the corresponding mining area. , , Preset weighting coefficients; Score the severity of accidents for the same statistical period; The hazard accumulation index is a weighted integral of the duration for which safety parameters exceed the warning threshold within the same time period.

3. The quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting according to claim 2, characterized in that: Accident Severity Score , is represented as: ； in, The number of accident fatalities during the statistical period. To count the number of seriously injured in accidents during the statistical period. This refers to the amount of direct economic loss from accidents during the statistical period. These are preset weighting coefficients.

4. The quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting according to claim 3, characterized in that: Step 3 also includes the calculation of the coupling strength coefficient: The normalized mutual information is used to calculate the first Quality parameters With the Security parameters Coupling strength coefficient between , is represented as: ； in, The information entropy function; when When the value of the quality parameter and the safety parameter exceeds the sum of the historical average coupling strength of the mining area and twice the standard deviation, the quality parameter and the safety parameter are marked as a strongly coupled pair and will be closely monitored in subsequent monitoring.

5. The quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting according to claim 2, characterized in that: In step 5, the method for determining the quality status level is as follows: Calculate the real-time quality parameter vector with the historical high-quality interval mean vector Mahalanobis distance between , is represented as: ； in, Covariance matrix The Moore-Penrose pseudo-inverse matrix; when the covariance matrix When the matrix is ​​singular, a pseudo-inverse matrix is ​​used instead of direct inverse operation; This is a vector transpose operation; when At that time, it was judged as "excellent quality"; when At that time, it was determined to be "quality qualified"; when At that time, it was judged as "quality concern"; when At that time, it was determined to be "quality hazard"; in, These are statistical boundary values ​​that are dynamically updated based on historical data from the mining area. The safety status level is determined using a fuzzy membership function, and is divided into four levels: "Safety Normal", "Safety Attention", "Safety Danger" and "Safety Emergency" based on the degree of deviation and trend of real-time safety parameters from preset safety thresholds.

6. The quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting according to claim 5, characterized in that: In step 5, the two-dimensional early warning matrix uses quality status as the first dimension and safety status as the second dimension. Quality status and safety status are cross-located in the two-dimensional early warning matrix to determine the joint early warning level. The joint early warning level includes Level I normal operation, Level II early warning attention, Level III joint control, and Level IV emergency response. Tiered response strategies include: During normal operation at Level I, the system automatically records data without triggering manual intervention; When a Level II alert is triggered, the system pushes the alert information to the terminal of the person in charge on duty; During Level III joint control, the system automatically generates a control plan and pushes it to the mine manager or chief engineer's terminal, which must be confirmed and handled within a preset time. In the event of a Level IV emergency, the system will automatically trigger an emergency shutdown or personnel evacuation procedure and simultaneously report to the superior regulatory authority.

7. The quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting according to claim 6, characterized in that: Step 6, multi-scenario future situation simulation includes: Based on the current quality-safety coupling situation And a lightweight, reduced-order model at the edge to predict the trajectory of future situation evolution. , The length of time for forward prediction; The study simulates the combined impact of different control strategies on the quality-safety coupling situation. The control strategies include at least one of adjusting the mining speed, changing ventilation parameters, and changing the washing and screening parameters. Output the range and confidence level of future quality and safety parameters under each control strategy.

8. The quality-safety coupled-driven digital twin monitoring method for mineral energy harvesting according to claim 7, characterized in that: In step 6, the quality-safety multi-objective collaborative optimization model is as follows: ； in, For quality utility function, , For the first Target values ​​for each quality parameter For the first Weighting coefficients for each quality parameter; For safety utility function, , For the first The upper limit of the security threshold for each security parameter For the first The weighting coefficients of each safety parameter; a negative sign indicates that the closer the safety parameter is to the threshold, the lower the safety utility. To couple the risk penalty function, , For the first The change in a quality parameter before and after adjustment For the first The change in each safety parameter before and after adjustment; The weighting coefficient is adjusted in real time using a linear weighting based on the dynamic ratio of the current mineral energy quality premium to the safety compensation cost in the mining area. The constraints include: , , In addition to equipment capacity constraints and personnel allocation constraints; The collaborative optimization model uses the NSGA-II multi-objective genetic algorithm to solve the Pareto optimal solution set, and selects the solution with the closest Euclidean distance to the current operating condition as the optimal control command.

9. A quality-safety coupled-driven digital twin monitoring system for mineral energy harvesting, used to implement the method according to any one of claims 1-8, characterized in that, include: The multi-source heterogeneous data acquisition and fusion module is used to collect quality parameter sets and safety parameter sets in real time through a multi-source heterogeneous sensor network deployed throughout the entire mineral energy acquisition chain, and to clean, align and standardize the collected data to form a unified quality-safety joint data pool. The quality-safety coupling correlation modeling module is used to construct a nonlinear coupling correlation model between quality parameters and safety parameters based on the joint quality-safety data pool. The coupling correlation model includes a coupling strength coefficient calculated using normalized mutual information. The edge-cloud collaborative digital twin building module includes a lightweight reduced-order model deployed on the edge computing device at the mining end and a full high-precision model deployed in the cloud. The lightweight reduced-order model is obtained through knowledge distillation from the full high-precision model and performs real-time data mapping and first-level early warning at a frequency of no less than 1Hz. The full high-precision model performs multi-scenario future situation inference. The quality-safety joint early warning and graded response module is used to calculate the quality status level based on the quality parameter set, calculate the safety status level based on the safety parameter set, and perform joint early warning level determination in a two-dimensional early warning matrix based on the quality status level and the safety status level to trigger a graded response strategy. The coupled-driven collaborative decision-making and forward-looking regulation module is used to run a quality-safety multi-objective collaborative optimization model based on the current quality-safety coupled situation and the inference results of the full high-precision model, and generate forward-looking collaborative regulation instructions. The closed-loop execution and feedback module is used to issue forward-looking collaborative control commands to the field execution equipment, collect execution effect data, and feed it back to the digital twin.

10. A quality-safety coupled-driven digital twin monitoring system for mineral energy harvesting according to claim 9, characterized in that: In the edge-cloud collaborative digital twin construction module, when the lightweight reduced-order model at the edge detects the comprehensive coupling situation value... When the value exceeds 80% of the preset safety threshold, the cloud emergency acceleration computing mode is automatically triggered, shortening the update cycle of cloud multi-scenario simulation from once every 5 minutes to less than 30 seconds; when the coupling situation value falls back to below 60% of the safety threshold, the normal computing cycle is restored. The forward-looking collaborative control instructions generated by the coupled-driven collaborative decision-making and forward-looking control module include at least one of the following types: mining parameter adjustment instructions, ventilation and environmental control instructions, washing and beneficiation parameter adjustment instructions, and equipment operation and maintenance scheduling instructions.

Images