Intelligent Decision-Making System and Method for Delamination Grouting Timing Based on Space-Ground Coordination

Abstract

This invention belongs to the field of intelligent monitoring and safety early warning technology. It provides an intelligent decision-making system and method for delamination grouting timing based on space-ground collaboration. The system includes a multi-source monitoring data synchronous acquisition and preprocessing module, an InSAR deformation field spatiotemporal feature extraction module, a microseismic event cluster multi-parameter analysis module, a delamination collaborative evolution stage intelligent identification module, and a grouting timing decision and early warning output module. Each module works together to utilize the InSAR macroscopic deformation field and microseismic micro-fracture sequence. Through multi-parameter feature extraction and machine learning intelligent identification, it accurately diagnoses the real-time evolution stage of mining-induced delamination and provides quantitative and scientific optimal grouting timing decisions, providing effective technical support for mine disaster prevention and control.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring and safety early warning technology, specifically a smart decision-making system and method for the timing of delamination grouting based on space-ground coordination. Background Technology

[0002] In mining operations, the development and sudden breakthrough of mining-induced delamination are key factors inducing severe surface subsidence, tunnel damage, and even rockbursts. Grouting to fill the delamination is a core engineering measure to control its hazards, but its effectiveness is highly dependent on the timing of grouting. Currently, the technology for selecting the timing of delamination grouting faces the following critical issues that urgently need to be addressed:

[0003] 1. Current methods for sensing delamination are limited and cannot fully reveal its entire evolutionary process. Existing technologies often rely on single monitoring methods, such as using only InSAR to monitor surface deformation or only microseismic monitoring to detect rock mass fracturing. While InSAR can accurately capture macroscopic deformation, it is insensitive to damage accumulation and pre-fracturing precursors within underground rock masses; microseismic monitoring, while directly reflecting fracturing, struggles to directly and accurately correlate with the macroscopic spatial evolution of delamination. This limitation prevents a comprehensive understanding of delamination dynamics.

[0004] 2. The timing of grouting decisions heavily relies on experience and lacks scientific quantitative basis. Currently, the decision on when to initiate grouting projects largely depends on the experience of engineers or simple deformation threshold alarms, making it impossible to accurately determine whether the delamination is in a safe, slow development phase or has entered a dangerous, critical instability phase. This often leads to reactive, remedial grouting, missing the optimal intervention window, resulting in not only high remediation costs but also significant safety risks.

[0005] 3. The phenomenon of "information silos" among multi-source monitoring data is serious, failing to form a joint decision-making force. Although modern mines deploy multiple monitoring systems simultaneously, InSAR, microseismic, and other data are usually analyzed independently by different teams, resulting in scattered conclusions. There is a lack of a unified framework to deeply integrate and collaboratively interpret this multi-dimensional and multi-scale information, and its profound value has not been fully explored, failing to provide strong joint evidence to support grouting decisions. Summary of the Invention

[0006] The purpose of this invention is to solve the above-mentioned technical problems. Based on the core principle that macroscopic deformation and microscopic fracture of rock mass are manifestations of the same mechanical process at different scales, this invention provides an intelligent decision-making system and method for delamination grouting timing based on space-ground collaboration. It collaboratively utilizes the InSAR macroscopic deformation field and microseismic microscopic fracture sequence, and through multi-parameter feature extraction and machine learning intelligent identification, it accurately diagnoses the real-time evolution stage of mining-induced delamination, and provides quantitative and scientific decisions on the optimal grouting timing accordingly.

[0007] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0008] A ground-space coordinated intelligent decision-making system for determining the timing of delamination grouting includes:

[0009] The multi-source monitoring data synchronous acquisition and preprocessing module is used to synchronously acquire, spatiotemporally register, and control the data quality of the input time-series SAR images and downhole microseismic monitoring data.

[0010] The InSAR deformation field spatiotemporal feature extraction module is used to extract key deformation features characterizing delamination evolution from time-series SAR image data.

[0011] The microseismic event cluster multi-parameter analysis module is used to identify event clusters from downhole microseismic monitoring data and extract key parameters characterizing the rock mass fracture process.

[0012] The intelligent identification module for the co-evolution stage of the delamination is based on deeply fused synthetic aperture radar images and microseismic features. Through a machine learning model enhanced with physical mechanisms, it automatically identifies the current evolution stage of the delamination.

[0013] The grouting timing decision and early warning output module automatically generates grouting timing suggestions and triggers early warnings based on the identified delamination evolution stage.

[0014] A smart decision-making method for the timing of delamination grouting based on space-ground coordination includes the following steps:

[0015] S1. The input time-series SAR images and downhole microseismic continuous waveform data are processed by the multi-source monitoring data synchronous acquisition and preprocessing module to obtain a spatiotemporally synchronized surface deformation and downhole microseismic dataset.

[0016] S2. The InSAR deformation field spatiotemporal feature extraction module is used to process the spatiotemporally synchronized surface deformation and downhole microseismic datasets to extract key deformation features characterizing delamination evolution.

[0017] S3. The spatiotemporally synchronized surface deformation and downhole microseismic datasets are processed through the microseismic event cluster multi-parameter analysis module to extract key parameters characterizing the rock mass fracture process.

[0018] S4. The intelligent identification module for the co-evolution stage of delamination integrates the key deformation features characterizing the delamination evolution with the key parameters characterizing the rock mass fracture process to determine the current evolution stage of the delamination.

[0019] S5. The grouting timing decision and early warning output module processes the identification results of the delamination evolution stage to generate grouting decision and early warning information.

[0020] Furthermore, the specific processing procedure of S1 includes:

[0021] S11. Spatiotemporal reference unification: Unify the spatiotemporal reference of all data from time-series SAR images and downhole microseismic monitoring to the same coordinate system and Beijing time, and set a unified analysis time window and window sliding step size;

[0022] S12. Temporal SAR Image Data Preprocessing: Temporal SAR images are processed using small baseline set interferometric radar technology to obtain the deformation time series of the study area and output the deformation rate of each pixel. The unit is mm / month. At the same time, the average coherence coefficient γ of each pixel is calculated as an InSAR data quality evaluation index.

[0023] S13. Microseismic data preprocessing: Event triggering, P-wave and S-wave first arrival picking, source location and magnitude calculation are performed on the continuous waveform data collected by the downhole microseismic monitoring network. The double-difference location method is used to accurately locate microseismic events, and the location error is controlled within 50 meters. At the same time, the location residual and signal-to-noise ratio of each event are recorded as microseismic data quality evaluation indicators.

[0024] S14. Data Quality Collaborative Control: For each analysis unit, a data quality collaborative control mechanism is implemented, including:

[0025] Calculate InSAR data quality score , which is the arithmetic mean of the average coherence coefficient γ of all pixels in the unit;

[0026] Calculate the quality fraction of microseismic data The calculation formula is as follows:

[0027]

[0028] in, For the first Location residuals of individual microseismic events (unit: seconds). Its signal-to-noise ratio. This represents the total number of microseismic events within the unit.

[0029] The Data Quality Co-Control Factor (DQF) of the analysis unit is calculated using the following formula:

[0030]

[0031] Set the first threshold ;when When this happens, the system automatically triggers a data quality warning and executes a data compensation strategy. This strategy includes extending the analysis window or using weighted supplementary calculations based on historical high-quality data from the previous three periods, until... .

[0032] S15. Data Fusion Interface Construction: The spatiotemporal coordinates of microseismic events controlled by quality are associated with the spatiotemporal coordinates of InSAR pixels to construct a spatiotemporally synchronized surface deformation and downhole microseismic dataset.

[0033] Furthermore, the specific processing procedure of S2 includes:

[0034] S21. Deformation Acceleration Calculation: Perform a quadratic difference on the deformation time series of each pixel to calculate its monthly deformation acceleration. The unit is mm / month²;

[0035] S22. Calculation of Deformation Spatial Gradient and Aggregation Characteristics: Calculation of the spatial gradient modulus of the deformation rate field. The system identifies abrupt deformation zones, uses the DBSCAN spatial clustering algorithm to identify the boundaries of the settling funnel, and calculates the expansion rate of the settling funnel. The unit is m / month.

[0036] Furthermore, the specific processing procedure of S3 includes:

[0037] S31. Microseismic event cluster identification: Based on the spatiotemporal density of microseismic events, the OPTICS clustering algorithm is used to identify microseismic event clusters by setting the minimum number of samples and the maximum neighborhood radius.

[0038] S32. Cluster Average Depth Calculation: For each microseismic event cluster, calculate the arithmetic mean of the depths of all its events, which will be used as the cluster average depth. The unit is meters (m).

[0039] S33. Microseismic Event Frequency Statistics: Count the total number of events for each cluster within the analysis time window, divide by the window duration, and obtain the frequency of microseismic events per unit time. The unit is times / day;

[0040] S34. Cluster Energy Release Rate Calculation: For each cluster of microseismic events, calculate its cumulative microseismic energy release rate per unit time. The calculation formula is: ,in For the first in the cluster The magnitude of the event The analysis window duration (days) is used, and the energy release rate is expressed in joules per day.

[0041] S35, b-value spatiotemporal scanning: For each micro-event seismic cluster, the b-value is calculated using the maximum likelihood method based on the GR relationship, and the calculation window length and step size are set;

[0042] S36. Focal Mechanism and Rupture Type Analysis: Focal mechanism inversion is performed on microseismic events with a moment magnitude greater than 1.0 within the microseismic event cluster, and the proportion of extensional rupture events within the cluster is calculated. .

[0043] Furthermore, the specific processing procedure of S4 includes:

[0044] S41. Construction of Deep Fusion Feature Vector: For each analysis unit, construct an eleven-dimensional deep fusion feature vector. ,vector The elements include: deformation rate Deformation acceleration Deformation gradient modulus , rate of expansion of settling funnel Microseismic energy release rate Frequency of microseismic events b-value, average cluster depth The proportion of tensile rupture Vertical distance Energy-deformation coupling ratio The vertical distance The vertical distance between the centroid depth of the microseismic cluster and the geometric center of the InSAR settlement funnel on the projection plane; the formula for calculating the energy-deformation coupling ratio R is: .

[0045] S42. Definition of Evolutionary Stages: Based on rock mechanics theory, delamination evolution is divided into three stages and assigned quantitative indicators:

[0046] Stage I (Generation Period): Deformation Rate < 5 mm / month, absolute value of deformation acceleration < 0.5 mm / month²; frequency of microseismic events < 2 times / day, b-value > 1.0, energy release rate < 1×10 8 Joules per day;

[0047] Phase II (Expansion Period): 5 mm / month ≤ ≤ 15 mm / month, 0.5 mm / month² ≤ ≤ 2.0 mm / month²; 2 times / day ≤ ≤ 10 times / day, 0.8 ≤ b value ≤ 1.0, 1×10 8 Joules / day ≤ ≤ 1×10 9 Joules per day;

[0048] Phase III (Pre-connection stage): > 15mm / month, > 2.0 mm / month²; > 10 times / day, b-value < 0.8 > 1×10 9 Joules per day, and > 60%.

[0049] S43, Intelligent identification model training and inference.

[0050] Furthermore, the specific implementation of S43 includes:

[0051] S431, Feature Grouping Identifier: Divide the features in the fused feature vector F into macroscopic deformation response groups ( ) and micro-fracture driving group ( ), and use this grouping information as metadata input during model training;

[0052] S432. Physical Constraints on the Loss Function: In the standard cross-entropy loss function... Based on this, add a physical logic penalty term. This constitutes the total loss function. ;

[0053] The The model will " < 2 mm / month and Samples with "< 0.5 times / day" are classified as Stage III, or samples with "b value < 0.7 and energy-deformation coupling ratio R increases by more than 50% per cycle" are classified as Stage I and penalized.

[0054] The intelligent identification model outputs two task results simultaneously: the main task is the classification probability of the off-layer evolution stage [P(I), P(II), P(III)]; the auxiliary task is the probability of entering stage III within a future window. ;

[0055] S433. Utilize the confirmed evolutionary stages in historical data as the main task labels, and combine them with the derived... As auxiliary task labels, train the XGBoost model enhanced with physical mechanisms;

[0056] S434. During inference, the fused feature vector generated in real time will be... By inputting the trained model, a comprehensive diagnostic report can be obtained, including stage identification results and short-term risk probability; simultaneously, the system is set to consider the maximum probability of the main task. Only then is the identification result considered valid.

[0057] Furthermore, the specific processing procedure of S5 includes:

[0058] S51. Setting grouting timing decision rules:

[0059] An early warning for optimal grouting timing will be triggered when any of the following conditions are met:

[0060] Condition 1: The identification result is Stage II, and the deformation acceleration... Two consecutive analysis periods showed positive values, and the energy release rate of the microseismic cluster was also positive. It exceeds 1.5 times the standard deviation of its historical trend value;

[0061] Condition 2: The identification result is Stage III, and the auxiliary task output is... > 0.7;

[0062] S52. Early Warning Information Generation and Push: Once an early warning is triggered, the system automatically generates a structured early warning report containing the location of the risk area, the evolution stage, the confidence level, key evidence diagrams, and decision recommendations, and pushes it to the production safety manager within 10 minutes via SMS, email, and platform alarms through the message interface.

[0063] Compared with the prior art, the system and method of the present invention have the following significant advantages:

[0064] 1. It has achieved advanced perception and accurate identification of the evolution state of the delamination layer, and the early warning time window has been significantly advanced;

[0065] This invention can identify the critical point at which delamination transitions from stable expansion to unstable connection by detecting the coordinated anomalies of microseismic activity and deformation acceleration before significant subsidence or collapse occurs on the ground surface, providing an advance warning period of several weeks or even months for grouting projects. The microseismic b-value and energy release rate extracted by this invention are microscopic precursors to the intensification of internal rock fracturing and impending instability. The intelligent identification model integrates these microscopic precursors with InSAR macroscopic deformation acceleration signals to accurately capture the critical state of delamination evolution.

[0066] 2. The decision-making process is highly scientific, solving the problems of reliance on experience and blind spots in selecting the timing of grouting;

[0067] The timing of grouting has been transformed from "remedial treatment after problems occur" or "regular construction based on experience" to "precise intervention based on the identification of the critical state of the system," avoiding the drawbacks of grouting too early (when the delamination is not fully developed and repeated grouting is required) or too late (when penetration failure has occurred and the treatment cost has increased dramatically). The decision-making rule for grouting timing is based on a deep understanding of the complete evolution law of delamination, and clarifies the best time window from "the end of the expansion stage" to "the precursor to penetration" to ensure grouting efficiency and effectively control risks.

[0068] 3. It has formed a comprehensive "space-ground integrated" perception capability, with strong anti-interference ability and high reliability;

[0069] Single monitoring methods are susceptible to interference or interpretation ambiguity. This invention cross-validates InSAR (sensitive to continuous surface deformation) with microseismic activity (sensitive to subsurface rock fracturing), significantly reducing false alarms and false negatives. For example, accelerated surface subsidence may originate from multiple factors, but if it is accompanied by a specific pattern of microseismic activity at key strata, it can be identified as a valid signal of delamination evolution. By constructing a fused feature vector, the model can learn the inherent correlation and coupling between the two data sources, thus making more robust and reliable identifications.

[0070] 4. It has achieved automation and intelligence in mine safety monitoring, significantly improving management efficiency;

[0071] The system of this invention is a complete "data input, decision output" intelligent system that can automatically process massive amounts of monitoring data, intelligently identify risks, and proactively push early warning reports, freeing security analysts from heavy manual data analysis and achieving 24 / 7 uninterrupted intelligent monitoring. Attached Figure Description

[0072] Figure 1 System structure block diagram of the present invention;

[0073] Figure 2 This is a flowchart of the method of the present invention;

[0074] Figure 3 This refers to the key process data in each analysis window of the analysis unit E52;

[0075] Figure 4 This is a comparison of key performance indicators between the present invention and three typical traditional methods. Detailed Implementation

[0076] To make the objectives, technical solutions, and advantages of this invention clearer, the following description, in conjunction with the accompanying drawings and embodiments, further illustrates the invention. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit it.

[0077] like Figure 1 As shown, the intelligent decision-making system for the timing of delamination grouting based on space-ground coordination includes:

[0078] The multi-source monitoring data synchronous acquisition and preprocessing module is used to synchronously acquire, spatiotemporally register, and control the data quality of the input time-series SAR images and downhole microseismic monitoring data.

[0079] The InSAR deformation field spatiotemporal feature extraction module is used to extract key deformation features characterizing delamination evolution from time-series SAR image data.

[0080] The microseismic event cluster multi-parameter analysis module is used to identify event clusters from downhole microseismic monitoring data and extract key parameters characterizing the rock mass fracture process.

[0081] The intelligent identification module for the co-evolution stage of the delamination is based on deeply fused synthetic aperture radar images and microseismic features. Through a machine learning model enhanced with physical mechanisms, it automatically identifies the current evolution stage of the delamination.

[0082] The grouting timing decision and early warning output module automatically generates grouting timing suggestions and triggers early warnings based on the identified delamination evolution stage.

[0083] A smart decision-making method for the timing of delamination grouting based on space-ground coordination, such as Figure 2 As shown, it includes the following steps:

[0084] S1. The input time-series SAR images and downhole microseismic continuous waveform data are processed by the multi-source monitoring data synchronous acquisition and preprocessing module to obtain a spatiotemporally synchronized surface deformation and downhole microseismic dataset.

[0085] The specific processing steps of S1 include:

[0086] S11. Spatiotemporal reference unification: Unify the spatiotemporal reference of all data from time-series SAR images and downhole microseismic monitoring to the same coordinate system and Beijing time, and set a unified analysis time window and window sliding step size;

[0087] S12. Temporal SAR Image Data Preprocessing: Temporal SAR images are processed using Small Baseline Set Interferometric Radar (SBAS-InSAR) technology to obtain the deformation time series of the study area and output the deformation rate of each pixel. The unit is mm / month. Simultaneously, the average coherence coefficient γ for each pixel is calculated as an InSAR data quality evaluation indicator;

[0088] S13. Microseismic data preprocessing: Event triggering, P-wave and S-wave first arrival picking, source location and magnitude calculation are performed on the continuous waveform data collected by the downhole microseismic monitoring network. The double-difference location method is used to accurately locate microseismic events, and the location error is controlled within 50 meters. At the same time, the location residual and signal-to-noise ratio of each event are recorded as microseismic data quality evaluation indicators.

[0089] S14. Data Quality Collaborative Control: For each analysis unit, a data quality collaborative control mechanism is implemented, including:

[0090] Calculate InSAR data quality score , which is the arithmetic mean of the average coherence coefficient γ of all pixels in the unit;

[0091] Calculate the quality fraction of microseismic data The calculation formula is as follows:

[0092]

[0093] in, For the first Location residuals of individual microseismic events (unit: seconds). Its signal-to-noise ratio. This represents the total number of microseismic events within the unit.

[0094] The Data Quality Co-Control Factor (DQF) of the analysis unit is calculated using the following formula:

[0095]

[0096] Set the first threshold ;when When this happens, the system automatically triggers a data quality warning and executes a data compensation strategy. This strategy includes extending the analysis window or using weighted supplementary calculations based on historical high-quality data from the previous three periods, until... ;

[0097] By dynamically evaluating and compensating for the quantitative DQF factor, the interference of the overall identification system caused by the instantaneous quality decline of a single data source is avoided from the mechanism, laying a reliable foundation for the deep integration of multi-source information and enabling the system to have strong robustness for stable operation in complex mining environments.

[0098] S15. Data Fusion Interface Construction: The spatiotemporal coordinates of microseismic events controlled by quality are associated with the spatiotemporal coordinates of InSAR pixels to construct a spatiotemporally synchronized surface deformation and downhole microseismic dataset.

[0099] S2. The InSAR deformation field spatiotemporal feature extraction module is used to process the spatiotemporally synchronized surface deformation and downhole microseismic datasets to extract key deformation features characterizing delamination evolution.

[0100] The specific processing procedure of S2 includes:

[0101] S21. Deformation Acceleration Calculation: Perform a quadratic difference on the deformation time series of each pixel to calculate its monthly deformation acceleration. The unit is mm / month²;

[0102] S22. Calculation of Deformation Spatial Gradient and Aggregation Characteristics: Calculation of the spatial gradient modulus of the deformation rate field. The system identifies abrupt deformation zones, uses the DBSCAN spatial clustering algorithm to identify the boundaries of the settling funnel, and calculates the expansion rate of the settling funnel. The unit is m / month.

[0103] S3. The spatiotemporally synchronized surface deformation and downhole microseismic datasets are processed through the microseismic event cluster multi-parameter analysis module to extract key parameters characterizing the rock mass fracture process.

[0104] The specific processing procedure of S3 includes:

[0105] S31. Microseismic event cluster identification: Based on the spatiotemporal density of microseismic events, the OPTICS clustering algorithm is used to identify microseismic event clusters by setting the minimum number of samples and the maximum neighborhood radius.

[0106] S32. Cluster Average Depth Calculation: For each microseismic event cluster, calculate the arithmetic mean of the depths of all its events, which will be used as the cluster average depth. The unit is meters (m).

[0107] S33. Microseismic Event Frequency Statistics: Count the total number of events for each cluster within the analysis time window, divide by the window duration, and obtain the frequency of microseismic events per unit time. The unit is times / day;

[0108] S34. Cluster Energy Release Rate Calculation: For each cluster of microseismic events, calculate its cumulative microseismic energy release rate per unit time. The calculation formula is: ,in For the first in the cluster The magnitude of the event The analysis window duration (days) is used, and the energy release rate is expressed in joules per day.

[0109] S35, b-value spatiotemporal scanning: For each micro-event seismic cluster, the b-value is calculated using the maximum likelihood method based on the GR relationship, and the calculation window length and step size are set;

[0110] S36. Focal Mechanism and Rupture Type Analysis: Focal mechanism inversion is performed on microseismic events with a moment magnitude greater than 1.0 within the microseismic event cluster, and the proportion of extensional rupture events within the cluster is calculated. .

[0111] S4. The intelligent identification module for the co-evolution stage of delamination integrates the key deformation features characterizing the delamination evolution with the key parameters characterizing the rock mass fracture process to determine the current evolution stage of the delamination.

[0112] The specific processing procedure of S4 includes:

[0113] S41. Construction of Deep Fusion Feature Vector: For each analysis unit, construct an eleven-dimensional deep fusion feature vector. ,vector The elements include: deformation rate Deformation acceleration Deformation gradient modulus , rate of expansion of settling funnel Microseismic energy release rate Frequency of microseismic events b-value, average cluster depth The proportion of tensile rupture Vertical distance Energy-deformation coupling ratio The vertical distance The vertical distance between the centroid depth of the microseismic cluster and the geometric center of the InSAR settlement funnel on the projection plane; the formula for calculating the energy-deformation coupling ratio R is: ;

[0114] The construction of deep fusion feature vectors is not a simple parameter stacking, but a feature engineering based on the rock mechanics principle that macroscopic deformation is controlled by microscopic fracture. It quantitatively correlates the surface deformation response with the underground fracture dynamics, especially the energy-deformation coupling ratio. and vertical distance These characteristics directly reflect the spatiotemporal linkage between energy release and surface subsidence during the development of the abscess. This enables machine learning models to learn the intrinsic physical laws of abscess evolution, rather than the superficial statistical correlations, thereby achieving mechanistic diagnosis of the evolutionary stages.

[0115] S42. Definition of Evolutionary Stages: Based on rock mechanics theory, delamination evolution is divided into three stages and assigned quantitative indicators:

[0116] Stage I (Generation Period): Deformation Rate < 5mm / month, absolute value of deformation acceleration < 0.5 mm / month²; frequency of microseismic events < 2 times / day, b-value > 1.0, energy release rate < 1×10 8 Joules per day;

[0117] Phase II (Expansion Period): 5 mm / month ≤ ≤ 15 mm / month, 0.5 mm / month² ≤ ≤ 2.0 mm / month²; 2 times / day ≤ ≤ 10 times / day, 0.8 ≤ b value ≤ 1.0, 1×10 8 Joules / day ≤ ≤ 1×10 9 Joules per day;

[0118] Phase III (Pre-connection stage): > 15mm / month, > 2.0 mm / month²; > 10 times / day, b-value < 0.8 > 1×10 9 Joules per day, and > 60%.

[0119] S43. Intelligent identification model training and inference, implemented in the following ways:

[0120] S431, Feature Grouping Identifier: Divide the features in the fused feature vector F into macroscopic deformation response groups ( ) and micro-fracture driving group ( ), and use this grouping information as metadata input during model training;

[0121] This grouping identifier injects a priori physical knowledge structure into the identification model. During training, the identification model can not only learn the overall experience relationship between all features and targets, but also distinguish and understand the respective action modes and contributions of the two different physical meanings of feature groups, namely, drivers and responses, within its internal structure. This significantly improves the physical consistency and interpretability of the identification model's identification results and reduces logical fallacies caused by data noise.

[0122] S432. Physical Constraints on the Loss Function: In the standard cross-entropy loss function... Based on this, add a physical logic penalty term. This constitutes the total loss function. ;

[0123] The The model will " < 2mm / month and Samples with "< 0.5 times / day" are classified as Stage III, or samples with "b value < 0.7 and energy-deformation coupling ratio R increases by more than 50% per cycle" are classified as Stage I and penalized.

[0124] This design is the core link connecting data-driven and physics-driven approaches. Traditional loss functions only focus on the difference between the predicted label and the true label, while the physics penalty term... This directly supervises whether the model's prediction results conform to the basic laws of rock mechanics. It forces the model to follow basic physical common sense, such as the fact that low deformation and low microseismic activity cannot correspond to the period of instability and penetration, during the learning process. This solidifies the expert's domain knowledge into the algorithm, which constrains the "absurd" predictions that the model may produce from a mechanism perspective. This ensures that the output results can maintain physical rationality even in scenarios not covered by training data, thereby enhancing the model's generalization ability and reliability.

[0125] The model outputs two task results simultaneously: the main task is the classification probability of the evolutionary stage [P(I), P(II), P(III)]; the auxiliary task is the probability of entering stage III within a future window. .

[0126] S433. Utilize the confirmed evolutionary stages in historical data as the main task labels, and combine them with the derived... As auxiliary task labels, train the XGBoost model enhanced with physical mechanisms.

[0127] S434. During inference, the fused feature vector generated in real time will be... By inputting the trained model, a comprehensive diagnostic report can be obtained, including stage identification results and short-term risk probability; simultaneously, the system is set to consider the maximum probability of the main task. Only then is the identification result considered valid.

[0128] S5. The grouting timing decision and early warning output module processes the identification results of the delamination evolution stage to generate grouting decision and early warning information.

[0129] The specific processing procedure of S5 includes:

[0130] S51. Setting grouting timing decision rules:

[0131] An early warning for optimal grouting timing will be triggered when any of the following conditions are met:

[0132] Condition 1: The identification result is Stage II, and the deformation acceleration... Two consecutive analysis periods showed positive values, and the energy release rate of the microseismic cluster was also positive. It exceeds 1.5 times the standard deviation of its historical trend value;

[0133] Condition 2: The identification result is Stage III, and the auxiliary task output is... > 0.7;

[0134] This decision-making rule is based on a deep understanding of the microscopic mechanisms of delamination evolution. Condition 1 (acceleration and energy surge in Stage II) captures the "critical point" where the delamination transitions from stable expansion to unstable connection. At this point, microcracks within the rock mass are fully developed and begin to coordinate and network, leading to a sharp increase in energy release. This is the window of opportunity with the lowest intervention cost and the best results. Condition 2 (high conversion probability in Stage III) provides a bottom line for dealing with emergency situations that are nearing instability. The combination of these two conditions, in principle, enables precise targeting of the critical risk range from the "end of delamination expansion" to the "precursor to connection," transforming grouting timing from passive, experiential, and periodic construction to proactive and precise intervention based on the identification of the system's inherent critical state.

[0135] S52. Early Warning Information Generation and Push: Once an early warning is triggered, the system automatically generates a structured early warning report containing the location of the risk area, the evolution stage, the confidence level, key evidence diagrams, and decision recommendations, and pushes it to the production safety manager within 10 minutes via SMS, email, and platform alarms through the message interface. Example

[0136] Taking a coal mine working face as an example, the analysis period is selected from April 12 to May 30, 2025, with a window sliding step of 12 days. The delamination evolution state identification and grouting timing decision are performed on the 500m×500m analysis unit E52 within the working face. Using key process data from May 18 to 30, 2025 as an example, the steps for delamination evolution state identification and grouting timing decision are explained as follows:

[0137] S1. Synchronous acquisition and preprocessing of multi-source monitoring data:

[0138] S11. Acquire time-series SAR images from Sentinel-1 satellite covering the working area, and simultaneously acquire continuous waveform data collected by 30 microseismic sensors downhole; unify the spatiotemporal reference of all data to the WGS84 UTM 49N coordinate system and Beijing time;

[0139] S12. Using small baseline set interferometric radar technology to process SAR images, deformation time series were obtained, and the average coherence coefficient γ of the unit was calculated to be 0.72. The InSAR data quality score was also calculated. =0.72;

[0140] S13. Fine positioning processing of microseismic data: (1) The first arrival times of P-waves and S-waves are automatically picked up using the Akaike information criterion algorithm; (2) Initial positioning is performed using the Geiger method to obtain the preliminary location parameters of the events; (3) Fine positioning is performed using the double difference positioning method, with the event pair spacing threshold set to 15 meters, the convergence threshold set to 0.01 seconds, and the maximum number of iterations set to 100; (4) The positioning error is controlled within 50 meters by iteratively solving the least squares QR decomposition algorithm.

[0141] Based on the precise positioning results, the microseismic data quality score was calculated, yielding an average positioning residual of 0.08 seconds and an average signal-to-noise ratio of 28 dB. =0.75;

[0142] S14. Calculate the collaborative control factor for data quality:

[0143] ;because >0.6, the data quality meets the requirements, and a spatiotemporally synchronized surface deformation and downhole microseismic dataset of microseismic events and InSAR pixels is established.

[0144] S2. Perform InSAR deformation field spatiotemporal feature extraction:

[0145] S21. The average deformation rate of the unit is calculated. =9.5mm / month, deformation acceleration =1.2mm / month²;

[0146] S22, Calculation of deformation space gradient modulus:

[0147] (1) Based on the deformation rate field of the analysis window, the Sobel operator is used to calculate the deformation gradient components of each pixel in the east-west and north-south directions respectively. and ;

[0148] (2) Calculate the deformation spatial gradient modulus of each pixel. ;

[0149] (3) Statistically analyze the average value of the gradient modulus of all pixels in cell E52 to obtain the representative deformation gradient modulus of the cell. = 0.15;

[0150] S23. Identify the boundary of the settling funnel using a density-based spatial clustering algorithm: (1) Convert the deformation rate field into point set data, set the neighborhood radius eps=50 meters, and the minimum number of samples min_samples=10; (2) Identify the core points and form an initial cluster; (3) Select the cluster with the largest average deformation rate as the main settling funnel; (4) Use the Alpha Shape algorithm to extract the boundary polygons, and set the Alpha parameter to 75 meters; (5) Calculate the expansion rate by comparing the position changes of adjacent periodic boundary polygons. =3.2m / month.

[0151] S3. Perform multi-parameter analysis of microseismic event clusters:

[0152] S31. Identify the main microseismic clusters using the OPTICS algorithm: (1) Construct a four-dimensional feature space, including three spatial dimensions of longitude, latitude, and depth, and a time dimension, with the time scaling factor set to 100; (2) Set clustering parameters: maximum neighborhood radius ξ = 500 meters, minimum number of samples min_samples = 5; (3) Calculate the core distance and reachability distance, and generate a reachability map; (4) Extract spatiotemporal clusters based on slope, and select the cluster with the most events and the largest total released energy as the main analysis object cluster A;

[0153] S32. Calculate the characteristic parameters of cluster A:

[0154] (1) Energy release rate Calculate for each event Summing and dividing by the analysis window duration of 12 days yields... Joules per day;

[0155] (2) Event frequency : Divide the total number of events by the analysis window duration to obtain =6 times / day;

[0156] (3) Calculation of b value: The maximum likelihood method is adopted, based on the Gutenberg-Richard law, using the formula ,in Taking 0.1, we get b = 0.85;

[0157] (4) Proportion of tensile fractures For events with a moment magnitude greater than 1.0, focal mechanism inversion was performed, and the proportion of events with a P-axis dip angle greater than 45° was statistically analyzed. ;

[0158] (5) Centroid depth Calculate the arithmetic mean of the depths of all events to obtain ;

[0159] (6) Vertical distance Vertical distance between the centroid depth of the microseismic cluster and the geometric center of the InSAR settlement funnel =15;

[0160] (7) Energy-deformation coupling ratio : ,get .

[0161] S4. Perform intelligent identification of the off-layer collaborative evolution stage:

[0162] S41. Construct the deep fusion feature vector F:

[0163] ;

[0164] S42. Input the feature vector into the pre-trained dual-task XGBoost model;

[0165] The training process of the XGBoost model includes:

[0166] (1) Collect 850 historical samples, each containing an eleven-dimensional feature vector and a confirmed stage label;

[0167] (2) Label each sample with a risk label based on whether a Phase III transition occurs within the next 12 days;

[0168] (3) The features are divided into a macroscopic deformation response group and a microscopic fracture driving group;

[0169] (4) Set model parameters: learning rate 0.1, maximum tree depth 8, subsampling ratio 0.9;

[0170] (5) Use a customized loss function ,in A value of 0.5 is used to penalize predictions that violate the laws of rock mechanics;

[0171] S43. The model output results are: stage probabilities P(I) = 0.12, P(II) = 0.83, P(III) = 0.05; risk probabilities =0.28; Based on the principle of maximum probability, during the analysis window from May 18 to 30, 2025, the 500m×500m analysis unit of E52 is determined to be in Phase II (Extended Phase).

[0172] S5. Decision-making and early warning output for grouting timing:

[0173] S51, Check Decision Rule Condition 1: Deformation acceleration is positive for two consecutive cycles, current cycle =1.2mm / month², previous cycle =0.8 mm / month²; Energy release rate 5.8 × 10⁻⁶ 8 The joules / day exceeds the historical trend value by 1.92 standard deviations and is greater than 1.5 times the threshold; all requirements of decision rule condition one are met, and the system triggers an early warning for the optimal grouting timing.

[0174] S52. Automatically generate early warning reports, including: location of the risk area, current evolution stage and confidence level, risk probability value, key evidence map and decision recommendations; push the early warning report to the person in charge of production safety within 10 minutes through the message interface of the mining area safety monitoring platform.

[0175] To clearly demonstrate the dynamic decision-making process of the system within a complete analysis cycle, the key process data of the four analysis windows of the analysis unit within one analysis cycle are listed as follows: Figure 3 As shown.

[0176] To objectively and quantitatively demonstrate the technological advancements of this invention, based on retrospective analysis of numerous historical cases and the application effects of this embodiment, the key performance indicators of the method of this invention are now compared with three typical traditional methods. Specific data are as follows: Figure 4 As shown, by Figure 4 Data comparison shows that this invention significantly outperforms any single technical path or experience-based decision-making model across the entire "decision-execution-verification" chain of delamination grouting treatment. Specifically:

[0177] (1) In terms of decision reliability, this invention improves the early warning accuracy to 96%, which is far higher than various traditional methods. Although the accuracy of the single InSAR data method is 78%, it is sensitive to surface deformation, but it cannot detect the precursors of underground rock mass rupture, and often leads to false alarms due to misjudging non-delamination deformation; although the accuracy of the single microseismic data method is 72%, it can directly capture rupture, but it is difficult to distinguish delamination rupture from other mining stress rupture, and false alarms coexist; the accuracy of the manual experience method is 63%, which relies heavily on personal judgment, with large fluctuations and poor reproducibility. This invention, based on deep fusion identification of space and ground, realizes cross-verification and accurate interpretation of delamination evolution signals.

[0178] (2) In terms of direct engineering effects, this invention increases the deformation convergence rate after grouting to 88%. The convergence rate of the single InSAR data method is 52%, which is poor because the early warning time is too late and the rock structure has been damaged. The convergence rate of the single microseismic data method is 37%, which is not possible to accurately assess the surface response and the timing of grouting is often out of sync with macroscopic deformation control. The convergence rate of the manual experience method is 48%, which is unstable due to the strong randomness of the intervention time. This invention intervenes at the critical point where the delamination is about to penetrate, and the grout can effectively fill and support the rock layer, thereby quickly curbing surface subsidence.

[0179] (3) In terms of technical and economic efficiency, this invention significantly reduces the repeated grouting rate to 8%. The repeated grouting rates of the three traditional methods are 48%, 58%, and 52%, respectively. Due to their respective blind spots and decision-making biases, they are all unable to solve the problem accurately in one go, and generally fall into a vicious cycle of grouting-failure-regrouting. This invention achieves fundamental treatment with a single grouting by accurately locking the optimal intervention window, resulting in extremely significant economic benefits.

[0180] (4) In terms of the scientific mechanism of treatment, this invention achieves rapid attenuation of microseismic events after grouting, with a weekly attenuation rate of 78%. The 28% attenuation rate of the single InSAR data method is completely undetectable and unverifiable; although the 46% attenuation rate of the single microseismic data method can be monitored, the timing is inappropriate, and grouting is difficult to effectively curb the fracturing process; the 35% attenuation rate of the manual empirical method has uncertain effects. The rapid cessation of microseismic activity proves that the intervention of this invention can not only stabilize macroscopic deformation, but also fundamentally curb the generation and development of micro-fractures in the rock mass, achieving both symptomatic and radical treatment.

[0181] The delamination grouting timing decision system and method described in this invention achieves accurate identification of the delamination evolution stage and scientific decision-making on grouting timing through multi-source data collaborative control, deep fusion feature extraction, and dual-task intelligent identification model, providing effective technical support for mine disaster prevention and control.

[0182] The embodiments described above are merely examples of the present invention and are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart decision-making system for the timing of delamination grouting based on space-ground coordination, characterized in that: Includes the following modules: The multi-source monitoring data synchronous acquisition and preprocessing module is used to synchronously acquire, spatiotemporally register, and control the data quality of input time-series SAR images and downhole microseismic monitoring data. The data quality control includes: calculating a data quality collaborative control factor for each analysis unit, which is calculated by a preset weighted combination formula from the InSAR data quality score and the microseismic data quality score; when the data quality collaborative control factor is lower than a preset threshold, a data compensation strategy is triggered until the data quality collaborative control factor meets the requirements. The InSAR deformation field spatiotemporal feature extraction module is used to extract key deformation features characterizing delamination evolution from time-series SAR image data. The key deformation features include at least: monthly deformation acceleration calculated by performing a second difference on the deformation time series of each pixel, and the subsidence funnel expansion rate identified by spatial clustering analysis of the deformation rate field. The microseismic event cluster multi-parameter analysis module is used to identify event clusters from downhole microseismic monitoring data and extract key parameters characterizing the rock mass fracture process. The key parameters include at least: microseismic event frequency, cluster energy release rate, b-value calculated based on GR relationship, and tensile fracture proportion. The intelligent identification module for the co-evolution stage of delamination is used to construct a deeply fused feature vector for each analysis unit. The feature vector includes at least the deformation acceleration, settlement funnel expansion rate, microseismic event frequency, cluster energy release rate, b-value, tensile fracture ratio, vertical distance between the centroid depth of the microseismic cluster and the geometric center of the InSAR settlement funnel, and energy-deformation coupling ratio. Based on the feature vector, a machine learning model enhanced with physical mechanisms automatically identifies the current evolution stage of the delamination. The physical mechanism enhancement includes: dividing the features in the feature vector into a macroscopic deformation response group and a microscopic fracture driving group as metadata input during model training; adding a physical logic penalty term to the total loss function to penalize prediction results that violate the basic laws of rock mechanics; and the physical logic penalty term supervising whether the model prediction results conform to the basic laws of rock mechanics. The grouting timing decision and early warning output module automatically generates grouting timing suggestions and triggers early warnings based on the identified delamination evolution stage and according to preset multi-condition decision rules that combine deformation acceleration trends and microseismic energy anomalies.

2. A smart decision-making method for the timing of delamination grouting based on space-ground coordination, characterized in that: Includes the following steps: S1. The input time-series SAR images and downhole microseismic continuous waveform data are processed by the multi-source monitoring data synchronous acquisition and preprocessing module to obtain a spatiotemporally synchronized surface deformation and downhole microseismic dataset; the data processing includes calculating the data quality collaborative control factor DQF for each analysis unit, the data quality collaborative control factor being determined by the InSAR data quality score Q. insar and microseismic data quality score Q ms The data quality collaborative control factor (DQF) is calculated using a preset weighted combination formula. When the DQF is lower than the preset threshold, a data compensation strategy is triggered until the DQF meets the requirements. The calculation formula for the Data Quality Co-control Factor (DQF) is defined as follows: ; S2. The dataset is processed by the InSAR deformation field spatiotemporal feature extraction module to calculate the monthly deformation acceleration of each pixel and identify the boundary of the settlement funnel to calculate its expansion rate. S3. The dataset is processed by the microseismic event cluster multi-parameter analysis module, including microseismic event cluster identification, counting the frequency of microseismic events in the cluster, calculating the energy release rate of the cluster, calculating the b value based on the GR relationship, and analyzing the source mechanism to obtain the proportion of tensional rupture. S4. Through the intelligent identification module of the delamination co-evolution stage, a deep fusion feature vector is constructed for each analysis unit. The feature vector includes at least deformation acceleration, settlement funnel expansion rate, microseismic event frequency, cluster energy release rate, b-value, tensile rupture ratio, and the vertical distance D between the centroid depth of the microseismic cluster and the geometric center of the InSAR settlement funnel. v The energy-deformation coupling ratio R; the feature vector is input into a machine learning model enhanced with physical mechanisms. During the training process of the machine learning model, the features in the feature vector F are divided into a macroscopic deformation response group and a microscopic fracture driving group as metadata input, and a physical logic penalty term L is used. physics Total loss function L total The physical logic penalty term L physics To monitor whether the model prediction results conform to the basic laws of rock mechanics; The formula for calculating the energy-deformation coupling ratio R is: ; in, For microseismic energy release rate, It is the deformation acceleration; The total loss function L total The calculation formula is: ; Among them, L ce The standard cross-entropy loss function; S5. Based on the identification results, the grouting timing decision and early warning output module generates grouting decision and early warning information when the decision rules are met, including "the identification result is stage II and the deformation acceleration is positive for two consecutive cycles and the energy release rate exceeds 1.5 times the standard deviation of its historical trend value" or "the identification result is stage III and the short-term risk probability output by the auxiliary task is > 0.7".

3. The intelligent decision-making method for the timing of delamination grouting based on ground-sea coordination as described in claim 2, characterized in that: The specific processing procedure of S1 includes: S11. Spatiotemporal reference unification: Unify the spatiotemporal reference of all data from time-series SAR images and downhole microseismic monitoring to the same coordinate system and Beijing time, and set a unified analysis time window and window sliding step size; S12. Temporal SAR Image Data Preprocessing: Temporal SAR images are processed using small baseline set interferometric radar technology to obtain the deformation time series of the study area and output the deformation rate of each pixel. Calculate the average coherence coefficient γ for each pixel; S13. Microseismic data preprocessing: The continuous waveform data collected by the downhole microseismic monitoring network is processed for event triggering, P-wave and S-wave first arrival picking, source location and magnitude calculation. The double-difference location method is used to accurately locate microseismic events, and the location error is controlled within 50 meters. At the same time, the location residual and signal-to-noise ratio of each event are recorded. S14. Data Quality Collaborative Control: For each analysis unit, a data quality collaborative control mechanism is implemented, including: Calculate InSAR data quality score , which is the arithmetic mean of the average coherence coefficient γ of all pixels in the unit; Calculate the quality fraction of microseismic data The calculation formula is as follows: in, For the first Location residuals of individual microseismic events Its signal-to-noise ratio. This represents the total number of microseismic events within the unit. The data quality collaborative control factor (DQF) of the analysis unit is set to a first threshold. ;when When this happens, the system automatically triggers a data quality warning and executes a data compensation strategy. This strategy includes extending the analysis window or using weighted supplementary calculations based on historical high-quality data from the previous three periods, until... ; S15. Data Fusion Interface Construction: The spatiotemporal coordinates of microseismic events controlled by quality are associated with the spatiotemporal coordinates of InSAR pixels to construct a spatiotemporally synchronized dataset of surface deformation and downhole microseismic events.

4. The intelligent decision-making method for the timing of delamination grouting based on ground-sea coordination as described in claim 2, characterized in that: The specific processing procedure of S2 includes: S21. Deformation Acceleration Calculation: Perform a quadratic difference on the deformation time series of each pixel to calculate its monthly deformation acceleration. ; S22. Calculation of Deformation Spatial Gradient and Aggregation Characteristics: Calculation of the spatial gradient modulus of the deformation rate field. The system identifies abrupt deformation zones, uses the DBSCAN spatial clustering algorithm to identify the boundaries of the settling funnel, and calculates the expansion rate of the settling funnel. .

5. The intelligent decision-making method for timing of delamination grouting based on ground-sea coordination as described in claim 2, characterized in that: The specific processing procedure of S3 includes: S31. Microseismic event cluster identification: Based on the spatiotemporal density of microseismic events, the OPTICS clustering algorithm is used to identify microseismic event clusters by setting parameters such as minimum number of samples and maximum neighborhood radius. S32. Cluster Average Depth Calculation: For each microseismic event cluster, calculate the arithmetic mean of the depths of all its events, which is taken as the cluster average depth. ; S33. Microseismic Event Frequency Statistics: Count the total number of events for each cluster within the analysis time window, divide by the window duration, and obtain the frequency of microseismic events per unit time. ; S34. Cluster Energy Release Rate Calculation: For each cluster of microseismic events, calculate its cumulative microseismic energy release rate per unit time. The calculation formula is: ,in For the first in the cluster The magnitude of the event To analyze window duration; S35, b-value spatiotemporal scanning: For each micro-event seismic cluster, the b-value is calculated using the maximum likelihood method based on the GR relationship, and the calculation window length and step size are set; S36. Focal Mechanism and Rupture Type Analysis: Focal mechanism inversion is performed on microseismic events with a moment magnitude greater than 1.0 within the microseismic event cluster, and the proportion of extensional rupture events within the cluster is calculated. .

6. The intelligent decision-making method for the timing of delamination grouting based on ground-sea coordination as described in claim 2, characterized in that: The specific processing procedure of S4 includes: S41. Construction of Deep Fusion Feature Vector: For each analysis unit, construct an eleven-dimensional deep fusion feature vector F. The elements of vector F include: deformation rate. Deformation acceleration Deformation gradient modulus , rate of expansion of settling funnel Microseismic energy release rate Frequency of microseismic events b-value, average cluster depth The proportion of tensile rupture The depth of the centroid of the microseismic cluster and the vertical distance between the geometric center of the InSAR settlement funnel on the projection plane. Energy-deformation coupling ratio ; S42. Definition of Evolution Stages: Based on rock mechanics theory, delamination evolution is divided into three stages and assigned quantitative indicators: Stage I: Deformation Rate < 5mm / month, absolute value of deformation acceleration < 0.5 mm / month²; frequency of microseismic events <2 times / day, b-value > 1.0, energy release rate < 1×10 8 Joules per day; Phase II: 5 mm / month ≤ ≤ 15 mm / month, 0.5 mm / month² ≤ ≤ 2.0 mm / month²; 2 times / day ≤ ≤ 10 times / day, 0.8 ≤ b value ≤ 1.0, 1×10 8 Joules / day ≤ ≤ 1×10 9 Joules per day; Phase III: > 15mm / month, > 2.0 mm / month²; > 10 times / day, b-value < 0.8 > 1×10 9 Joules per day, and > 60%; S43, Intelligent identification model training and inference.

7. The intelligent decision-making method for the timing of delamination grouting based on ground-sea coordination as described in claim 6, characterized in that: The specific implementation of S43 includes: S431, Feature Grouping Identifier: Divide the features in the fused feature vector F into macroscopic deformation response groups, including... With micro-fracture driving group including And this grouping information is used as metadata input during model training; S432, The physical constraint for the loss function is the total loss function. ; The The model will " < 2mm / month and Samples with "< 0.5 times / day" are classified as Stage III, or samples with "b value < 0.7 and energy-deformation coupling ratio R increases by more than 50% per cycle" are classified as Stage I and penalized. The model outputs two task results simultaneously: the main task is the classification probability of the exoplane evolution stage [P(I), P(II), P(III)]; the auxiliary task is the probability of entering stage III within a future window. ; S433. Utilize the confirmed evolutionary stages in historical data as the main task labels, and combine them with the derived... As auxiliary task labels, train the XGBoost model enhanced with physical mechanisms; S434. During inference, the fused feature vector generated in real time will be used. Inputting the trained model yields a comprehensive diagnostic report, including stage-specific assessment results and short-term risk probabilities; simultaneously, the system is configured to determine the maximum probability of the main task. Only then is the identification result considered valid.

8. The intelligent decision-making method for the timing of delamination grouting based on ground-sea coordination as described in claim 2, characterized in that: The specific processing procedure of S5 includes: S51. Setting grouting timing decision rules: An early warning for optimal grouting timing will be triggered when any of the following conditions are met: Condition 1: The identification result is Stage II, and the deformation acceleration... Two consecutive analysis periods showed positive values, and the energy release rate of the microseismic cluster was also positive. It exceeds 1.5 times the standard deviation of its historical trend value; Condition 2: The identification result is Stage III, and the auxiliary task output is... > 0.7; S52. Early Warning Information Generation and Push: Once an early warning is triggered, the system automatically generates a structured early warning report containing the location of the risk area, the stage of evolution, the confidence level, key evidence diagrams, and decision-making suggestions, and pushes it to the production safety manager via SMS, email, and platform alarms through the message interface.

Images