Intelligent identification and grading early warning method, system and equipment based on mine production abnormity

Abstract

The invention discloses an intelligent identification and grading early warning method based on mine production abnormity, and a system and equipment thereof. The method comprises the following steps: obtaining space-time evolution data, earth surface deformation data and environment monitoring data of rock mass stress in a target area; performing abnormal mode detection on the data by using a space-time convolutional network to obtain accident risk probability distribution data; using an ultra wide band positioning system to track the movement track of the person entering the target area in real time to obtain the movement track data of the person; and constructing a three-dimensional early warning index system in combination with the accident risk probability distribution data in the target area and a dynamic early warning mechanism, performing dynamic early warning grading on different positions in the target area by using the three-dimensional early warning index system, and generating a grading response strategy matched with the personnel activity trajectory data. According to the invention, the problems of incapability of tracking a personnel movement track in real time and low timeliness of mine production abnormity early warning are solved.

Description

Technical Field

[0001] This invention relates to the field of mine production early warning technology, and more specifically, to a method, system and equipment for intelligent identification and graded early warning of mine production anomalies. The patent title of this invention may also be "Method and System for Intelligent Identification and Graded Early Warning of Mine Production Anomalies". Background Technology

[0002] Mining, as a core link in the nation's energy and mineral resource supply, occupies a pivotal position in the industrial system. Its production environment is complex and ever-changing, involving multiple critical aspects such as underground mining, tunnel support, equipment operation, and ventilation and drainage. Any anomaly in any of these areas can trigger safety accidents, threatening not only the lives of workers but also causing significant economic losses and environmental impacts. Therefore, timely and accurate identification and early warning of anomalies in mining production processes, along with effective response measures, are crucial for ensuring safe mining operations and improving production efficiency. However, current anomaly early warning and response technologies in the mining sector still face many unresolved issues. Among related technologies, there is a significant tendency to emphasize detection over response. Most existing early warning systems can only perform simple judgments of anomalies, outputting normal or abnormal binary alarm signals, failing to differentiate based on the risk level of the anomaly, such as accurately classifying low, medium, and high risk levels. This directly leads to the difficulty in matching different levels of risk with corresponding differentiated response plans in practical applications, such as partial equipment inspection in low-risk situations, temporary shutdown for troubleshooting in medium-risk situations, and emergency evacuation and equipment reinforcement in high-risk situations. This greatly reduces the practical value of early warning information and makes it unable to effectively guide on-site response work.

[0003] Prior art 1 (application number: 202411803404.4; application date: 2024.12.10) discloses a method for dynamic monitoring and early warning of the mine environment, referring to... Figure 1 As shown, the scheme includes the following steps: Multi-source data acquisition: collecting dynamic data of the mine environment through remote sensing imagery equipment, IoT sensors, and meteorological station equipment, including remote sensing imagery data, ground sensor data, and meteorological data; Data preprocessing: cleaning, temporally aligning, and unifying spatial reference coordinates of the data; Spatiotemporal data fusion: mapping multi-source data to a unified feature space through feature extraction, spatiotemporal alignment, and cross-modal fusion algorithms; Intelligent analysis and disaster prediction: predicting surface deformation, landslide risk, and tailings dam stability in the mining area based on a multi-task joint learning model; Risk assessment and early warning response: classifying risk levels based on comprehensive risk scores and triggering corresponding early warning measures and emergency response strategies. However, this scheme cannot track personnel activity trajectories in real time, resulting in low timeliness of early warnings for abnormal mine production.

[0004] Prior art 2 (Application No.: 202410824814.0; Application Date: 2024.06.25) discloses a method for predicting geological landslide displacement in mining areas based on radial basis function neural networks, including: acquiring time-series parameter data of the mining area; extracting statistical features of the time-series parameter data using a temporal convolutional network; inputting the statistical features into a trained radial basis function neural network model to obtain the geological landslide displacement in the mining area; the radial basis function neural network model is trained using the statistical features of historical time-series parameter data of the mining area as input and the historical cumulative landslide displacement of the mining area as output. This scheme cannot track personnel activity trajectories in real time, and the timeliness of early warning of abnormal mine production is not high.

[0005] Therefore, how to design a method for intelligent identification and hierarchical early warning of mine production anomalies that can track personnel activity trajectories in real time is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] In view of this, the present invention provides a method, system and equipment for intelligent identification and hierarchical early warning of mine production anomalies, in order to solve the problem of the inability to track personnel activity trajectories in real time and the low timeliness of early warning of mine production anomalies.

[0007] This application provides a method for intelligent identification and graded early warning of mine production anomalies, including the following steps: using a microseismic sensor array to acquire real-time spatiotemporal evolution data of rock mass stress within a target area; using InSAR satellites to acquire remote sensing data within the target area, and employing interferometric analysis to calculate surface deformation data from the remote sensing data; using a dust concentration sensor network, temperature and humidity sensors, and wind speed sensors to monitor the environment within the target area in real-time, obtaining environmental monitoring data; wherein, acquiring remote sensing data within the target area using InSAR satellites includes: periodically scanning the target area using InSAR satellites; the scanning cycle of the InSAR satellites is dynamically adjusted according to the satellite model, networking mode, and risk level of the mining area; in high-risk mining areas such as the working face advancement zone or old goaf area, when the InSAR satellite is a dual-satellite network, the scanning cycle is 3 days; when the InSAR satellite is a land exploration satellite in dual-satellite follow-fly mode, the scanning cycle is 4 days; when I When the nSAR satellite is a single-satellite land exploration satellite, the scanning cycle is 8 days. Anomaly pattern detection is performed on spatiotemporal evolution data, surface deformation data, and environmental monitoring data using a spatiotemporal convolutional network to obtain accident risk probability distribution data within the target area. The movement trajectories of personnel entering the target area are tracked in real time using an ultra-wideband positioning system to obtain personnel movement trajectory data. A three-dimensional early warning indicator system is constructed by combining the accident risk probability distribution data and dynamic early warning mechanism within the target area. This system is used to dynamically classify early warnings for different locations within the target area and generate graded response strategies that match the personnel movement trajectory data. The process of constructing a three-dimensional early warning indicator system by combining the accident risk probability distribution data and dynamic early warning mechanism within the target area, and using this system to dynamically classify early warnings for different locations within the target area and generate graded response strategies that match the personnel movement trajectory data, also includes fusing the processed data using Kalman filtering to obtain the risk decay rate. The expression for the risk decay rate is:

[0008]

[0009] Where, α t β is the risk diffusion damping coefficient, P0 is the initial baseline risk value, and β i γ represents the weight of the i-th treatment measure. i P represents the actual effectiveness coefficient of the i-th measure, where i ranges from 0 to 1, and n is the total number of measures implemented at the current moment; t Let P be the real-time risk assessment value within the target area at time t. t-1 The value represents the historical risk assessment value within the target area at time t-1, where time t is the current time and time t-1 is the time preceding the current time, which is one monitoring cycle apart.

[0010] Compared with existing technologies, the intelligent identification and hierarchical early warning method, system, and equipment for mine production anomalies provided by this invention achieve at least the following beneficial effects:

[0011] This invention provides a method, system, and equipment for intelligent identification and graded early warning of mine production anomalies. The method first utilizes a microseismic sensor array to acquire real-time spatiotemporal evolution data of rock mass stress within a target area; then, it uses InSAR satellites to acquire remote sensing data within the target area and employs interferometric analysis to calculate surface deformation data; finally, it uses a dust concentration sensor network, temperature and humidity sensors, and wind speed sensors to monitor the environment within the target area in real time, obtaining environmental monitoring data. Next, it uses a spatiotemporal convolutional network to perform anomaly pattern detection on the spatiotemporal evolution data, surface deformation data, and environmental monitoring data, obtaining accident risk probability distribution data within the target area. Then… This approach utilizes an ultra-wideband positioning system to track the movement trajectories of personnel entering the target area in real time, obtaining personnel movement trajectory data. Finally, by combining the accident risk probability distribution data within the target area with a dynamic early warning mechanism, a three-dimensional early warning indicator system is constructed. This system is then used to dynamically classify early warnings for different locations within the target area and generate graded response strategies that match the personnel movement trajectory data. By adopting this approach, the efficiency of mine production anomaly diagnosis can be improved, the movement trajectories of personnel entering the target area can be tracked in real time, the possibility of accidents caused by human violations can be reduced, and the overall safety management level of the mine production site can be improved. At the same time, differentiated response plans can be matched for different levels of risk, improving the timeliness of mine production anomaly early warning.

[0012] Of course, any product implementing this invention does not necessarily need to achieve all of the technical effects described above at the same time.

[0013] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description

[0014] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0015] Figure 1 This is a flowchart of the method for dynamic monitoring and early warning of the mine environment in existing technology 1;

[0016] Figure 2 This is a flowchart of a method for intelligent identification and hierarchical early warning of mine production anomalies provided by the present invention;

[0017] Figure 3This is a schematic diagram of the structure of a mine production anomaly intelligent identification and hierarchical early warning system provided by the present invention;

[0018] Figure 4 This is a schematic diagram of the structure of a mine production anomaly intelligent identification and hierarchical early warning device provided by the present invention. Detailed Implementation

[0019] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0020] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0021] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0022] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0023] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0024] Example 1

[0025] Reference Figure 2 , Figure 2 This is a flowchart of a method for intelligent identification and graded early warning of mine production anomalies provided by the present invention. This embodiment provides a method for intelligent identification and graded early warning of mine production anomalies, including the following steps:

[0026] S1: Real-time acquisition of spatiotemporal evolution data of rock mass stress within the target area using a microseismic sensor array; acquisition of remote sensing data within the target area using InSAR satellites, and calculation of surface deformation data using interferometric analysis methods; real-time monitoring of the environment within the target area using a dust concentration sensor network, temperature and humidity sensors, and wind speed sensors to obtain environmental monitoring data.

[0027] Specifically, refer to Figure 2As shown, the microseismic sensor array is a monitoring network formed by multiple high-sensitivity vibration sensors arranged in a specific topology. It continuously captures minute vibration signals generated by rock masses under stress within the target mining area. These vibration signals are recorded and transmitted in real time, and after analysis and processing, spatiotemporal evolution data is obtained. The microseismic sensor array synchronously collects information such as the time, location, and energy of rock mass vibration under stress, tracking the distribution and changing trends of stress within the rock mass in real time. This provides basic data for judging rock strata stability and predicting potential geological hazards. These vibration sensors are typically installed in key locations on the roadway walls, inside boreholes, or on the surface. The topology can be selected from star, ring, tree, or improved layered cluster structures, etc., without specific limitations here. Optionally, the vibration sensors are powered by DC, with a frequency response range of 10Hz to 1000Hz, a temperature range of -40℃ to 100℃, and a power supply range of 12V to 24V. Further information about this vibration sensor can be found at the following link:

[0028] https: / / item.jd.com / 10098023595500.html#switch-sku.

[0029] Continue to refer to Figure 2As shown, InSAR satellites, or Synthetic Aperture Radar Interferometry Satellites, are space observation devices that use microwave remote sensing technology to acquire information on surface deformation. By transmitting radar signals and receiving the echoes reflected from the surface, they calculate the minute displacements of the surface by combining the phase difference between the two observations. This enables large-scale, high-precision monitoring of surface deformation and allows for accurate remote sensing observation of target areas, obtaining remote sensing data within those areas. In mining operations, InSAR satellites can periodically scan the surface of mining areas to capture subtle changes such as surface subsidence, uplift, or displacement caused by underground mining. This provides crucial data support for analyzing changes in rock mass structure and assessing the impact of mining on surface stability, thereby obtaining remote sensing data for the target area. The scanning cycle of InSAR satellites is dynamically set based on key factors in the actual mining production scenario. For example, key factors used to set the scanning cycle include, but are not limited to: the mining intensity of the target area (such as the frequency of underground tunneling and mining operations), the basic condition of rock mass stability (such as the development of rock fractures and historical deformation rates), and the accident risk level of mining operations (such as high-risk areas requiring more intensive monitoring to capture potential dangerous changes). In mining production scenarios, the scanning cycle of InSAR satellites is dynamically adjusted according to the satellite model, networking mode, and risk level of the mining area. Optionally, for high-risk mining areas (such as working face advancement areas and old goaf areas), the scanning cycle can be set to 3 days for satellite equipment with dual-satellite networking, achieving high-density scanning twice a week to capture millimeter-level surface subsidence. Alternatively, when land exploration satellites can adopt a dual-satellite follow-fly mode, the scanning cycle can be set to an ultra-short cycle of 4 days. In single-satellite mode, the scanning cycle can be set to 8 days to meet the hourly response requirements for sudden deformations such as landslides.

[0030] Continue to refer to Figure 2As shown, technicians use InSAR satellites to acquire remote sensing data of the target area. Through specialized interferometric analysis methods, they calculate the surface deformation within the target area, obtaining surface deformation data. This data includes subtle changes such as subsidence, uplift, or displacement, thus enabling multi-dimensional monitoring of rock mass stability. The dust concentration sensor network is a distributed monitoring system composed of numerous dust concentration sensors. These sensors are deployed in key areas such as underground mine roadways, working faces, and goafs, enabling real-time detection of dust gas concentrations in the surrounding environment and ensuring continuous tracking of potential dust accumulation risks. Dust concentration sensors transmit monitoring data to a ground control center via wired or wireless means. When the dust concentration exceeds a preset threshold, the distributed monitoring system immediately issues an alarm signal, reminding staff to take measures such as ventilation and shutdown to prevent safety accidents such as explosions and poisoning caused by dust accumulation. Optionally, the dust concentration sensor has a power supply voltage of 5V to 7V, an operating temperature range of -10℃ to 65℃, and a sensitivity of 0.5V / (0.1mg / m3). Further information about this dust concentration sensor can be found at the following link: https: / / ic-item.jd.com / 10085319324572.html. Temperature and humidity sensors are distributed at different locations in the production area to collect air temperature and humidity data at any time, providing data to determine whether the environment is suitable for production operations. Optionally, the temperature and humidity sensor has a temperature range of -40℃ to 80℃ and a humidity range of 0 to 99.9% RH. Further information about this sensor can be found at the following link: https: / / item.jd.com / 100056068484.html#switch-sku. The wind speed sensor is installed at key nodes in the roadway ventilation system to record airflow velocity in real time, assessing whether the ventilation effectively dilutes harmful gases and regulates air quality. Optionally, the wind speed sensor has a frequency range of 200Hz to 1000Hz and a current range of 4mA to 20mA. Further information about this sensor can be found at the following link: https: / / item.jd.com / 10129497388942.html. All data from the dust concentration sensor, temperature and humidity sensor, and wind speed sensor are collected uniformly to form comprehensive environmental monitoring data for the target area, providing detailed environmental information for subsequent anomaly identification and risk warning. It should be noted that in step S1 above, three types of data need to be acquired: spatiotemporal evolution data, surface deformation data, and environmental monitoring data. The order in which these three types of data are acquired does not affect each other and can be adjusted adaptively according to the actual situation. No specific restrictions are made here.

[0031] S2: Use a spatiotemporal convolutional network to perform anomaly pattern detection on spatiotemporal evolution data, surface deformation data and environmental monitoring data to obtain accident risk probability distribution data within the target area; it includes the following steps: S2-1: Project the spatiotemporal evolution data and surface deformation data onto the original geological spatiotemporal model to obtain the target geological spatiotemporal model.

[0032] For details, please refer to... Figure 2 As shown, the original geological spatiotemporal model is a geological spatiotemporal model in existing technology. A geological spatiotemporal model is used to describe and analyze the temporal and spatial changes of geological phenomena. It typically consists of several key components, including a spatial element model, a spatial entity model, and a spatiotemporal object model. These models work together to provide a structured descriptive framework for the complexity and variability of geological phenomena. For example, the spatial element model focuses on the basic constituent units of a disaster; the spatial entity model handles the spatial characteristics of disaster entities; and the spatiotemporal object model considers the dynamic nature of disaster changes over time and space. By projecting spatiotemporal evolution data and surface deformation data onto the original geological spatiotemporal model, the target geological spatiotemporal model can be obtained.

[0033] S2-2: Input environmental monitoring data into the gas diffusion physics model to obtain the gas diffusion dynamics model.

[0034] For details, please refer to... Figure 2 As shown, the physical model of gas diffusion is a theoretical tool used to describe and predict the diffusion behavior of gas molecules in space. Based on physical laws and mathematical equations, the physical model of gas diffusion can explain the process of gas molecules moving from a high concentration region to a low concentration region. The dynamic model of gas diffusion is used to describe the diffusion behavior of gas in space. The dynamic model of gas diffusion simulates the motion of gas molecules through mathematical equations, helping users understand how gas diffuses in different environments, and thus know how gas diffuses in a target area.

[0035] S2-3: Anomaly pattern detection is performed on the target geological spatiotemporal model and gas diffusion dynamics model using a spatiotemporal convolutional network to obtain accident risk probability distribution data within the target area; this includes the following steps:

[0036] S2-3-1: The STL algorithm is used in the target geological spatiotemporal model and gas diffusion dynamics model to separate the long-term cumulative characteristic data of rock movement in the target area according to the time variation.

[0037] For details, please refer to... Figure 2As shown, the STL (Standard Template Library) algorithm, a seasonal-trend decomposition algorithm based on locally weighted regression, is a method for decomposing time series data. It can break down complex time series data into three parts: a trend term, a seasonal term, and a residual term, thus clearly presenting the data's changing characteristics at different time scales. Its core principle lies in separating the components through an iterative process. The trend term reflects the long-term trend of the data and is unaffected by short-term fluctuations, such as the gradually accumulating displacement changes in rock movement data as mining progresses. The seasonal term captures the recurring fluctuation patterns of the data within a fixed period, such as the minor deformation of the rock mass affected by seasonal temperature changes. The residual term is the random disturbance remaining after removing the trend and seasonal terms, mainly caused by accidental factors. The STL algorithm uses locally weighted regression fitting (Loess) technology to locally smooth the data in each iteration step, gradually optimizing the estimation results of each component, ultimately achieving a stable decomposition effect. In mine rock movement analysis, the STL algorithm can effectively isolate short-term random disturbances (such as instantaneous displacement fluctuations caused by blasting vibrations) from rock movement time series, accurately extracting features reflecting the long-term deformation trend of the rock mass, and providing a reliable basis for judging the development law of rock strata fractures. Through the separated trend term, the cumulative process of rock mass displacement over time can be clearly observed. Combined with geological structural characteristics, it can more accurately predict changes in rock strata stability, laying a solid data foundation for geological accident risk assessment. Simultaneously, for rock movement data containing seasonal influences (such as periodic deformation caused by groundwater during the rainy season), the STL algorithm can identify such patterns through seasonal term decomposition, avoiding misjudging normal periodic fluctuations as abnormal risks and improving the accuracy of risk assessment.

[0038] Taking the rock movement monitoring data of the No. 5 stope of a copper mine as an example, assuming that the stope collects surface displacement observations once a day for a year, a time series of 365 data points is formed. The original rock movement data series is in millimeters. The calculation process of the STL algorithm for this series can be specifically described in the context of a copper mine as follows: First, data preparation and parameter initialization are performed, and the seasonal cycle, a core parameter of the STL algorithm, needs to be clearly defined. Considering the climate characteristics of the copper mine area, the rainy season has a periodic impact on rock mass stability every three months. Also, taking into account the mining rhythm, the seasonal cycle is set to 90 days, meaning that the rock movement data shows a repeating fluctuation trend every 90 days. Simultaneously, a smoothing window for local weighted regression is determined. When calculating the seasonal term, the window is set to 30 days to focus on fluctuations within the short-term cycle, and when calculating the trend term, the window is set to 120 days to cover long-term changes over a longer period, ensuring compatibility with the characteristics of long-term accumulation, short-term cycles, and random disturbances in copper mine rock movement. Subsequently, a multi-round iterative decomposition process is entered, iterating 10 to 15 times until the components stabilize. Each iteration calculates the seasonal term, trend term, and residual term in sequence. The first step is to calculate the seasonal term. First, the raw rock movement data is detrended. Assuming a preliminary trend estimate is already available for the current cycle, a simple moving average of the raw data can be used as the initial trend in the first iteration. This trend term is subtracted from the raw data to obtain the detrended data. The purpose is to eliminate interference from long-term cumulative effects and extract cyclical fluctuations separately. Next, the detrended data is divided into multiple seasonal subsequences based on a 90-day seasonal cycle. For example, data from days 1, 91, 181, and 271 are grouped into the first subsequence, and data from days 2, 92, 182, and 272 are grouped into the second subsequence, for a total of 90 subsequences. For each subsequence, local weighted regression is used for local smoothing. Taking the first subsequence containing 4 data points as an example, the smoothed value of day 1 is calculated by assigning a weight of 0.8 to the data of day 1 (which is closest to the current point), a weight of 0.5 to the data of day 91, and a weight of 0.2 to the data of day 181 and day 271. The weight decreases as the data becomes farther away. The smoothed curve of this subsequence is fitted by weighted regression. Then, the smoothed values ​​of all subsequences are concatenated in chronological order to obtain the initial seasonal term for the current round. Finally, the initial seasonal term is smoothed overall using local weighted regression to eliminate the discontinuities in the concatenation of subsequences and obtain the final seasonal term for this round. This seasonal term can clearly show the fluctuations every 90 days, such as the 0.5 to 1 mm higher rock movement volume due to groundwater infiltration during the rainy season from June to August compared to other months. The second step is to calculate the trend term. First, the seasonal term obtained in this round is subtracted from the original rock movement data to obtain deseasonal data, thereby isolating cyclical fluctuations and focusing on long-term changes. Local weighted regression is used to smooth the deseasonal data with a long window of 120 days.For example, when calculating the trend value for day 100, 120 points of deseasoned data from days 40 to 160 are selected. The data for day 100 is assigned the highest weight of 0.9, and the weights of the data at both ends are gradually reduced to 0.1. A smooth curve is fitted through weighted regression, and this curve is the trend term for this round. In this copper mine scenario, the trend term shows a continuous increase from January to December, with a displacement of 5 mm in January, increasing to 18 mm in June, and reaching 32 mm in December, accurately reflecting the long-term cumulative characteristics of rock mass displacement as the mining depth increases with the advancement of the mining area. The third step is to calculate the residual term by directly subtracting the seasonal term and the trend term for this round from the original rock movement data. The difference obtained is the residual term. This component corresponds to random disturbances. For example, on day 89, due to blasting operations in the mining area, the original data is 2 mm higher than the theoretical value of the trend term plus the seasonal term, and the residual term is 2 mm higher. For other days without special disturbances, the residual term mostly fluctuates within ±0.5 mm, which is a normal random error. After each iteration, it is necessary to determine whether the components are stable. The seasonal and trend terms of the current iteration are compared with those of the previous iteration. If the displacement change of all data points in the trend term is less than 0.1 mm, and the difference in the periodic fluctuation amplitude of the seasonal term is less than 0.05 mm, the decomposition is considered stable and iteration stops. If these criteria are not met, the weighting coefficients of the local weighted regression are adjusted based on the residual term of the current iteration. For example, the weights of outliers with large residuals are reduced to avoid their interference with subsequent calculations, and the next iteration begins. After the final iteration, three independent components are obtained: the trend term (long-term cumulative displacement), the seasonal term (90-day periodic fluctuation), and the residual term (random disturbance). This process successfully isolates short-term random disturbances (such as blasting effects) and periodic fluctuations (such as rainy season effects) from the original rock movement data, accurately extracting the trend term reflecting the long-term deformation of the rock mass. This step provides core data support for subsequent prediction of rock fracture density based on the long-term cumulative characteristics of rock movement. For example, by observing the growth rate of the trend term, which shows an average monthly increase of 2.25 mm in December compared to January, it can be predicted that the rock fracture density may increase by 15% in the next quarter, thereby calculating the probability of geological accidents. In this embodiment, the STL algorithm can effectively separate long-term trends from short-term fluctuations in time series, making it suitable for extracting the long-term cumulative characteristics of rock movement and providing a foundation for geological accident risk analysis. First, the STL algorithm is used to decompose the data in the target geological spatiotemporal model and gas diffusion dynamics model according to the time dimension, separating the long-term cumulative characteristics of rock movement in the target area and eliminating the interference of short-term random fluctuations. Based on these long-term cumulative characteristics, the formation and development laws of rock fractures are analyzed, their density change trends are predicted, and the probability distribution of geological accident risks at different locations is determined. The extraction of long-term rock movement characteristics by the STL algorithm makes geological accident risk assessment more closely aligned with the actual laws of rock strata changes.

[0039] S2-3-2: Based on the long-term cumulative characteristic data of rock movement in the target geological spatiotemporal model, predict the trend of rock fracture density change and obtain the probability distribution data of geological accident risk at different locations in the target area.

[0040] S2-3-3: The spatial aggregation of dust concentration in the gas diffusion dynamics model is predicted by Moran's index. The prediction results are obtained, and the dust concentration change characteristic data at different locations in the target area are divided according to the prediction results. The dust accident risk probability distribution data at different locations in the target area are obtained.

[0041] Specifically, Moran's I is a key tool in geostatistics used to quantify spatial autocorrelation. Proposed in 1950 by British statistician Patrick Alfred Pierce Moran, its core function is to reveal the distribution patterns of attribute values ​​among geographic spatial units, determining whether data exhibits clustering, randomness, or regularity. In this embodiment, Moran's I quantifies the degree of spatial data clustering, accurately identifying the spatial aggregation state of dust concentration and aiding in dust accident risk assessment. This embodiment uses Moran's I to analyze the spatial distribution of dust concentration in a gas diffusion dynamics model, determining the presence and degree of clustering, and thereby classifying the dust concentration variation characteristics at different locations to obtain the dust accident risk probability distribution. Moran's I analysis of dust spatial clustering makes dust accident risk assessment more accurate.

[0042] S2-3-4: Extract seasonal diurnal variation characteristic data and / or sudden gas parameter variation characteristic data within the target area from the gas diffusion dynamics model, and use the Bayesian change point detection algorithm to predict the environmental accident risk probability distribution data at different locations within the target area;

[0043] For details, please refer to... Figure 2As shown, the Bayesian change point detection algorithm is a time series analysis technique based on Bayesian statistical methods. It is used to identify potential "change points" in data—that is, points in time where the data generation mechanism undergoes significant changes. Its core idea is to recursively update the segmentation hypothesis, combine prior knowledge with observed data to calculate the posterior probability distribution, and ultimately quantify the probability and location of change points. The Bayesian change point detection algorithm can not only detect abrupt changes in statistical properties such as mean and variance, but also capture nonlinear trend changes or interruptions in periodic patterns. In this embodiment, the Bayesian change point detection algorithm excels at capturing abrupt changes in data, and can promptly detect abnormal changes in gas parameters for environmental accident risk assessment. This embodiment extracts the normal fluctuation characteristics of gas parameters caused by seasonal diurnal variations and the characteristics of possible sudden abnormal changes from the gas diffusion dynamics model. Using the Bayesian change point detection algorithm, it identifies anomalous points in these characteristics and predicts the probability distribution of environmental accident risks at different locations. The Bayesian change point detection algorithm's ability to capture abrupt changes in gas parameters enhances the timeliness of environmental accident risk prediction.

[0044] S2-3-5: Integrate the probability distribution data of geological accidents, dust accidents, and environmental accidents at different locations within the target area to obtain the probability distribution data of accidents within the target area.

[0045] For details, please refer to... Figure 2 As shown, integrating geological accident risk probability distribution data, dust accident risk probability distribution data, and environmental accident risk probability distribution data can take into account multiple factors and obtain more comprehensive overall accident risk probability distribution data for the target area. The integration of the above-mentioned multiple types of risk probability distribution data avoids the limitations of single risk assessment, can comprehensively reflect the risk status of the target area, provide a more scientific basis for subsequent early warning and disposal, and help to formulate more precise prevention and control measures to improve the safety of mine production.

[0046] Step S2 of this embodiment: Anomaly pattern detection is performed on spatiotemporal evolution data, surface deformation data, and environmental monitoring data using a spatiotemporal convolutional network to obtain accident risk probability distribution data within the target area. This process achieves accurate assessment and fusion of various accident risks through targeted processing of specific data using different algorithms. From a technical perspective, this approach improves the accuracy and relevance of various risk assessments, and the overall risk distribution after fusion is more comprehensive and reliable.

[0047] In principle, this process combines multi-source monitoring data with specialized models to accurately characterize mine accident risks. The target geological spatiotemporal model provides a unified spatial carrier for rock mass stress and surface deformation data, intuitively presenting the dynamic changes in geological structure; the gas diffusion dynamics model simulates the propagation of harmful gases such as dust under different environmental conditions based on the physical laws of gas diffusion. When combined, the spatiotemporal convolutional network can deeply mine the hidden spatiotemporal correlation features in the data, identify abnormal patterns deviating from normal states, and thus provide a scientific basis for calculating risk probability distributions, ensuring that risk assessment not only conforms to geological and physical laws but also captures complex dynamic changes. In practical implementation, firstly, the spatiotemporal evolution data of rock mass stress and surface deformation data are accurately mapped onto the three-dimensional target geological spatiotemporal model according to their actual spatial locations, making these data correspond to spatial information such as the mine's geological structure and tunnel distribution, forming a holistic description of geological dynamics. Next, combining real-time collected environmental monitoring data such as temperature, humidity, and wind speed, a gas diffusion dynamics model is constructed based on the physical characteristics of gas diffusion to simulate the flow, diffusion paths, and concentration distribution of gases such as dust in the tunnel. Then, a spatiotemporal convolutional network is used to jointly analyze the data from the target geological spatiotemporal model and the gas diffusion dynamics model, identifying abnormal patterns such as sudden changes in rock mass stress, abnormal surface deformation, and excessive dust concentration. These abnormal patterns are then transformed into accident risk probability distributions at different locations, clearly demonstrating the spatial distribution differences of risks within the target area. From a technical perspective, this approach improves the accuracy and spatial resolution of accident risk assessment. The target geological spatiotemporal model makes the spatial correlation of geological data clearer, facilitating the capture of the linkages between geological changes in different areas; the gas diffusion dynamics model makes the simulation of gas distribution more realistic, avoiding one-sided judgments about gas risks. Furthermore, the accurate identification of abnormal patterns by the spatiotemporal convolutional network effectively distinguishes between normal fluctuations and danger signals, reducing misjudgments. The resulting probability distribution of accident risks at different locations provides a precise basis for subsequent graded early warning and targeted response, helping staff to focus on high-risk areas, take preventive measures in advance, and reduce the possibility of accidents.

[0048] S2-4: Based on the topographic structure, mine functional zoning, and mine type of the target area, establish an accident chain simulation model matching the target area. Input the spatiotemporal evolution data, surface deformation data, and environmental monitoring data collected in real time into the accident chain simulation model to obtain spatiotemporal causal chain data at different locations within the target area.

[0049] For details, please refer to... Figure 2As shown, after obtaining the accident risk probability distribution data within the target area, the next step is to construct an adapted accident chain simulation model based on the actual conditions of the target area. This process fully considers the terrain structure of the target area, such as whether it is a steep mountainous mining area or a gentle hilly mining area, as different terrains have a direct impact on rock mass stability and water flow direction. It also considers the functional zoning of the mine, as different areas such as mining areas, transport roadways, and ventilation rooms have different operational characteristics and potential risks. Furthermore, the type of mine is also taken into consideration; whether it is a metal mine or a non-metal mine, the main types of risks and disaster-causing factors faced during production are different. The accident chain simulation model constructed by integrating these factors can better fit the actual production scenario of the target area, laying the foundation for subsequent causal relationship analysis.

[0050] Taking a copper open-pit mine as an example, firstly, an accident chain simulation model is constructed based on the actual scenario of the open-pit mine. The terrain is a gently sloping mountain, requiring a focus on the impact of the slope gradient on the stability of the slope and soil. The functional zones include mining operation areas, ore transfer stations, and drainage pumping stations; different zones have different risk focuses, such as the operation area focusing on slope safety and the transfer station focusing on equipment operation. The mine type is a metal mine, with the main risks concentrated on slope instability and equipment transfer failures. Based on these characteristics, the risk factors are first decomposed using the analytic hierarchy process (AHP), categorized into three types: geological disaster-causing factors (slope stability, slope angle), equipment operating parameters (transfer belt operation status, pump station motor load), and environmental monitoring indicators (mining precipitation, wind speed). Then, the importance of each type of factor is compared layer by layer. Because slope instability in open-pit mines can trigger large-scale safety accidents, slope stability has a higher weight than other geological disaster-causing factors. This is used to construct a multi-level accident chain network. The bottom layer consists of the aforementioned basic factors; the middle layer includes related nodes such as slope loosening, abnormal belt vibration, and excessive motor load; and the top layer clarifies the two core accident outcomes: "slope landslide" and "transfer interruption." Each node corresponds to a specific risk event or indicator status. Next, the Bayesian network probability inference stage is entered. After inputting the real-time collected monitoring data into the accident chain inference model, the initial state probability of the bottom-level basic factor nodes is first determined. For example, if slight displacement of the mining slope is detected, the initial probability of the "abnormal slope displacement" node is determined based on historical data. Furthermore, considering the influence of related factors, such as increased recent rainfall in the mining area exacerbating soil moisture and reducing stability, the trigger probability of mid-level related nodes is deduced through the probabilistic correlation relationships of a Bayesian network. The probability of the "slope soil loosening" node increases due to the combined effects of "abnormal slope soil displacement" and "increased rainfall." Further extrapolating upwards, the increased probability of "slope soil loosening" directly increases the risk probability of a "slope landslide" accident at the top level. Simultaneously, regarding equipment, if irregular vibration is detected during conveyor belt operation, the initial probability of the "abnormal belt vibration" node is first determined. Then, combined with the status of related nodes such as "pump station motor load," the probability of the "motor overload" node is deduced, further deriving the possible evolution path of a "transfer interruption" accident, clearly presenting the probability transmission process from initial factors to accident outcome. Subsequently, the node probabilities are dynamically updated using abnormal pattern features extracted by a spatiotemporal convolutional network.Spatiotemporal convolutional networks capture easily overlooked spatiotemporal anomalies from real-time monitoring data. For example, they might detect short-term abrupt changes in soil and rock displacement on the eastern slope of the mining area within a specific time period, and observe that the vibration amplitude of the conveyor belt during the morning peak transport period is significantly higher than at other times. These anomalies are fed back into the accident chain network to adjust the conditional probability tables of the corresponding nodes. The probability of the "slope soil and rock loosening" node is further increased due to the abrupt displacement change, and the probability of the "abnormal conveyor belt vibration" node is corrected based on the abnormal vibration during peak hours. This ensures that the state of each node in the network better reflects the real-time risk situation of the mining area, avoiding inference biases caused by static probabilities. Finally, the accident chain inference model integrates all inference results with spatiotemporal information to output a complete spatiotemporal causal chain. For instance, it might infer that within a certain time period, the eastern slope of the mining area first experiences increased soil and rock moisture due to increased precipitation, followed by abnormal soil and rock displacement, which in turn triggers soil and rock loosening, ultimately leading to a continuous increase in the risk of landslides. Meanwhile, during the morning peak transport period, the conveyor belt at the transfer station first experienced abnormal vibration due to excessive material load, which then led to overload of the pump station motor, gradually increasing the risk of transfer interruption. This causal chain clearly marks the temporal sequence and specific spatial location of each link, allowing staff to intuitively grasp the possible evolution path of accidents in different areas and providing clear guidance for developing targeted prevention and control measures. Subsequently, real-time collected data on the spatiotemporal evolution of rock mass stress, surface deformation, and environmental monitoring data will be uniformly input into the constructed accident chain simulation model. The accident chain simulation model will conduct in-depth analysis of these data, sort out the correlation between different data, and then deduce the spatiotemporal causal chain at different locations within the target area. For example, in a certain mining area, the accident chain simulation model may, through data analysis, derive a causal chain such as "the continuous increase in rock mass stress leads to significant surface deformation, which in turn causes abnormal stress in the local tunnel support structure. If the dust concentration in the environment increases and ventilation is insufficient at this time, it may exacerbate the risk of accident," and clearly define the temporal sequence and specific spatial distribution of these changes, allowing staff to clearly grasp the possible evolution path of the accident.

[0051] In an optional embodiment, it is assumed that the accident chain inference model consists of a multi-level accident chain network, which contains accident chain nodes. Based on this, in the above embodiment, the step of inputting real-time collected spatiotemporal evolution data, surface deformation data, and environmental monitoring data into the accident chain inference model to obtain spatiotemporal causal chains at different locations within the target area can be implemented as follows:

[0052] Based on spatiotemporal evolution data, surface deformation data, and environmental monitoring data, the Analytic Hierarchy Process (AHP) is used to determine the weight vectors of geological disaster-causing factors, equipment operating parameters, and environmental monitoring indicators, and a multi-level accident chain network is constructed. Through the joint probability distribution calculation of Bayesian networks, the evolution paths of various types of accidents are deduced in the multi-level accident chain network. Combining the abnormal pattern features extracted by spatiotemporal convolutional networks, the conditional probability table of accident chain nodes in the multi-level accident chain network is dynamically updated, and a causal chain sequence with timestamps and spatial coordinates is output as the spatiotemporal causal chain at different locations within the target area.

[0053] Among them, the evolution path of multiple types of accidents includes one of the following: microseismic activity to rock fissures expanding to rock mass protrusion, equipment vibration exceeding limits to component fatigue fracture to production interruption.

[0054] From a theoretical perspective, this method of obtaining spatiotemporal causal chains based on multi-level accident chain networks is fundamentally about quantifying the influence weights of different disaster-causing factors using scientific methods and combining this with probabilistic models to predict the accident development path. The Analytic Hierarchy Process (AHP) can decompose complex mine production risk factors into different levels, clearly distinguishing the importance of geological disaster-causing factors, equipment operating parameters, and environmental monitoring indicators in the accident evolution, providing a logical foundation for constructing a hierarchical accident chain network. Bayesian networks, leveraging probabilistic reasoning capabilities, capture the correlations between nodes in the accident chain, enabling the accident chain prediction model to dynamically determine the possible path from the initial trigger to the final occurrence of the accident based on real-time data. Furthermore, the extraction of abnormal patterns by spatiotemporal convolutional networks further enhances the accident chain prediction model's ability to perceive complex spatiotemporal features, making the state updates of accident chain nodes more closely reflect the dynamic changes in actual production. In terms of implementation, firstly, based on the Analytic Hierarchy Process (AHP), a layer-by-layer comparative analysis is conducted on various factors such as geology, equipment, and environment to determine the weight vector of each factor in the accident evolution. This establishes a multi-layered accident chain network containing bottom-level basic factors, mid-level related nodes, and top-level accident results, with each node corresponding to a specific risk indicator or event. Subsequently, various types of data collected in real time are input into a Bayesian network. By calculating the joint probability distribution between nodes, the complete evolution path of various types of accidents is deduced, such as "microseismic activity triggering the expansion of rock fissures, leading to rock mass outbursts" and "excessive equipment vibration causing fatigue fracture of components, ultimately leading to production interruption." Simultaneously, a spatiotemporal convolutional network is used to extract abnormal pattern features from real-time data, such as abrupt changes in rock mass stress and abnormal distributions of dust concentration. Based on this, the conditional probability tables of each node in the accident chain network are dynamically adjusted, making the node status more accurately reflect the current risk situation. The final output causal chain sequence includes the timestamp and spatial coordinates of each event, precisely corresponding to different locations in the target area. From a technical perspective, this approach improves the accuracy and relevance of accident evolution path prediction. By determining weights using the analytic hierarchy process (AHP), subjective judgments about the impact of risk factors are avoided, making the construction of the accident chain network more scientific. The probabilistic extrapolation capabilities of Bayesian networks enable the accident chain extrapolation model to clearly present the development path of different accident types, whether it's a rock outburst dominated by geological factors or a production interruption caused by equipment problems, a clear evolution path can be obtained. Furthermore, combining spatiotemporal convolutional networks to update node states gives the accident chain extrapolation model dynamic adaptability, enabling it to respond promptly to new changes in production and reduce misjudgments caused by data lag. The causal chain sequence with spatiotemporal information provides staff with precise risk positioning data, helping to quickly identify high-risk areas and key time points, laying a solid foundation for developing targeted prevention and control measures, thereby effectively improving the reliability and efficiency of mine production anomaly early warning and response.

[0055] S3: Use an ultra-wideband positioning system to track the movement trajectory of people entering the target area in real time and obtain personnel movement trajectory data.

[0056] For details, please refer to... Figure 2 As shown, an ultra-wideband (UWB) positioning system is a high-precision positioning solution based on wireless communication technology. It achieves communication and positioning functions by sending and receiving extremely narrow pulses at the nanosecond or microsecond level. Compared with traditional technologies (such as GPS, Bluetooth, and Wi-Fi), its core advantage lies in its extremely wide signal bandwidth, providing centimeter-level or even millimeter-level positioning accuracy in complex environments, while also possessing strong anti-interference capabilities and real-time performance. The UWB positioning system uses Time-of-Flight (TOF) ranging technology to achieve millimeter-level positioning of personnel, enabling real-time tracking of personnel movement trajectories within a target area and obtaining personnel movement trajectory data. TOF ranging technology involves emitting an extremely short light pulse to the object being measured and then receiving the light pulse reflected back from the object. By measuring the flight (round trip) time of the light pulse from transmission to reception, the distance from the transmitter to the object can be calculated. The core of this ranging method lies in using the flight time of the light pulse to accurately obtain distance information.

[0057] Furthermore, after tracking the movement trajectories of personnel entering the target area in real time, a 3D mining area geographic model within the target area can be used to calculate the spatial distance between the real-time location of personnel and the risk source. When the distance is less than a safety threshold, a near-field warning is triggered. Subsequently, the YOLOv8 target detection algorithm is used to identify personnel behavior and construct a behavioral risk assessment model. When personnel within the target area are detected not wearing safety equipment or illegally entering restricted areas, an alarm message is generated and pushed to the safety supervision platform.

[0058] In principle, real-time tracking of personnel activity trajectories and risk warning is achieved through the synergistic effect of high-precision positioning technology, spatial distance calculation, and intelligent behavior recognition, constructing a comprehensive personnel safety monitoring system. The UWB positioning system utilizes the propagation characteristics of electromagnetic waves to achieve precise positioning by measuring signal time of flight, providing fundamental data for personnel trajectory tracking. The 3D mining area geographic model provides a digital scenario for spatial distance calculation, accurately reflecting the relative positional relationship between personnel and risk sources. The YOLOv8 target detection algorithm, leveraging deep learning's ability to extract image features, can quickly identify abnormal personnel behavior, providing a basis for behavioral risk assessment. The combination of these three elements forms a complete logical chain from location tracking to risk assessment to warning triggering, ensuring the real-time nature and accuracy of personnel safety monitoring. In terms of specific implementation, the UWB positioning system is first deployed, with positioning base stations strategically placed within the target area. Personnel carry positioning tags, and the distance between the tag and the base station is obtained in real time using Time-of-Flight (TOF) ranging technology, thereby calculating the personnel's precise location and continuously recording location information to form an activity trajectory. Next, real-time personnel location data is imported into a 3D mining area geographic model. This model pre-marks the spatial coordinates of each risk source. The actual distance between personnel and risk sources is calculated using these coordinates. When the actual distance is less than a set safety threshold, the UWB positioning system automatically triggers a near-field warning, alerting personnel to stay away from dangerous areas. Simultaneously, cameras are installed at key locations in the mining area to capture images of personnel activities. The YOLOv8 target detection algorithm is used to analyze these images, identifying whether personnel are wearing safety equipment and have entered restricted areas. The identification results are input into a behavioral risk assessment model. Once a violation is detected, the model immediately generates an alarm and pushes it to the safety monitoring platform, facilitating timely intervention by supervisors. From a technical perspective, this method significantly improves the accuracy and response speed of mine personnel safety management. The millimeter-level positioning capability of the UWB positioning system allows for more accurate recording of personnel activity trajectories, providing a reliable foundation for subsequent risk assessment. Combined with spatial distance calculations from the 3D mining area geographic model, it enables timely detection of safety hazards between personnel and risk sources. The real-time triggering of near-field warnings effectively prevents personnel from accidentally entering dangerous areas. The YOLOv8 object detection algorithm's rapid identification of human behavior enables dangerous behaviors such as not wearing safety equipment and entering restricted areas in a timely manner. The real-time push of alarm information makes safety supervision more proactive, reduces the possibility of accidents caused by human violations, and improves the overall level of safety management at the mine production site.

[0059] S4: Construct a three-dimensional early warning indicator system by combining accident risk probability distribution data and dynamic early warning mechanism within the target area. Use the three-dimensional early warning indicator system to dynamically classify early warnings for different locations within the target area and generate a graded response strategy that matches personnel activity trajectory data.

[0060] Specifically, continue to combine Figure 2 As shown, the three-dimensional early warning indicator system includes potential casualties, economic losses, risk diffusion rate, and the number of related equipment. The weights of each indicator in the three-dimensional early warning indicator system are determined using the fuzzy hierarchical analysis method (FAHP), and the mapping and transformation between qualitative concepts and quantitative values ​​are realized through a cloud model. Based on the early warning classification matrix of the three-dimensional early warning indicator system, the probability distribution of accident risk is mapped to a multi-level early warning state, forming a dynamic early warning heat map to indicate the early warning state at different locations. The number of early warning state levels is obtained by coordinating the parameter scale of the early warning classification matrix and the risk type.

[0061] In principle, dynamic early warning grading for different locations within a target area achieves precise risk classification by constructing a multi-dimensional indicator system and combining scientific weight allocation and qualitative-quantitative conversion methods. The three-dimensional early warning indicator system covers indicators reflecting the severity of risk, such as the potential number of casualties and economic losses, as well as indicators reflecting the scope and urgency of risk impact, such as the risk diffusion rate and the number of associated equipment, comprehensively considering the potential consequences and development trends of an accident. The Fuzzy Analytic Hierarchy Process (FAHP) effectively handles the fuzzy relationships between indicators and rationally determines the weight of each indicator in the grading process; the cloud model solves the problem of converting qualitative descriptions to quantitative data, providing corresponding quantitative standards for qualitative concepts such as high risk and medium risk, and offering scientific logical support for early warning grading.

[0062] Taking the dynamic early warning classification requirement of a copper mine open-pit mine as an example, for the FAHP (Fast-Only Hierarchical Analysis) approach, the following steps are taken: First, a hierarchical structure adapted to the copper mine is constructed. The top layer is set to determine the weights of the three-dimensional early warning indicators for the copper mine open-pit mine to support the core objective of dynamic early warning classification. The middle layer, the criterion layer, corresponds to the four indicators clearly defined in the document: potential casualties (reflecting the harm to workers caused by accidents such as slope landslides), economic losses (reflecting the property impact of equipment damage or production interruption), risk diffusion rate (the spread speed of slope instability or dust diffusion in open-pit scenarios), and the number of associated equipment (the total number of equipment such as conveyor belts and crushers related to the risk area in the mine). The bottom layer does not require an additional scheme; the weights are directly derived from the relative importance of the indicators in the criterion layer to ensure that the hierarchy matches the risk association logic of the copper mine. Next, the process of constructing a fuzzy judgment matrix through pairwise comparisons is initiated. A review group composed of copper mine safety engineers, open-pit mine managers, and early warning algorithm personnel is invited to conduct pairwise importance judgments on the four indicators in conjunction with the actual risk scenarios of copper mine open-pit mining (such as the rapid spread of slope instability and the large associated impact due to the clustered operation of equipment). Because the risk diffusion rate in open-pit mining directly determines the speed at which the impact of an accident expands (e.g., rapid spread of small-scale slope loosening could trigger a landslide), the review panel determines that the risk diffusion rate is more important than the potential number of casualties, economic losses, and the number of associated equipment. The potential number of casualties, being directly related to personnel safety, is more important than economic losses and the number of associated equipment. When quantifying these judgments, instead of setting specific values, fuzzy numbers are used to represent the uncertainty of the judgments. For example, "risk diffusion rate is slightly more important than potential casualties" corresponds to a set of fuzzy numbers containing a lower bound, the most likely value, and an upper bound; "risk diffusion rate is significantly more important than the number of associated equipment" corresponds to another set of fuzzy numbers. For reverse comparisons, the reciprocal of the corresponding fuzzy number is taken, ultimately forming a 4×4 fuzzy judgment matrix. Each element in the matrix corresponds to a pair of fuzzy importance relationships between indicators. Subsequently, a consistency check of the fuzzy judgment matrix is ​​performed to avoid logical contradictions in the review panel's judgments. First, the fuzzy numbers in the matrix are converted into clear values ​​using a specific method, transforming the fuzzy matrix into a computable clear matrix. Then, the maximum eigenvalue of the clear matrix is ​​calculated, and a consistency index is calculated based on the number of indicators (4). Simultaneously, the standard values ​​of the random consistency index, which are compatible with the four indicators, are retrieved. The ratio of the consistency index to the random consistency index is used to determine if the matrix conforms to logic. If the ratio is less than a preset threshold, the review group's judgment is consistent, and the matrix is ​​valid. If the ratio exceeds the threshold, contradictory indicator pairs are traced back (e.g., the original judgment that the risk diffusion rate is more important than the potential number of casualties, and the potential number of casualties is more important than the number of associated equipment, but a misjudgment that the number of associated equipment is more important than the risk diffusion rate). The review group is then reorganized to adjust the judgment based on the actual risks of the copper open-pit mine until the matrix passes the consistency test. Finally, the indicator weights are calculated. Fuzzy operations are performed on each row of the fuzzy judgment matrix that has passed the test to obtain the fuzzy weight of each indicator.These ambiguous weights are then transformed into clear weight values ​​to eliminate ambiguity. Finally, all clear weight values ​​are normalized to ensure that the sum of the weights of the four indicators is 1. Considering the risk priority of copper open-pit mines, the normalized risk diffusion rate will have a higher weight than other indicators, followed by the potential number of casualties, and then the weights of economic losses and the number of associated equipment. The final weight results will be directly used to construct the three-dimensional early warning indicator system for the copper open-pit mine, providing a quantitative basis for mapping the probability distribution of subsequent accident risks to the early warning state.

[0063] In the specific implementation of the above optional embodiments, firstly, combining the accident risk probability distribution and dynamic early warning mechanism of the target area, the potential number of casualties, economic losses, risk diffusion rate, and number of associated equipment are selected as the core indicators of the three-dimensional early warning indicator system, clarifying the evaluation scope and standards of each indicator. Next, FAHP is used to compare each indicator pairwise, and a judgment matrix is ​​constructed to determine the weight of different indicators; for example, the risk diffusion rate may be given a higher weight in mine accidents. Subsequently, a cloud model is used to convert qualitative risk descriptions such as severe, moderate, and minor risks into specific numerical ranges, achieving an effective connection between qualitative and quantitative analysis. Then, according to the preset early warning classification matrix, the accident risk probability distribution is matched with the quantitative results of each indicator, mapping them to multi-level early warning states, ultimately forming a dynamic early warning heat map that intuitively displays the early warning status at different locations. The number of early warning levels is flexibly adjusted according to the parameter scale and risk type of the early warning classification matrix to adapt to the needs of different scenarios.

[0064] For example, the level of the early warning status can be flexibly adjusted according to the parameter size and risk type of the early warning classification matrix. Essentially, this makes the early warning system more suitable for the complex scenarios of mine production, ensuring that the classification results are both accurate and practical.

[0065] From the perspective of the parameter scale of the early warning classification matrix, the level of detail and assessment dimensions of the indicators included in the matrix directly affect the level setting. If the matrix parameters cover more detailed indicators, such as subdividing different worker ranges within the potential number of casualties or distinguishing between equipment loss and production stoppage loss within economic losses, the parameter scale will expand accordingly. In this case, the early warning level can be appropriately increased to more accurately match different levels of risk. For example, a three-level early warning might not be able to distinguish subtle differences within the medium risk range. When the parameter scale is expanded, a level one early warning can be added to provide a more realistic definition of situations bordering on medium risk. The difference in risk type is another important basis for adjusting the level. Different types of mine risks have different evolution speeds, impact ranges, and severity. Risks like rock outbursts develop rapidly and have serious consequences, requiring more detailed level classifications so that targeted measures can be triggered in the early stages through clear early warning levels. For example, levels can be gradually upgraded from level one to level four, corresponding to states such as slightly exceeding the dust concentration standard, continuously rising, approaching the critical value, and reaching the dangerous value. In contrast, some risks involving slow deformation of geological structures have a more gradual change process and do not require overly detailed levels. A three-level early warning system (low, medium, and high) is sufficient to meet control needs, avoiding the waste of response resources due to too many levels. This flexible approach allows the early warning system to dynamically adapt to the actual situation. For highly complex and high-risk production scenarios, increasing the number of levels enhances the resolution of the early warning; for relatively simple scenarios, simplifying the number of levels improves response efficiency, ultimately achieving a precise match between the early warning level and the actual risk situation, providing clearer guidance for differentiated handling.

[0066] From a technical perspective, the above steps enhance the comprehensiveness and accuracy of early warning grading. The three-dimensional early warning indicator system characterizes risks from multiple dimensions, avoiding the limitations of single-indicator assessment; the indicator weights determined by FAHP are more in line with the actual risk assessment logic, allowing the impact of each indicator to be reasonably reflected; the cloud model achieves accurate conversion between qualitative and quantitative analysis, reducing the bias of human judgment. The dynamic early warning heat map can intuitively display the risk level of different locations, making it easier for staff to quickly grasp the overall risk situation, while the flexible adjustment of the early warning status level enhances the adaptability of the UWB positioning system to different risk types and scenarios, making early warning grading more in line with actual production needs, providing a clear and reliable basis for subsequent graded responses, and further improving the safety management efficiency of mine production.

[0067] Optionally, a three-dimensional early warning indicator system is used to dynamically classify early warnings for different locations within the target area and generate a graded response strategy that matches personnel activity trajectory data, including:

[0068] Within the high-risk warning area, an emergency evacuation procedure is automatically triggered, the optimal escape route is pushed to users, and a drone swarm is activated for evacuation guidance and situation monitoring. Within the medium-risk warning area, the risk evolution trend is simulated through a digital twin platform, a suggested equipment shutdown range is generated, and a disposal plan is pushed to relevant personnel through a safety supervision platform. Within the low-risk warning area, the equipment maintenance history database is linked, and potential fault points are overlaid on the equipment using augmented reality technology. Maintenance work orders are simultaneously generated and sent to the MES (Manufacturing Execution System). Based on the Dijkstra algorithm, personnel evacuation routes are dynamically planned to ensure optimal safety based on the mine's roadway congestion coefficient and risk spread speed.

[0069] Specifically, the above steps generate a tiered response strategy that matches personnel activity trajectories. Its core principle is a differentiated handling logic based on risk levels, combined with real-time personnel location information, to achieve precise and efficient emergency response. High, medium, and low risk levels correspond to different probabilities of accident occurrence and degrees of harm; therefore, it is necessary to match response measures with increasing intensity. Simultaneously, by binding the response strategy to specific locations through personnel activity trajectories, it ensures that measures directly affect the risk-related areas and relevant personnel, avoiding resource waste or insufficient response caused by indiscriminate responses.

[0070] In terms of specific implementation, for areas corresponding to high-risk warning states, the MES system automatically triggers emergency evacuation procedures. At this time, the real-time location of affected personnel is determined by combining their activity trajectories, and the optimal escape route is pushed to them via smart terminals. This route avoids high-risk areas and congested sections. Simultaneously, a drone swarm is activated, utilizing the flexible mobility of drones for aerial evacuation guidance, transmitting real-time situational images to help command personnel monitor the evacuation progress. For areas corresponding to medium-risk warning states, a digital twin platform is used to combine real-time risk data with a 3D model of the mining area to simulate the future evolution trend of the risk, generating reasonable recommendations for equipment shutdown. The safety supervision platform then pushes the handling plan, including the shutdown scope and key inspection points, to relevant operations and maintenance personnel, guiding them to conduct investigations in an orderly manner. In low-risk warning areas, the MES system connects to the equipment maintenance history database, retrieving past fault records and maintenance information of the equipment in that area. Augmented reality technology is used to overlay potential fault points as visual markers onto the equipment, facilitating rapid problem location by maintenance personnel. Simultaneously, a maintenance work order is generated and sent to the MES system, automating the maintenance process. In addition, when dynamically planning personnel evacuation routes based on the Dijkstra algorithm, the congestion coefficient and risk spread speed of the tunnels in the mining area are continuously collected to continuously optimize the route selection and ensure that the personnel evacuation route always avoids high-risk spread areas and congested sections.

[0071] From a technical perspective, this tiered response strategy enhances the accuracy and efficiency of emergency response. Emergency evacuation procedures in high-risk areas guide personnel away from danger in the shortest possible time, while the involvement of drone swarms enhances the flexibility and situational awareness of this guidance. Digital twin simulations and targeted response plans in medium-risk areas can control the spread of risk while minimizing unnecessary production interruptions. Augmented reality fault alerts and automated work order generation in low-risk areas can quickly resolve potential hazards and prevent risk escalation. Furthermore, the combination of personnel trajectories and dynamic path planning ensures that every response measure accurately reaches its target, protecting personnel safety while minimizing the impact on normal production, achieving a balance between safety and efficiency.

[0072] Optionally, a three-dimensional early warning indicator system is constructed by combining accident risk probability distribution data and dynamic early warning mechanisms within the target area. This system is then used to dynamically classify early warnings for different locations within the target area and generate graded response strategies that match personnel activity trajectory data. Furthermore, the system utilizes Kalman filtering to fuse post-treatment data and calculate the risk mitigation rate. The expression for the risk mitigation rate is:

[0073]

[0074] Where, α t Let α be the risk diffusion damping coefficient. t Related to the type of risk, α t Used to characterize the inertial characteristics of natural risk decay; P0 is the initial baseline risk value to avoid calculation deviations caused by an excessively small initial value; β i The weight of the i-th disposal measure is determined using the analytic hierarchy process (AHP); the β corresponding to equipment shutdown is... i Take 0.3, the β corresponding to personnel evacuation i Take 0.5, and adjust the corresponding β for ventilation. i Take 0.2; γ i γ is the actual effectiveness coefficient of the i-th measure, obtained by fitting it using a BP neural network based on historical treatment data; i The value ranges from 0 to 1; n represents the total number of actions taken at the current moment; P t P represents the real-time risk assessment value within the target area at time t; t-1 This represents the historical risk assessment value within the target area at time t-1. Time t can be the current time, and time t-1 can be the time preceding the current time, which is one monitoring cycle prior. Further optionally, P... tThe data obtained can be obtained by fusing multi-source monitoring data after treatment using the Kalman filter algorithm. The fused data includes, but is not limited to: spatiotemporal evolution data of rock mass stress at time t, surface deformation data, environmental monitoring data (dust concentration, temperature and humidity, etc.), personnel activity trajectory correlation data, and actual feedback data of implemented treatment measures (such as equipment shutdown and ventilation adjustment); P t-1 The method for obtaining P is similar and will not be repeated here. It should be noted that if P... t Less than P t-1 A positive risk regression rate indicates that the risk is receding and the mitigation measures may be effective; conversely, a negative risk regression rate indicates that the risk has not receded or has escalated, and it is necessary to determine whether to trigger a response strategy escalation based on whether the risk regression rate is lower than a preset value. When the risk regression rate remains below the preset value for a period of time, the response strategy escalation mechanism is triggered.

[0075] Optionally, newly added rockburst cases automatically extract spatiotemporal features, construct a feature template library, and update the parameters of the spatiotemporal convolutional network model through online learning algorithms to achieve adaptive optimization of the spatiotemporal convolutional network model. After completing dynamic early warning classification by combining the accident risk probability distribution and dynamic early warning mechanism within the target area, further processing revolves around risk decline monitoring and continuous optimization of the spatiotemporal convolutional network model. The principle is to evaluate the handling effect through precise data fusion technology and rely on intelligent learning mechanisms to allow the spatiotemporal convolutional network model to adapt to new risk features. Kalman filtering can effectively fuse multi-source data after handling, filter out noise interference, accurately calculate the risk decline, and provide a basis for adjusting the response strategy; online learning algorithms can enable the spatiotemporal convolutional network model to continuously absorb the features of new cases, continuously optimize parameters, and improve the ability to identify new risks.

[0076] In terms of specific implementation, Kalman filtering is used to fuse various data after the response, including changes in rock mass stress, dust concentration, and personnel evacuation status. Through recursive calculation using the filtering algorithm, a smooth and accurate risk regression rate is obtained. When the risk regression rate is consistently lower than the preset value, and this state lasts for a set duration, it indicates that the current response measures have failed to effectively control the risk. The MES system will automatically trigger a response strategy upgrade mechanism, such as upgrading the original partial shutdown to a complete shutdown, or dispatching more rescue forces. For new rockburst cases, the MES system automatically extracts their spatiotemporal characteristics, such as the stress change patterns at the time of occurrence and the spatial distribution characteristics of microseismic signals, and incorporates these features into a feature template library. Subsequently, an online learning algorithm is used to update the parameters of the spatiotemporal convolutional network model, allowing the model to quickly learn and master new risk patterns without retraining the entire network, achieving adaptive optimization of the spatiotemporal convolutional network model. From a technical perspective, Kalman filtering, by fusing data to calculate risk mitigation rates, improves the accuracy of risk assessment, avoids misjudgments caused by fluctuations in single data points, and ensures more reliable triggering of response strategy upgrade mechanisms, enabling timely responses to situations where risk control is inadequate. Feature extraction of newly added rockburst cases and online optimization of the spatiotemporal convolutional network model allow the MES system to continuously accumulate new risk identification experience, continuously improving its early warning capabilities for novel or rare risks and avoiding missed or false alarms due to lag in the spatiotemporal convolutional network model. This mechanism creates a closed loop in the entire early warning system, encompassing monitoring, response, assessment, and optimization, enhancing its dynamic adaptability to complex and ever-changing mine risks and further ensuring mine production safety.

[0077] Alternatively, the expression for the risk decay rate is:

[0078]

[0079] Where, α t Let α be the risk diffusion damping coefficient. t Related to the type of risk, α t The inertial characteristics used to characterize the natural decay of risk are 0.8-1.2 for geological disasters and 1.2-1.5 for dust exceeding standards accidents. P0 is the initial baseline risk value to avoid calculation deviations caused by excessively small initial values; β i The weight of the i-th disposal measure is determined using the analytic hierarchy process (AHP); the β corresponding to equipment shutdown is... i Take 0.3, the β corresponding to personnel evacuation i Take 0.5, and adjust the corresponding β for ventilation. i Take 0.2; γ i γ is the actual effectiveness coefficient of the i-th measure, obtained by fitting it using a BP neural network based on historical treatment data; iThe value ranges from 0 to 1; n represents the total number of actions taken at the current moment; P t P represents the real-time risk assessment value within the target area at time t; t-1 This represents the historical risk assessment value within the target area at time t-1. Time t can be the current time, and time t-1 can be the time preceding the current time, separated by one monitoring cycle. This formula retains the core change in risk probability while incorporating dynamic parameters such as the characteristics of the risk itself and the impact of response measures, thus more realistically reflecting the risk decay patterns under different scenarios. For example, in a dust exceeding standard accident, when ventilation adjustment (β... i =0.2) and the measured efficiency γ i When the coefficient is 0.8, the effect of the measures on accelerating risk mitigation can be reflected by superimposing the coefficients.

[0080] In this embodiment, through multi-dimensional and high-precision data collection and intelligent analysis, the problems of emphasizing detection and neglecting response in the current mine production anomaly early warning and handling technologies are effectively solved, thereby improving the safety of mine production and the scientific nature of early warning and handling.

[0081] To address the issues of overemphasis on detection, insufficient response, and inadequate differentiation of risk levels, this application's embodiments form a complete closed-loop optimization from data acquisition to early warning response. First, by simultaneously acquiring spatiotemporal evolution data of rock mass stress, surface deformation data, and environmental monitoring data through various monitoring devices such as microseismic sensor arrays, InSAR satellite remote sensing, and dust concentration sensor networks, comprehensive and three-dimensional perception of mine production anomalies is achieved. This overcomes the limitations of traditional single-detection methods and provides a rich and accurate data foundation for subsequent risk assessment. Second, the use of spatiotemporal convolutional networks for anomaly pattern detection of multi-source data enables in-depth mining of hidden spatiotemporal correlation features within the data, accurately calculating the probability distribution of accident risks within the target area, rather than simply outputting normal or abnormal binary signals. This provides a scientific basis for the refined classification of risk levels, accurately distinguishing between low, medium, and high risk levels. Based on this, a dynamic early warning mechanism is combined to dynamically classify early warnings for different locations within the target area and generate graded response strategies that match personnel activity trajectories, completely changing the current situation where traditional early warning systems cannot match differentiated plans. For example, for low-risk areas, a response strategy for local equipment inspection can be triggered. For medium-risk areas, the system automatically generates instructions for temporary shutdown and investigation. For high-risk areas, emergency evacuation and equipment reinforcement plans are activated, ensuring that each level of risk has corresponding and precise response measures. This dynamic binding of risk levels with response strategies transforms early warning information from simple alarm signals into action guidelines that directly guide on-site handling, greatly enhancing the practical value of early warning information and effectively solving the problem of disconnect between early warning and response, and the inability to effectively guide on-site work in traditional technologies. Furthermore, this application's embodiments also demonstrate real-time risk assessment and targeted response strategies. By tracking personnel activity trajectories in real time, the tiered response strategy can be dynamically matched with personnel locations, avoiding unnecessary total shutdowns or evacuations while ensuring personnel safety. This balances safety and efficiency, improving both. In addition, the fusion analysis of multi-source data and the application of intelligent algorithms significantly improve the accuracy of abnormal pattern detection, reduce false alarms and missed alarms, further enhancing the reliability of the early warning system and providing strong technical support for the safety management of mine production.

[0082] Compared with existing technologies, the intelligent identification and hierarchical early warning method for mine production anomalies provided in this embodiment achieves at least the following beneficial effects:

[0083] The intelligent identification and hierarchical early warning method for mine production anomalies provided in this embodiment firstly acquires spatiotemporal evolution data of rock mass stress in the target area in real time using a microseismic sensor array; secondly, it acquires remote sensing data of the target area using InSAR satellites, and calculates surface deformation data using interferometric analysis; and thirdly, it monitors the environment in the target area in real time using a dust concentration sensor network, temperature and humidity sensors, and wind speed sensors to obtain environmental monitoring data. Finally, it uses a spatiotemporal convolutional network to perform anomaly pattern detection on the spatiotemporal evolution data, surface deformation data, and environmental monitoring data to obtain accident risk probability distribution data of the target area. Then, it uses an ultra-wideband positioning system to track... The system tracks the activity trajectories of personnel entering the target area to obtain personnel activity trajectory data. Finally, it constructs a three-dimensional early warning indicator system by combining the accident risk probability distribution data and dynamic early warning mechanism within the target area. Using this three-dimensional early warning indicator system, dynamic early warning classifications are applied to different locations within the target area, and graded response strategies matching the personnel activity trajectory data are generated. By adopting the above scheme, the efficiency of mine production anomaly diagnosis can be improved, the activity trajectories of personnel entering the target area can be tracked in real time, the possibility of accidents caused by human violations can be reduced, the overall safety management level of the mine production site can be improved, and corresponding differentiated handling plans can be matched for different levels of risk, thereby improving the timeliness of mine production anomaly early warning.

[0084] Example 2

[0085] Reference Figure 3 As shown, Figure 3 This is a schematic diagram of a mine production anomaly intelligent identification and hierarchical early warning system provided by the present invention. This embodiment provides a mine production anomaly intelligent identification and hierarchical early warning system, including:

[0086] Acquisition module 1, coupled to detection module 2, is used to acquire in real time spatiotemporal evolution data of rock mass stress, environmental monitoring data, and surface deformation data within the target area. The surface deformation data is obtained by acquiring remote sensing data within the target area using InSAR satellites. Acquiring remote sensing data within the target area using InSAR satellites includes periodic scanning of the target area. The scanning cycle of the InSAR satellites is dynamically adjusted based on the satellite model, networking mode, and risk level of the mining area. In high-risk mining areas such as the working face advancement zone or old goaf areas, the scanning cycle is 3 days when the InSAR satellite is a dual-satellite network; 4 days when the InSAR satellite is a land exploration satellite in dual-satellite follow-fly mode; and 8 days when the InSAR satellite is a land exploration satellite in single-satellite mode.

[0087] The detection module 2 is coupled to the acquisition module 1 and the prediction module 4 respectively. It is used to receive spatiotemporal evolution data, surface deformation data and environmental monitoring data in the target area. It uses a spatiotemporal convolutional network to perform abnormal pattern detection on the spatiotemporal evolution data, surface deformation data and environmental monitoring data to obtain accident risk probability distribution data in the target area.

[0088] Tracking module 3, coupled to prediction module 4, is used to provide personnel activity trajectory data to prediction module 4, and to use the ultra-wideband positioning system to track the personnel activity trajectory entering the target area in real time to obtain personnel activity trajectory data;

[0089] The prediction module 4, coupled to the detection module 2 and the tracking module 3 respectively, is used to receive accident risk probability distribution data and personnel activity trajectory data in the target area. Based on the accident risk probability distribution data and personnel activity trajectory data in the target area, a three-dimensional early warning indicator system is constructed. The three-dimensional early warning indicator system is used to dynamically classify early warnings for different locations in the target area and generate a graded response strategy that matches the personnel activity trajectory data.

[0090] It should be noted that: the microseismic sensor array is used to acquire spatiotemporal evolution data of rock mass stress in the target area in real time; the dust concentration sensor network, temperature and humidity sensor, and wind speed sensor are used to acquire environmental monitoring data in the target area; and the InSAR satellite is used to acquire remote sensing data, thereby obtaining surface deformation data in the target area.

[0091] The intelligent identification and hierarchical early warning system for mine production anomalies provided in this embodiment can realize all the steps in the intelligent identification and hierarchical early warning method for mine production anomalies in Embodiment 1, which will not be repeated here.

[0092] Example 3

[0093] Reference Figure 4 As shown, Figure 4This is a schematic diagram of a mine production anomaly intelligent identification and hierarchical early warning device provided by the present invention. The device includes a processor 10, a communication interface 20, a memory 30, and a communication bus 40. The processor 10, communication interface 20, and memory 30 communicate with each other via the communication bus 40. The memory 30 stores computer programs. When the processor 10 executes the program stored in the memory 30, it implements the mine production anomaly intelligent identification and hierarchical early warning method in Embodiment 1. The communication bus 40 mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 40 can be divided into an address bus, a data bus, a control bus, etc.; for ease of representation, Figure 4 The bus is represented by a single thick line, but this does not imply that there is only one bus or one type of bus. Communication interface 20 is used for communication between the aforementioned electronic device and other devices. Memory 30 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor. The aforementioned processor 10 may be a general-purpose processor, including artificial intelligence processors, graphics processing units (GPUs), machine learning units (MLUs), central processing units (CPUs), network processors (NPs), etc.; it may also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0094] Accordingly, this application also provides a computer-readable storage medium storing a computer program, which, when executed, can perform the steps that can be executed by an electronic device in Embodiment 1.

[0095] While specific embodiments of the invention have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of the invention. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of the invention. The scope of the invention is defined by the appended claims.

Claims

1. A method for intelligent identification and hierarchical early warning of mine production anomalies, characterized in that, The method comprises the following steps: Real-time acquisition of the spatio-temporal evolution data of the rock mass stress in the target area by using a microseismic sensor array; real-time acquisition of remote sensing data in the target area by using an InSAR satellite; calculation of the remote sensing data by using an interferometric measurement analysis method to obtain surface deformation data; real-time monitoring of the environment in the target area by using a dust concentration sensor network, a temperature and humidity sensor, and a wind speed sensor to obtain environmental monitoring data; wherein the real-time acquisition of the remote sensing data in the target area by using the InSAR satellite comprises: periodic scanning of the target area by using the InSAR satellite; the scanning period of the InSAR satellite is dynamically adjusted according to the satellite model, networking mode, and mine risk level of the InSAR satellite; in the high-risk mining area such as the working face advancing area or the old goaf, when the InSAR satellite is a satellite device in a double-satellite networking mode, the scanning period is 3 days; when the InSAR satellite is a land exploration satellite in a double-satellite follow-up mode, the scanning period is 4 days; when the InSAR satellite is a land exploration satellite in a single-satellite mode, the scanning period is 8 days; Abnormal pattern detection of the spatio-temporal evolution data, the surface deformation data, and the environmental monitoring data by using a spatio-temporal convolution network to obtain accident risk probability distribution data in the target area; Real-time tracking of the personnel activity trajectory entering the target area by using an ultra-wideband positioning system to obtain personnel activity trajectory data; Combination of the accident risk probability distribution data in the target area and a dynamic early warning mechanism to construct a three-dimensional early warning index system, dynamic early warning grading of different positions in the target area by using the three-dimensional early warning index system, and generation of a graded response strategy matched with the personnel activity trajectory data; Wherein, the combination of the accident risk probability distribution data in the target area and the dynamic early warning mechanism to construct a three-dimensional early warning index system, the dynamic early warning grading of different positions in the target area by using the three-dimensional early warning index system, and the generation of a graded response strategy matched with the personnel activity trajectory data further comprise using Kalman filtering to fuse the disposed data to obtain a risk recession rate, and the expression of the risk recession rate is: wherein, α t is a risk diffusion damping coefficient, P0 is an initial reference risk value, β i is a weight of the i th treatment measure, γ i is an actual effectiveness coefficient of the i th measure, i ranges from 0 to 1, and n is a total number of treatment measures executed at the current time; P t is a real-time risk assessment value in the target region at time t, P t-1 is a historical risk assessment value in the target region at time t-1, time t is the current time, and time t-1 is a previous time interval from the current time by one monitoring period.

2. The mine production anomaly intelligent identification and grading early warning method according to claim 1, characterized in that, The abnormal pattern detection of the spatio-temporal evolution data, the surface deformation data, and the environmental monitoring data by using a spatio-temporal convolution network to obtain accident risk probability distribution data in the target area comprises the following steps: Projecting the spatio-temporal evolution data and the surface deformation data into a original geological spatio-temporal model to obtain a target geological spatio-temporal model; Inputting the environmental monitoring data into a gas diffusion physical model to obtain a gas diffusion dynamics model; Abnormal pattern detection of the target geological spatio-temporal model and the gas diffusion dynamics model by using the spatio-temporal convolution network to obtain the accident risk probability distribution data in the target area; Based on the topographic structure, mine functional zoning and mine type of the target area, an accident chain deduction model matched with the target area is established, the real-time collected spatio-temporal evolution data, ground surface deformation data and environmental monitoring data are input into the accident chain deduction model, and spatio-temporal causal chain data at different positions in the target area are obtained.

3. The method according to claim 2, characterized in that, The abnormal pattern detection of the target geologic spatio-temporal model and the gas diffusion dynamics model by using the spatio-temporal convolution network includes the following steps: The STL algorithm is adopted in the target geologic spatio-temporal model and the gas diffusion dynamics model to separate the rock movement long-term cumulative characteristic data in the target area according to the time change; According to the rock movement long-term cumulative characteristic data in the target geologic spatio-temporal model, the change trend of the rock fracture density is predicted, and the geologic accident risk probability distribution data at different positions in the target area are obtained; The dust concentration spatial aggregation in the gas diffusion dynamics model is predicted by the Moran index, and the prediction result is obtained, and the dust concentration change characteristic data at different positions in the target area are obtained according to the prediction result, and the dust accident risk probability distribution data at different positions in the target area are obtained; The seasonal diurnal variation characteristic data and / or the burst gas parameter change characteristic data in the target area are extracted from the gas diffusion dynamics model, and the environmental accident risk probability distribution data at different positions in the target area are predicted by using the Bayesian variable point detection algorithm; The geologic accident risk probability distribution data, the dust accident risk probability distribution data and the environmental accident risk probability distribution data at different positions in the target area are fused, and the accident risk probability distribution data in the target area are obtained.

4. The method according to claim 2, characterized in that, The accident chain deduction model matched with the target area is established based on the topographic structure, mine functional zoning and mine type of the target area, the real-time collected spatio-temporal evolution data, ground surface deformation data and environmental monitoring data are input into the accident chain deduction model, and spatio-temporal causal chain data at different positions in the target area are obtained. Based on the spatio-temporal evolution data, ground surface deformation data and environmental monitoring data, the weight vectors of geologic disaster-causing factors, equipment operation parameters and environmental monitoring indexes are determined by using the analytic hierarchy process, and a multi-level accident chain network is constructed; Through the joint probability distribution calculation of the Bayesian network, the multi-type accident evolution path is deduced in the multi-level accident chain network; Combined with the abnormal pattern characteristics extracted by the spatio-temporal convolution network, the conditional probability table of the accident chain node in the multi-level accident chain network is dynamically updated, the causal chain sequence with time stamp and spatial coordinates is output, and the spatio-temporal causal chain data at different positions in the target area are obtained.

5. The method for intelligent identification and grading early warning of mine production anomalies according to claim 1, characterized in that, The personnel activity trajectory data obtained by using the ultra-wideband positioning system to track the personnel activity trajectory in the target area in real time includes the following steps: Deploying an ultra-wideband positioning system, laying positioning base stations in the target area, personnel carrying positioning tags, real-time acquisition of the distance between the tags and the positioning base stations through TOF ranging technology, calculation of the accurate position information of the personnel, continuous recording of the accurate position information to form the personnel activity trajectory; Introducing the personnel activity trajectory into a three-dimensional mine geographic model, pre-labeling the spatial coordinates of each risk source in the three-dimensional mine geographic model, calculating the actual distance between the personnel and the risk source through the spatial coordinates, and when the actual distance between the personnel and the risk source is less than a set safety threshold, the ultra-wideband positioning system automatically triggers a near-field early warning to remind the personnel to stay away from the dangerous area; Installing cameras at key positions in the target area to collect activity images of the personnel, analyzing the activity images using a YOLOv8 target detection algorithm, identifying whether the personnel wear safety equipment, whether they enter a restricted area, and the like, inputting the identification results into a behavior risk assessment model to obtain the personnel activity trajectory data.

6. The method for intelligent identification and grading early warning of mine production anomalies according to claim 1, characterized in that, The three-dimensional early warning index system is constructed by combining the accident risk probability distribution data in the target area and the dynamic early warning mechanism, the three-dimensional early warning index system is used for dynamic early warning grading of different positions in the target area, and a graded response strategy matched with the personnel activity trajectory data is generated, the graded response strategy includes: In the position area corresponding to the high-risk early warning state, an emergency evacuation program is automatically triggered, an optimal escape path is pushed to relevant personnel, and a drone group is started to guide evacuation and monitor the situation; In the position area corresponding to the medium-risk early warning state, a risk evolution trend is simulated through a digital twin platform, a device shutdown range suggestion is generated, and a disposal scheme is pushed to relevant personnel through a safety supervision platform; In the position area corresponding to the low-risk early warning state, a device maintenance history database is associated, a fault hidden point is superimposed on the device through augmented reality technology, and a maintenance work order is generated to an MES system.

7. A mine production anomaly intelligent identification and grading early warning system based on, characterized in that, It includes: The acquisition module is coupled with the detection module and is used for real-time acquisition of the spatio-temporal evolution data of rock mass stress, environmental monitoring data and surface deformation data in the target area; wherein the surface deformation data is obtained by using an InSAR satellite to obtain remote sensing data in the target area, and the InSAR satellite is used to obtain remote sensing data in the target area, which includes: using the InSAR satellite to periodically scan the target area; the scanning period of the InSAR satellite is dynamically adjusted according to the satellite model, networking mode and mine risk level of the InSAR satellite; in the high-risk mining area such as working face advancing area or old goaf, when the InSAR satellite is a satellite device in double-star networking mode, the scanning period is 3 days; when the InSAR satellite is a land exploration satellite in double-star follow-up mode, the scanning period is 4 days; when the InSAR satellite is a land exploration satellite in single-star mode, the scanning period is 8 days; The detection module is coupled with the collection module and the prediction module respectively, and is configured to receive the spatio-temporal evolution data, the ground deformation data and the environmental monitoring data in the target area, perform abnormal pattern detection on the spatio-temporal evolution data, the ground deformation data and the environmental monitoring data by using a spatio-temporal convolution network, and obtain accident risk probability distribution data in the target area; The tracking module is coupled with the prediction module, and is configured to provide the prediction module with personnel activity trajectory data, track the personnel activity trajectory entering the target area in real time by using an ultra-wideband positioning system, and obtain the personnel activity trajectory data; The prediction module is coupled with the detection module and the tracking module respectively, and is configured to receive the accident risk probability distribution data and the personnel activity trajectory data in the target area, construct a three-dimensional early warning index system according to the accident risk probability distribution data and the personnel activity trajectory data in the target area, perform dynamic early warning grading on different positions in the target area by using the three-dimensional early warning index system, and generate a graded response strategy matched with the personnel activity trajectory data, and the prediction module is further configured to obtain a risk recession rate by fusing post-disposal data by using Kalman filtering, and the expression of the risk recession rate is: wherein, α t is a risk diffusion damping coefficient, P0 is an initial reference risk value, β i is a weight of the i th countermeasure, γ i is an actual effectiveness coefficient of the i th countermeasure, i ranges from 0 to 1, and n is a total number of countermeasures executed at the current time; P t is a real-time risk assessment value in the target region at time t, P t-1 is a historical risk assessment value in the target region at time t-1, time t is the current time, and time t-1 is a previous time interval one monitoring period from the current time.

8. A mine production anomaly intelligent identification and grading early warning device based on, characterized in that, Comprise: A processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus; The memory is used to store a computer program; When the processor executes the computer program stored on the memory, the method for intelligent identification and hierarchical early warning based on mine production anomaly according to any one of claims 1-6 is implemented.

Images