Risk identification method and system based on mine data

Abstract

The invention discloses a risk identification method and system based on mine data, and belongs to the technical field of computers. According to the method, effective fusion and collaborative analysis of multi-source heterogeneous monitoring data are realized, and the problem of data islands in a traditional method is solved to a certain extent. Through multi-level feature modeling of a dynamic space-time diagram, a space-time propagation rule of mine risks is captured. And in combination with a dynamic correction mechanism of a real-time mining state, the adaptability of a risk identification result to a complex mining environment is improved. The application of the data quality evaluation matrix reduces the influence of low-quality data on an analysis result to a certain extent, and improves the reliability of risk early warning.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a risk identification method and system based on mining data. Background Technology

[0002] Mine safety production monitoring systems have accumulated a large amount of multi-source heterogeneous data during long-term operation. However, existing risk identification technologies often use simple data splicing or independent analysis when processing this data, which cannot achieve true multi-source data collaborative analysis.

[0003] Due to the lack of an effective data fusion mechanism, the inherent correlations between different types of data are difficult to fully utilize. This results in significant limitations in the risk identification process, making it difficult to grasp the overall risk situation in the complex environment of a mine.

[0004] Therefore, the challenge lies in overcoming the obstacles to collaborative analysis among multi-source heterogeneous data and establishing a risk identification method that can fully explore the inherent correlations between different types of data, thereby achieving comprehensive risk perception of the entire mining system. Summary of the Invention

[0005] This application provides a risk identification method and system based on mine data, which can comprehensively perceive the risks of the entire mining system. The technical solution is as follows: On the one hand, a risk identification method based on mine data is provided, the method comprising: In response to mine risk monitoring instructions for the target mine, multi-source heterogeneous monitoring data of the target mine is acquired. This multi-source heterogeneous monitoring data includes sensor monitoring data, geological structure data, mining process data, and environmental monitoring data. The multi-source heterogeneous monitoring data is processed to obtain a data quality assessment matrix, a standardized data stream, and feature-enhanced data. The data quality assessment matrix includes quality assessment parameters across four dimensions: equipment operating status, spatiotemporal consistency, data completeness, and abnormal fluctuations. The feature-enhanced data includes time-domain statistical features, frequency-domain energy features, and spatiotemporal correlation features. Based on the data quality assessment matrix and feature-enhanced data, a multi-level feature representation of the target mine's dynamic spatiotemporal map is determined. This multi-level feature representation includes node-level micro-features, sub-graph-level regional features, and graph-level macro-features. Based on the multi-level feature representation, the data quality assessment matrix, and the real-time mining status of the target mine, the risk identification results of the target mine are determined. The real-time mining status includes the location of the mining face, the advance speed, and equipment distribution information.

[0006] Furthermore, this application proposes processing multi-source heterogeneous monitoring data to obtain a data quality assessment matrix, a standardized data stream, and feature-enhanced data, including: performing quality assessment and credibility analysis on multi-source heterogeneous monitoring data to obtain a data quality assessment matrix and a credible data stream; and performing spatiotemporal alignment and multi-scale feature enhancement on the credible data stream based on geological structural data and mining process data to obtain a standardized data stream and feature-enhanced data.

[0007] Furthermore, this application proposes to conduct quality assessment and reliability analysis on multi-source heterogeneous monitoring data to obtain a data quality assessment matrix and a reliable data stream, including: determining the working status score of each sensor data based on the equipment operating parameters in the sensor monitoring data; determining the spatial consistency index based on the spatial correlation between geological structure data and sensor monitoring data; determining the temporal consistency index based on the temporal series consistency between environmental monitoring data and sensor monitoring data; generating a data quality assessment matrix based on the working status score, spatial consistency index, and temporal consistency index; and filtering and repairing the multi-source heterogeneous monitoring data based on the data quality assessment matrix to obtain a reliable data stream.

[0008] Furthermore, this application proposes to perform spatiotemporal alignment and multi-scale feature enhancement on a reliable data stream based on geological structural data and mining technology data to obtain a standardized data stream and feature-enhanced data. This includes: performing spatiotemporal alignment on the reliable data stream to obtain a data sequence under a unified spatiotemporal reference; extracting temporal statistical features and frequency domain energy features from the data sequence under the unified spatiotemporal reference to obtain temporal and frequency domain feature components; determining spatiotemporal correlation feature components based on geological structural data and mining technology data, whereby the spatiotemporal correlation feature components represent the intensity of spatiotemporal interaction between monitoring data points based on geological structural relationships and mining technology dynamics. The spatial attenuation model and temporal influence factor are used to quantify the spatial constraint effect of geological structures on monitoring points and the temporal influence of mining technology on the state of monitoring points. The spatial attenuation model is an exponential attenuation model or a Gaussian model, and the temporal influence factor is determined based on historical mining technology data; fusing the temporal feature components, frequency domain feature components, and spatiotemporal correlation feature components to obtain feature-enhanced data; and standardizing the reliable data stream based on the feature-enhanced data to obtain a standardized data stream.

[0009] Furthermore, this application proposes a multi-level feature representation of the dynamic spatiotemporal graph of a target mine based on a data quality assessment matrix and feature enhancement data. This includes: constructing a multi-scale node feature tensor and a multi-modal edge relationship matrix of the dynamic spatiotemporal graph of the target mine based on the data quality assessment matrix and feature enhancement data. The multi-modal edge relationship matrix includes five edge types: spatial topological relationship, geological structural association, stress field coupling relationship, mining disturbance propagation, and data flow correlation. Spatiotemporal feature extraction and cross-modal fusion are performed on the multi-scale node feature tensor and the multi-modal edge relationship matrix to obtain a multi-level feature representation. Node-level micro-features are used to represent the feature information of a single node in the dynamic spatiotemporal graph, subgraph-level regional features are used to represent the feature information of a sub-region composed of adjacent nodes in the dynamic spatiotemporal graph, and graph-level macro-features are used to represent the overall feature information of the dynamic spatiotemporal graph.

[0010] Furthermore, this application proposes to construct a multi-scale node feature tensor and a multi-modal edge relation matrix for a dynamic spatiotemporal graph of a target mine based on a data quality assessment matrix and feature enhancement data. This includes: constructing initial node features based on feature enhancement data, where each initial node feature vector is formed by fusing temporal statistical features, frequency energy features, and spatiotemporal correlation features of its corresponding location; performing reliability weighting processing on the initial node features based on the data quality assessment matrix to obtain weighted node features; extracting multi-scale features from the weighted node features to obtain a multi-scale node feature tensor, where node-level features are directly derived from the weighted node features, regional-level features are obtained through aggregation calculation of adjacent node features, and global-level features are obtained through statistical analysis of the full graph features; and constructing a multi-modal edge relation matrix based on the spatiotemporal correlation feature components in the feature enhancement data.

[0011] Furthermore, this application proposes to extract spatiotemporal features and perform cross-modal fusion on multi-scale node feature tensors and multi-modal edge relation matrices to obtain multi-level feature representations, including: performing spatial feature propagation and aggregation on multi-scale node feature tensors based on spatial topological relationships and geological structural correlation edges in the multi-modal edge relation matrix to obtain local feature patterns and regional feature distributions in the spatial dimension; performing temporal feature modeling and evolution analysis on multi-scale node feature tensors based on data flow correlation edges and mining disturbance propagation edges in the multi-modal edge relation matrix to obtain short-term fluctuation features and long-term trend features in the time dimension; and generating a unified multi-level feature representation by performing spatiotemporal feature interaction and multi-scale fusion through a cross-modal attention mechanism based on local feature patterns and regional feature distributions in the spatial dimension, as well as short-term fluctuation features and long-term trend features in the time dimension.

[0012] Furthermore, this application proposes to determine the risk identification results of a target mine based on multi-level feature representation, data quality assessment matrix, and real-time mining status of the target mine. This includes: conducting risk propagation mechanism analysis and dynamic evolution prediction on the multi-level feature representation and data quality assessment matrix to obtain a risk level distribution map and risk evolution prediction trajectory; and conducting risk identification and confidence assessment based on the risk level distribution map, risk evolution prediction trajectory, multi-level feature representation, and real-time mining status, combined with the data quality assessment matrix, to obtain risk identification results. These risk identification results are used to represent the identification information of four risk types: roof collapse risk, water inrush risk, gas outburst risk, and rock burst risk.

[0013] Furthermore, this application proposes to analyze the risk propagation mechanism and predict the dynamic evolution of multi-level feature representations and data quality assessment matrices to obtain a risk level distribution map and a risk evolution prediction trajectory. This includes: extracting spatial correlations and risk propagation paths between regions based on subgraph-level regional features in the multi-level feature representation, with the propagation path determined by analyzing the spatial continuity and evolutionary synchronicity of features of adjacent subgraphs; determining the dynamic trend and intensity of global risk propagation based on graph-level macro-features in the multi-level feature representation; initializing the risk status of each node in the dynamic spatiotemporal graph based on node-level micro-features in the multi-level feature representation, and using the data quality assessment matrix to weighted correct the credibility of the node status; constructing a risk propagation dynamic equation based on the risk propagation path, the global dynamic trend, and the weighted corrected node risk status, and obtaining a risk level distribution map and a risk evolution prediction trajectory through numerical simulation of the risk propagation dynamic equation.

[0014] Furthermore, this application proposes a method for risk identification and confidence assessment based on risk level distribution maps, risk evolution prediction trajectories, multi-level feature representations, and real-time mining status, combined with a data quality assessment matrix. This results in the following steps: generating preliminary risk classifications and levels based on the risk level distribution maps and risk evolution prediction trajectories; verifying the preliminary risk classifications using local features based on node-level micro-features in the multi-level feature representations to obtain a verified set of risk types; simulating the risk impact range based on the verified set of risk types and sub-graph-level regional features to obtain a preliminary risk impact boundary; dynamically correcting the preliminary risk impact boundary based on the real-time mining status to obtain a target risk impact range matching the current mining activities; calculating the confidence scores for the verified set of risk types and the target risk impact range based on the data quality assessment matrix; and generating risk identification results based on the verified set of risk types, the target risk impact range, and the corresponding confidence scores. The risk identification results include the probability of each candidate risk type in the risk type set.

[0015] As can be seen from the above, the risk identification method and system based on mine data provided in this application realizes the deep integration of multi-source heterogeneous data and dynamic risk evolution analysis by constructing a multi-level feature representation of a dynamic spatiotemporal map and combining it with a real-time mining status and data quality assessment mechanism. It has the technical effects of improving mine risk identification, enhancing multi-source data collaborative analysis capabilities, and realizing dynamic risk evolution prediction.

[0016] On the one hand, a risk identification system based on mining data is provided, the system comprising: Multi-source heterogeneous monitoring data, which includes sensor monitoring data, geological structure data, mining process data, and environmental monitoring data; The processing module is used to process the multi-source heterogeneous monitoring data to obtain a data quality assessment matrix, a standardized data stream, and feature-enhanced data. The data quality assessment matrix includes quality assessment parameters in four dimensions: equipment operating status, spatiotemporal consistency, data completeness, and abnormal fluctuations. The feature-enhanced data includes time-domain statistical features, frequency-domain energy features, and spatiotemporal correlation features. The feature representation determination module is used to determine the multi-level feature representation of the dynamic spatiotemporal graph of the target mine based on the data quality assessment matrix and feature enhancement data. The multi-level feature representation includes node-level micro features, subgraph-level regional features and graph-level macro features. The identification result determination module is used to determine the risk identification result of the target mine based on the multi-level feature representation, the data quality assessment matrix, and the real-time mining status of the target mine. The real-time mining status includes the location of the mining face, the advance speed, and equipment distribution information.

[0017] On one hand, a computer device is provided, the computer device including one or more processors and one or more memories, the one or more memories storing at least one computer program, the computer program being loaded and executed by the one or more processors to implement the risk identification method based on mining data.

[0018] On the one hand, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, the computer program being loaded and executed by a processor to implement the risk identification method based on mining data.

[0019] On the one hand, a computer program product or computer program is provided, which includes program code stored in a computer-readable storage medium. The processor of a computer device reads the program code from the computer-readable storage medium and executes the program code, causing the computer device to perform the aforementioned risk identification method based on mining data. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the implementation environment of a risk identification method based on mine data provided in an embodiment of this application; Figure 2 This is a flowchart of a risk identification method based on mine data provided in an embodiment of this application; Figure 3 This is a partial flowchart of a risk identification method based on mine data provided in an embodiment of this application; Figure 4 This is a partial flowchart of another risk identification method based on mine data provided in an embodiment of this application; Figure 5 This is a partial flowchart of another risk identification method based on mine data provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of a risk identification system based on mine data provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a server provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0023] In this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0024] Artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain better results.

[0025] Machine Learning (ML) is a multidisciplinary field that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and many other disciplines. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge sub-models to continuously improve their performance.

[0026] Target mine: refers to the mining production area that serves as the specific application object of risk identification, including its underground roadways, mining areas, working faces, and related facilities, forming a complete mining system.

[0027] Sensor monitoring data: Physical quantity monitoring data collected by sensing devices deployed within the mine area, including real-time or near-real-time measurements of stress, displacement, gas concentration, micro-seismic events, equipment vibration, etc.

[0028] Geological structural data: Structured data describing the geological conditions of a mine, including geological features such as fault distribution, stratum occurrence, fracture development, and rock mass strength parameters.

[0029] Mining process data: Records the technical parameters of the mining process, including the coal mining machine's operating trajectory, the working status of support equipment, blasting operation parameters, the operating status of the transportation system, and other production process information.

[0030] Environmental monitoring data: Monitoring data reflecting the environmental status of the mine, including measured values ​​of environmental parameters such as wind speed, temperature, humidity, gas composition, and dust concentration.

[0031] Data quality assessment matrix: A two-dimensional matrix structure used to quantitatively assess data quality. Its rows represent different data sources, and its columns represent four quality assessment dimensions: device operating status, spatiotemporal consistency, data completeness, and abnormal fluctuations.

[0032] Standardized data stream: A standardized data sequence that has undergone format unification, dimension normalization, and anomaly handling, and has a unified sampling frequency and data processing standard.

[0033] Feature-enhanced data: Enhanced feature representations obtained through multi-scale feature extraction and fusion processing, including three types of feature components: time-domain statistical features, frequency-domain energy features, and spatiotemporal correlation features.

[0034] Equipment operating status: Evaluation indicators characterizing the working status of monitoring equipment, including a comprehensive evaluation result of operating parameters such as equipment online rate, signal strength, and self-test status.

[0035] Spatiotemporal consistency: an indicator that assesses the degree of coordination between multi-source data in the temporal and spatial dimensions, reflecting the spatiotemporal alignment accuracy and logical consistency between different data sources.

[0036] Data completeness: A metric for evaluating the completeness of a dataset, which assesses data missingness by calculating the ratio of valid data points to the theoretically required number of data points.

[0037] Abnormal fluctuations: an indicator that characterizes the degree of abnormal changes in a data sequence, and identifies and quantifies outliers and abnormal patterns in the data through statistical analysis methods.

[0038] Time-domain statistical features: Data statistical features extracted from the time dimension, including time-domain statistics such as mean, variance, peak factor, and waveform factor.

[0039] Frequency domain energy characteristics: Energy distribution characteristics obtained through frequency domain analysis, including the energy proportion of each frequency band, dominant frequency characteristics, spectral entropy and other frequency domain statistics.

[0040] Spatiotemporal correlation characteristics: Simultaneously considering the correlation characteristics of spatial and temporal dimensions, reflecting the interaction and evolution of data points in the spatiotemporal coordinate system.

[0041] Dynamic spatiotemporal graph: A graph structure used to characterize the spatiotemporal evolution of a mining system, where nodes represent spatial locations, edges represent spatiotemporal relationships, and the graph topology is dynamically updated over time.

[0042] Subgraph-level region features: Feature representations describing the statistical characteristics of local regions in a dynamic spatiotemporal graph, obtained through aggregation calculation of node features within the region.

[0043] Graph-level macroscopic features: Feature representations describing the overall statistical characteristics of a dynamic spatiotemporal graph, obtained through comprehensive analysis of the relationships between nodes and edges in the entire graph.

[0044] Location of the mining face: Spatial coordinate information of the mining face, including spatial parameters such as the face direction, dip angle, and advance position.

[0045] Advance speed: The distance the mining face advances per unit time, reflecting the progress and intensity of the mining operation.

[0046] Equipment distribution information: The spatial layout of mining equipment underground, including distribution parameters such as equipment type, location coordinates, and working status.

[0047] Equipment operating parameters: Technical parameters of the monitoring equipment during operation, including equipment performance indicators such as sampling frequency, measurement range, and accuracy level.

[0048] Operating status score: A quantitative assessment score of the operating status of sensor equipment, calculated based on the degree of deviation between the equipment's operating parameters and standard parameters.

[0049] Spatial correlation: The degree of statistical correlation between monitoring data from different spatial locations, which is quantified and calculated using spatial statistical methods.

[0050] Spatial consistency index: A quantitative indicator that assesses the degree of consistency in the spatial distribution of multi-source data and reflects the reliability of spatial relationships.

[0051] Time consistency index: A quantitative indicator that assesses the degree of synchronization of multi-source data over time, reflecting the temporal correlation.

[0052] Spatiotemporal alignment: A data preprocessing process that unifies multi-source data to the same spatiotemporal reference, including time synchronization, spatial registration and other processing steps.

[0053] Spatiotemporal correlation feature components: The specific components of spatiotemporal correlation features, which are the correlation strength values ​​calculated through the spatial decay model and the temporal influence factor.

[0054] Spatial decay model: A mathematical model that describes how the strength of spatial association decreases with increasing distance, using exponential decay or Gaussian decay function forms.

[0055] Time-series impact factor: A parameter that quantifies the degree of influence of mining technology on the time-series changes in monitoring data, obtained based on statistical analysis of historical process data.

[0056] Standardization: Data processing methods that convert data into a unified standard and scale, including data transformation operations such as normalization and standardization.

[0057] Spatial topological relationships: Definitions that describe the topological connections such as adjacency and connectivity between spatial locations.

[0058] Geological structure correlation: The correlation established based on geological structure characteristics reflects the influence path of geological conditions on monitoring data.

[0059] Stress field coupling relationship: a definition describing the interaction between stress distribution and monitoring data.

[0060] Mining disturbance propagation: A definition that characterizes the relationship between the propagation path and the range of influence of disturbances generated by mining activities in the rock mass.

[0061] Data stream correlation: The statistical correlation between different data streams in time, reflecting the synchronicity and correlation of data changes.

[0062] Cross-modal fusion: A processing method that unifies and merges feature data from different sources and with different properties.

[0063] Spatial feature propagation and aggregation: The computational process of transferring feature information and converging local features on a graph structure.

[0064] Local feature patterns: Pattern recognition results that describe the distribution patterns of features within a small area.

[0065] Regional feature distribution: Statistical features that describe the spatial distribution patterns of features over a relatively large area.

[0066] Time series feature modeling: The modeling process of feature extraction and pattern recognition for time series data.

[0067] Evolutionary analysis: A method for analyzing how the state of a system changes over time.

[0068] Short-term fluctuation characteristics: time-frequency characteristics that characterize the short-term change patterns of data.

[0069] Long-term trend characteristics: Trend characteristics that characterize the long-term direction of data changes.

[0070] Risk propagation mechanism analysis: an analytical method for studying the propagation patterns of risks in the spatiotemporal dimensions.

[0071] Dynamic evolution prediction: Predictive analysis of the future state change trend of a system.

[0072] Spatial correlation: Statistical correlation between monitoring data at spatial locations.

[0073] Risk propagation paths: the possible paths and scope of impact of risks propagation within the mining system.

[0074] Spatial continuity: the smooth transition of features between adjacent spatial locations.

[0075] Evolutionary synchronicity: The coordination and consistency of data changes at different locations over time.

[0076] Dynamic trend and intensity: Description of the dynamic characteristics and intensity of changes in the system state.

[0077] Risk propagation dynamics equations: mathematical equation models that describe the risk propagation process.

[0078] Preliminary risk classification: Initial risk type identification based on simple rules or thresholds.

[0079] Risk Level Classification: A classification system based on the degree of risk.

[0080] Local feature verification: The process of verifying preliminary identification results using local features.

[0081] Risk impact range simulation: Determine the spatial range of the risk's potential impact through calculation and simulation.

[0082] Dynamic correction: The process of dynamically adjusting and optimizing existing results based on real-time data.

[0083] Figure 1 This is a schematic diagram illustrating the implementation environment of a risk identification method based on mine data provided in this application embodiment. See also... Figure 1 This implementation environment may include node 110 and server 140.

[0084] Node 110 is connected to server 140 via a wireless or wired network. Node 110 is deployed around the target mine, functioning as an edge device. Node 110 has an application installed and running that supports risk identification based on mine data.

[0085] Server 140 is an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms. Server 140 can provide background services for applications running on node 110. In this embodiment, due to the numerous and complex calculation processes involved, the computing power of node 110 may be insufficient, requiring server 140 to perform the corresponding calculations. Furthermore, mine safety is of paramount importance and is very difficult to identify, hence the involvement of the aforementioned complex calculation processes.

[0086] In related technologies, mine safety monitoring systems have accumulated a large amount of multi-source heterogeneous data over a long period. However, traditional risk identification technologies only employ data splicing or independent analysis methods, failing to achieve collaborative analysis of multi-source data. Due to the lack of an effective data fusion mechanism, the inherent correlations between different types of data are difficult to uncover, resulting in significant limitations in risk identification. For example, during the advancement of mining faces, the spatiotemporal correlation between sensor monitoring data and geological structural data is not effectively utilized, making it difficult to predict rockburst risks.

[0087] To address the aforementioned issues, it was first observed that existing methods suffer from isolated analysis limitations when processing multi-source data, failing to establish cross-data type correlation models. Further analysis revealed that mine risk propagation exhibits spatiotemporal dynamic characteristics, but traditional static analysis models cannot adapt to the dynamic changes in mining processes. Based on this, an attempt was made to construct a dynamic spatiotemporal graph structure to uniformly represent the spatiotemporal correlations of multi-source data, and a data quality assessment mechanism was introduced to address the issue of inconsistent reliability among heterogeneous data. Ultimately, a technical approach was developed that integrates static data and dynamic mining conditions through multi-level feature modeling.

[0088] Therefore, this application proposes the following technical solution, see below. Figure 2 Taking the server as the executing entity as an example, the steps include the following: 201. In response to the mine risk monitoring instructions for the target mine, acquire multi-source heterogeneous monitoring data of the target mine; 202. Process the multi-source heterogeneous monitoring data to obtain the data quality assessment matrix, standardized data stream, and feature-enhanced data of the multi-source heterogeneous monitoring data; 203. Based on the data quality assessment matrix and feature enhancement data, determine the multi-level feature representation of the dynamic spatiotemporal map of the target mine; 204. Based on multi-level feature representation, data quality assessment matrix and real-time mining status of the target mine, determine the risk identification results of the target mine.

[0089] The multi-source heterogeneous monitoring data includes sensor monitoring data, geological structure data, mining process data, and environmental monitoring data. The data quality assessment matrix includes quality assessment parameters across four dimensions: equipment operating status, spatiotemporal consistency, data completeness, and abnormal fluctuations. Feature-enhanced data includes time-domain statistical features, frequency-domain energy features, and spatiotemporal correlation features. Multi-level feature representation includes node-level micro-features, sub-graph-level regional features, and graph-level macro-features. Real-time mining status includes the location of the mining face, advance speed, and equipment distribution information. Multi-source heterogeneous monitoring data refers to mine operation data acquired through different acquisition devices and data sources. Specifically, it can be achieved by using sensor networks to collect equipment operating parameters, 3D geological modeling systems to acquire structural data, mining equipment monitoring systems to record process parameters, and environmental monitoring stations to collect meteorological data, covering all elements of the mine's physical environment and production activities. The data quality assessment matrix is ​​a quantitative evaluation system for the reliability of raw data. It can be constructed through four dimensions: equipment operating status scoring, spatiotemporal consistency verification, missing data ratio calculation, and abnormal fluctuation detection. It is used to screen high-reliability data and correct low-quality data. Feature-enhanced data refers to deeply correlated features extracted through feature engineering. Specifically, it can be generated using methods such as time-domain mean and variance calculation, frequency-domain wavelet transform, and spatiotemporal correlation modeling, transforming the implicit correlations of heterogeneous data into computable features. Dynamic spatiotemporal graphs refer to graph structures representing the spatiotemporal dynamic relationships of a mining system. Specifically, they can be constructed by defining monitoring points as nodes and spatiotemporal correlations as edges, supporting dynamic updates of node features and edge weights. Multi-level feature representation refers to data structures describing risk characteristics at different scales. Specifically, it can be achieved by extracting node-level local features, subgraph-level regional features, and graph-level global features through graph neural networks, forming a multi-dimensional risk analysis framework. Real-time mining status refers to a set of parameters reflecting the current dynamics of mining operations. Specifically, it can be obtained in real-time from the face positioning system, equipment operation logs, and production scheduling system, used to dynamically adjust the risk propagation model.

[0090] Specifically, upon receiving a risk monitoring instruction, the system first integrates four types of monitoring data from different systems. Valid sensor data is filtered through equipment operating status scores, and spatiotemporal alignment is performed based on the spatial distribution of geological structures to eliminate timestamp biases and spatial benchmark differences between different data sources. The standardized data stream undergoes temporal statistics and frequency domain transformation to extract feature components characterizing equipment status fluctuations and geological stress changes. A spatiotemporal correlation model is constructed based on mining process parameters to quantify the impact of mining activities on the surrounding area. The processed feature data is mapped to node attributes and edge weights of a dynamic spatiotemporal graph. A graph convolutional network is used to extract node-level features characterizing the status of individual monitoring points, subgraph-level features reflecting regional correlations, and graph-level features describing the overall risk situation. Finally, the parameters of the risk propagation model are dynamically adjusted based on the working face advance speed and equipment distribution information to output risk identification results matching the current mining activities.

[0091] Compared to related technologies, traditional methods, which rely on single data source analysis or simple data overlay, fail to establish cross-data type correlation models. This solution, however, achieves unified representation and collaborative analysis of multi-source data by constructing a dynamic spatiotemporal graph structure. While related technologies lack systematic data quality assessment, this solution utilizes a four-dimensional quality assessment matrix to achieve quantitative control of data reliability. Compared to traditional static risk models, this solution integrates real-time mining status parameters, enabling risk identification results to dynamically adapt to changes in mining operations.

[0092] Through the above technical solutions, this application achieves effective fusion and collaborative analysis of multi-source heterogeneous monitoring data, which to some extent solves the data silo problem in traditional methods. By using multi-level feature modeling of dynamic spatiotemporal maps, the spatiotemporal propagation patterns of mine risks are captured. Combined with a dynamic correction mechanism based on real-time mining status, the adaptability of risk identification results to complex mining environments is improved. The application of a data quality assessment matrix reduces the impact of low-quality data on analysis results to some extent, improving the reliability of risk warnings.

[0093] This application further proposes the following technical solutions, see [link / reference] Figure 3 Taking the server as the executing entity as an example, the following steps are included.

[0094] 301. Conduct quality assessment and reliability analysis on multi-source heterogeneous monitoring data to obtain a data quality assessment matrix and reliable data stream; 302. Based on geological structure data and mining technology data, spatiotemporal alignment and multi-scale feature enhancement are performed on the reliable data stream to obtain standardized data stream and feature-enhanced data.

[0095] Quality assessment and reliability analysis refer to evaluating data reliability across four dimensions: equipment operating status, spatiotemporal consistency, data completeness, and abnormal fluctuations. This can be achieved through sensor operating status scoring, spatial correlation analysis, time series consistency detection, and missing data imputation methods, used to filter out high-reliability data streams. Spatiotemporal alignment involves unifying data from different sources to the same spatiotemporal benchmark, achieved through spatial interpolation algorithms and time series resampling methods, used to eliminate differences in spatial location and temporal sampling frequency between different monitoring devices. Multi-scale feature enhancement involves extracting data features from three dimensions: time domain, frequency domain, and spatiotemporal correlation. This can be achieved through time-domain statistical calculations, fast Fourier transform, and spatial attenuation models based on geological structures, used to enhance the expression of correlation characteristics between data.

[0096] Specifically, the process begins by calculating the sensor's operational status score using equipment operating parameters. This score, combined with geological structure data and environmental monitoring data, generates spatial and temporal consistency indicators, forming a data quality assessment matrix. Based on this matrix, the raw data is filtered and repaired to obtain a reliable data stream. Subsequently, using fault distribution information from the geological structure data and mining progress information from the mining process data, spatial interpolation and time series alignment are performed on the reliable data stream to generate a standardized data stream under a unified spatiotemporal benchmark. Further, through temporal statistical feature extraction, frequency domain energy analysis, and correlation feature calculation based on a spatial decay model, feature-enhanced data containing multi-dimensional information is fused. For example, the spatial decay model can use an exponential decay function to quantify the constraint strength of geological structures on monitoring points, and the temporal influence factor can be dynamically adjusted through historical mining process parameters.

[0097] Compared to related technologies, existing methods typically only assess the quality of a single data source, failing to consider the spatiotemporal correlations between multiple data sources, thus limiting the effectiveness of data fusion. This proposed solution, however, introduces geological structure and mining process data to guide spatiotemporal alignment, combined with multi-scale feature enhancement methods, which to some extent addresses the issues of inconsistent benchmarks and insufficient feature representation in collaborative analysis of multi-source data.

[0098] Through the above technical solutions, this application can screen out high-reliability data streams and eliminate spatiotemporal benchmark differences to form standardized data inputs; at the same time, by enhancing multi-dimensional features, it can fully explore the spatiotemporal correlation characteristics between data, providing a high-quality feature expression basis for subsequent risk identification models, thereby improving the characteristics of risk situation perception in complex mining environments.

[0099] This application further proposes to conduct quality assessment and credibility analysis on multi-source heterogeneous monitoring data, and obtain a data quality assessment matrix and a credible data stream. This includes determining the working status score of each sensor data based on the equipment operating parameters in the sensor monitoring data, determining the spatial consistency index based on the spatial correlation between geological structure data and sensor monitoring data, determining the temporal consistency index based on the temporal series consistency between environmental monitoring data and sensor monitoring data, generating a data quality assessment matrix based on the working status score, spatial consistency index, and temporal consistency index, and filtering and repairing the multi-source heterogeneous monitoring data based on the data quality assessment matrix to obtain a credible data stream.

[0100] The operational status score is a reliability indicator calculated by analyzing the operating parameters of sensor equipment, such as voltage, signal strength, and sampling frequency. It can be implemented using a weighted scoring model, which maps different parameters to standardized scores and then sums them in a weighted manner to reflect whether the sensor hardware is functioning normally. The spatial consistency index is a quantitative value derived by calculating the degree of matching between sensor monitoring data and geological structure data in spatial distribution. It can be achieved by comparing spatial interpolation algorithms with geological structure models to identify data deviations caused by changes in geological conditions. The temporal consistency index is a synchronization measure derived by comparing the temporal correlation between environmental monitoring data and sensor data. It can be achieved by using dynamic time warping algorithms to calculate the similarity of time series and to identify data acquisition timing misalignments. The data quality assessment matrix is ​​a multi-dimensional assessment table composed of the operational status score, spatial consistency index, and temporal consistency index. It can be stored using a matrix data structure to comprehensively reflect the quality level of different data sources. The reliable data stream refers to the set of valid data retained after filtering and repair. It can be generated using a combination of threshold filtering and interpolation completion to ensure the reliability of subsequent data analysis.

[0101] Specifically, in the quality assessment process, the working status score is first calculated based on the operating parameters of the sensor equipment. For example, the score weight is reduced when the voltage is below a threshold, and a score reduction mechanism is triggered when the signal strength fluctuates abnormally, thus directly reflecting the impact of hardware status on data quality. Next, geological structural data and sensor monitoring data are spatially overlaid and analyzed. For example, after generating a geological stress field model through Kriging interpolation, the spatial correlation coefficient between the sensor data and the model's predicted values ​​is calculated to quantify the data's spatial rationality. Simultaneously, environmental monitoring data and sensor data are temporally aligned. For example, a sliding window is used to calculate the trend similarity between the two data points within the same time period to identify time misalignment issues caused by acquisition delays. The assessment results from these three dimensions are integrated into a data quality assessment matrix. For example, a three-dimensional score vector is generated after normalization processing, forming a comprehensive quality evaluation for each data point. Based on this matrix, the original data is filtered, for example, abnormal data points with scores below a preset threshold are removed, and missing data is repaired through interpolation based on spatiotemporal correlation, ultimately generating a reliable data stream with spatiotemporal consistency.

[0102] Compared to related technologies, traditional methods typically focus on a single dimension of data quality assessment, such as judging data validity solely based on equipment status or performing simple time-series verification. This solution innovatively integrates assessment indicators from three dimensions: equipment operating status, spatial geological correlation, and temporal environmental correlation, constructing a comprehensive quality assessment system covering the entire data acquisition chain from acquisition to application. For example, related technologies fail to consider the spatial constraints of geological structures on sensor data, making it difficult to effectively identify data anomalies in fault areas. This solution, however, quantifies the degree of matching between geological conditions and monitoring data through spatial consistency indicators, improving the accuracy of data quality assessment in complex geological environments. Furthermore, traditional data restoration methods often employ single interpolation algorithms, while this solution combines multi-dimensional information from the data quality assessment matrix to dynamically select the optimal restoration strategy. For instance, spatial interpolation is used for areas with high spatial consistency, while temporal prediction is used for periods with high temporal consistency.

[0103] Through the aforementioned technical solutions, this application addresses, to some extent, the obstacle to correlation mining caused by insufficient reliability in multi-source data collaborative analysis. By employing a multi-dimensional quality assessment system to accurately identify data quality issues such as equipment failure, spatial misalignment, and temporal deviations, and combining this with a dynamic filtering and repair mechanism to generate a highly reliable data stream, a reliable data foundation is provided for subsequent spatiotemporal alignment and feature enhancement. For example, in areas with aging sensor nodes, the integrity of the data stream is maintained by reducing their data weight and supplementing data from adjacent nodes; in areas with complex geological structures, spatial consistency analysis corrects abnormal values ​​in monitoring data caused by rock fracture zones; and during periods of abrupt changes in environmental parameters, temporal consistency verification avoids data contradictions caused by acquisition delays. These technical effects collectively ensure the inherent consistency of multi-source heterogeneous data in the spatiotemporal dimensions, enhancing the adaptability of the risk identification model to complex mining environments.

[0104] This application further proposes a method for spatiotemporal alignment and multi-scale feature enhancement of trusted data streams based on geological structural data and mining process data. The method includes spatiotemporal alignment of trusted data streams to obtain data sequences under a unified spatiotemporal reference, extraction of temporal statistical features and frequency domain energy features from the data sequences, determination of spatiotemporal correlation feature components based on geological structural data and mining process data, fusion of temporal feature components, frequency domain feature components and spatiotemporal correlation feature components to obtain feature-enhanced data, and standardization of trusted data streams based on feature-enhanced data to obtain standardized data streams.

[0105] Spatiotemporal alignment refers to the process of mapping reliable data from different sources to a unified spatiotemporal coordinate system. This can be achieved using spatiotemporal interpolation algorithms or coordinate transformation matrices to eliminate spatial reference differences between sensor data and geological structural data. Temporal statistical features refer to the mean, variance, and kurtosis indices extracted from time-series data. This can be achieved using sliding window statistical methods to characterize the fluctuation characteristics of monitoring data. Frequency domain energy features refer to the frequency band energy distribution characteristics extracted through Fourier transform. This can be achieved using wavelet packet decomposition methods to reveal the periodic variation patterns of monitoring signals. Spatiotemporal correlation feature components refer to feature vectors constructed using spatial attenuation models and temporal influence factors. This can be achieved by combining the geological structural spatial constraint matrix with the mining process influence coefficient to quantify the spatial influence of geological fault zones on monitoring points and the dynamic impact of mining speed on the monitoring status.

[0106] Specifically, the spatiotemporal alignment process first establishes a unified spatiotemporal benchmark including a geographic coordinate system, aligning sensor data from different acquisition frequencies to the same timestamp using linear interpolation, while simultaneously converting geological structural data into a three-dimensional grid coordinate system. The extraction of temporal statistical features employs a sliding time window to calculate the mean, variance, and kurtosis of each monitoring point. For example, for microseismic monitoring data, the window length can be set to ten minutes to capture short-term characteristics of rock mass stress changes. The extraction of frequency domain energy features decomposes the time-domain signal into multiple frequency bands using a fast Fourier transform, calculating the energy proportion of each band. For instance, the energy in the 0-10Hz band is used as a characteristic indicator of low-frequency vibrations. The construction of spatiotemporal correlation feature components includes two core parts: calculating the spatial influence weight of geological structures on monitoring points based on a Gaussian attenuation model, where the attenuation coefficient is determined based on the positional relationship between the fault zone and the monitoring point; and establishing a time-series influence factor matrix based on historical mining records to reflect the cumulative effect of process parameters at different mining stages on the state of the monitoring points. The feature fusion process employs feature concatenation and normalization, concatenating time-domain feature vectors, frequency-domain feature vectors, and spatiotemporally correlated feature vectors, followed by batch normalization to form feature-enhanced data with a uniform scale. In the standardization stage, based on the statistical characteristics extracted from the feature-enhanced data, Z-score standardization is applied to the original reliable data stream to eliminate dimensional differences between different monitoring parameters.

[0107] Compared to related technologies, existing methods typically use fixed spatial grids for data alignment, which cannot adapt to the dynamically changing distribution of monitoring points during mining operations, and the analysis of the correlation between geological structures and mining processes is mostly static. This scheme introduces a dynamic spatiotemporal benchmark transformation mechanism, which can automatically adapt to changes in spatial coordinates during mining operations. Simultaneously, the constructed spatiotemporal correlation feature components effectively integrate the spatial constraints of geological structures and the dynamic influence of mining processes. For example, an exponential decay model is used to characterize the gradient influence of fault zones on monitoring data in surrounding areas, which is more consistent with actual geomechanical laws than traditional methods with fixed influence radii. Furthermore, the time-series influence factor matrix constructed based on historical process data can dynamically reflect the influence weights of process parameters on monitoring data at different mining stages, improving the timeliness of feature representation compared to the static weight allocation methods used in related technologies.

[0108] Through the aforementioned technical solution, this application addresses, to some extent, the challenge of collaborative analysis caused by inconsistencies in the spatiotemporal benchmarks of multi-source data. By constructing spatiotemporal correlation features that incorporate spatial constraints of geological structures and the dynamic influence of mining processes, the correlation between monitoring data and actual mine conditions is improved. This solution not only achieves spatiotemporal consistency processing of multi-source heterogeneous data, but more importantly, by quantifying the dynamic coupling effect of geological structures and mining processes, it enables subsequent risk identification models to capture the causal relationship between changes in rock mass stress and mining activities, providing a reliable data foundation for accurately identifying mine risks such as roof falls and water inrushes.

[0109] This application further proposes the following technical solutions, see [link / reference] Figure 4 Taking the server as the executing entity as an example, the following steps are included.

[0110] 401. Construct multi-scale node feature tensors and multi-modal edge relation matrices for the dynamic spatiotemporal graph of the target mine based on the data quality assessment matrix and feature-enhanced data; 402. Spatiotemporal feature extraction and cross-modal fusion are performed on multi-scale node feature tensors and multi-modal edge relation matrices to obtain multi-level feature representations.

[0111] The multimodal edge relationship matrix includes five edge types: spatial topological relationships, geological structural associations, stress field coupling relationships, mining disturbance propagation, and data flow correlation. It refers to a heterogeneous graph structure relationship matrix containing these five edge types. Specifically, it can be implemented using spatial proximity calculation, geological structural similarity measurement, stress field coupling coefficient modeling, mining disturbance propagation path tracing, and data flow correlation analysis. This matrix describes the interaction relationships between mine monitoring nodes from different dimensions. The multi-scale node feature tensor refers to a multidimensional data representation that integrates node-level, region-level, and global-level features. This can be achieved through weighted node feature extraction, adjacent node feature aggregation, and full-graph feature statistical analysis. It is used to capture risk characteristics at different levels, from micro to macro. Node-level micro features represent the feature information of a single node in the dynamic spatiotemporal graph; subgraph-level region features represent the feature information of a sub-region composed of adjacent nodes in the dynamic spatiotemporal graph; and graph-level macro features represent the overall feature information of the dynamic spatiotemporal graph. Spatiotemporal feature extraction and cross-modal fusion refer to interactive modeling of local feature patterns in the spatial dimension and evolutionary trends in the temporal dimension. Specifically, graph convolutional networks can be used to extract spatial features, temporal attention mechanisms can be used to capture dynamic changes, and feature fusion can be achieved through cross-modal attention weight allocation to generate a unified multi-level risk representation.

[0112] Specifically, in constructing the dynamic spatiotemporal map, initial node feature vectors are first generated based on feature-enhanced data, fusing temporal statistical features, frequency domain energy features, and spatiotemporal correlation features. Then, a data quality assessment matrix is ​​used to weight the node features based on reliability, for example, using equipment operating status scores as weighting coefficients to suppress the interference of low-quality data on feature construction. Next, a multi-scale feature extraction method is employed to perform local aggregation and global statistics on the weighted node features, forming feature tensors covering node, region, and global levels. Simultaneously, a multimodal edge relationship matrix is ​​constructed based on geological structural correlations and mining disturbance propagation paths, for example, using a Gaussian model to quantify the spatial constraint strength of geological structures on monitoring points, or calculating temporal influence factors based on historical mining data. Finally, a spatiotemporal feature extraction module jointly models node features and edge relationships, utilizing a cross-modal attention mechanism to fuse spatial propagation features and temporal evolution features, generating a multi-level feature representation that includes micro-node states, regional risk distribution, and global trends.

[0113] Compared to related technologies, traditional methods typically construct mine monitoring networks using only a single spatial topological relationship, failing to reflect the impact of physical mechanisms such as geological structural relationships and stress field coupling on risk propagation. For example, edge relationships in related technologies rely solely on the spatial proximity of sensor locations, ignoring the propagation paths of disturbances caused by dynamic changes in mining processes. This approach, by introducing five edge types, can simultaneously model the mine's physical spatial structure, geomechanical relationships, and the dynamic impacts of mining. For instance, stress field coupling edges capture the chain effect of rock stress changes on adjacent areas, or data flow correlation edges identify the intrinsic correlations across sensor data. Furthermore, existing methods often employ single-scale feature analysis, making it difficult to distinguish between local anomalies and global trends. This approach, however, achieves hierarchical feature representation from micro to macro through multi-scale node feature tensors. For example, node-level features retain fine-grained fluctuation information from sensors, regional-level features reflect local rock stratum stability changes, and graph-level features characterize the overall risk propagation trend.

[0114] Through the above technical solutions, this application addresses to some extent the problem that traditional methods cannot effectively integrate the spatiotemporal correlation features and quality assessment parameters of multi-source data, achieving multi-scale feature capture of mine risks in micro-nodes, regional subgraphs, and macro-graph structures. Specifically, by modeling geological structural correlations and mining disturbance propagation paths using a multimodal edge relation matrix, the propagation mechanism of risks in physical space and mining dynamics can be identified; through reliability-weighted processing and multi-scale feature extraction, low-quality data interference can be suppressed and risk features can be expressed hierarchically; through a spatiotemporal feature fusion mechanism, both local spatial anomaly patterns and temporal evolution trends can be captured simultaneously, providing multi-dimensional feature support for risk identification. For example, in the scenario of roof collapse risk identification, node-level features can detect local rock layer displacement anomalies, subgraph-level features can analyze stress concentration phenomena in adjacent areas, and graph-level features can assess overall stability trends, thereby improving the comprehensiveness and accuracy of risk identification.

[0115] This application further proposes constructing initial node features based on feature-enhanced data. Each initial node feature vector is formed by fusing temporal statistical features, frequency energy features, and spatiotemporal correlation features of the corresponding location. The initial node features are weighted based on the data quality assessment matrix to obtain weighted node features. Multi-scale feature extraction is performed on the weighted node features to obtain multi-scale node feature tensors. Node-level features are directly obtained from the weighted node features, regional-level features are obtained by aggregating the features of adjacent nodes, and global-level features are obtained by statistical analysis of the full-map features. A multimodal edge relationship matrix is ​​constructed based on the spatiotemporal correlation feature components in the feature-enhanced data.

[0116] Among them, time-domain statistical features refer to the mean, variance, and extreme value features extracted from time-series data, which can be implemented using the sliding window statistical method to capture short-term fluctuation patterns in monitoring data. Frequency-domain energy features refer to the spectral energy distribution features extracted through Fourier transform, which can be implemented using the Fast Fourier Transform algorithm to characterize the periodic variation patterns of monitoring data. Spatiotemporal correlation feature components refer to quantitative indicators reflecting the geological structural correlation and mining disturbance propagation intensity between monitoring points, which can be calculated using a spatial attenuation model combined with temporal influence factors to model cross-regional dynamic correlation relationships. Reliability weighting processing refers to generating feature weight coefficients based on equipment operating status and spatiotemporal consistency parameters in the data quality assessment matrix, which can be implemented using linear weighting or entropy weighting methods to suppress the interference of low-quality data on feature expression. The multimodal edge relationship matrix refers to a dynamic relationship network containing various edge types such as geological structural correlation and mining disturbance propagation, which can be constructed using adjacency matrix and tensor decomposition methods to characterize complex spatiotemporal interaction patterns.

[0117] Specifically, the initial node features, by fusing temporal statistical features, frequency domain energy features, and spatiotemporal correlation features, address the issue of insufficient representational power of a single feature dimension to some extent. For example, temporal features capture sudden anomalies in equipment vibration data, frequency domain features identify periodic stress changes, and spatiotemporal correlation features reflect the intensity of the impact of fault zones on monitoring points. Reliability-weighted processing transforms equipment operating status scores and spatiotemporal consistency indicators into feature weights; for example, reducing the data weight of sensors in case of failure avoids outliers interfering with model training. Multi-scale feature extraction employs node-level preservation of original feature details, region-level aggregation of the mean of adjacent nodes, and global-level calculation of variance statistics, forming a feature expression hierarchy from micro to macro. The multimodal edge relationship matrix reflects fault spatial constraints through geological structural correlation edges and simulates the chain effects of blasting operations through mining disturbance propagation edges; for example, a Gaussian model is used to quantify the fault influence range, and time-series factors are used to characterize the dynamic impact of mining progress on the surrounding area.

[0118] Compared to related technologies, traditional methods only use the temporal characteristics of sensor data for node modeling, failing to integrate frequency domain energy characteristics and geological mining correlation characteristics, resulting in missing feature information. Node weighting in related technologies is often based on a single quality indicator, while this scheme uses multi-dimensional parameters such as equipment status and spatiotemporal consistency for dynamic weight allocation. Existing edge relationship construction is limited to physical spatial topology, while this scheme introduces multi-modal edge types such as geological structural correlation and mining disturbance propagation, more accurately reflecting the dynamic risk propagation path in the mine. For example, existing methods cannot characterize the impact of fault zones on non-adjacent areas, while this scheme establishes cross-regional correlation edges through a spatial attenuation model, effectively capturing the implicit risk transmission of geological structures.

[0119] Through the above technical solutions, this application achieves deep feature fusion and dynamic spatiotemporal correlation modeling of multi-source monitoring data, which to some extent solves the problems of one-sided feature expression and coarse correlation modeling in mine risk identification. The complementary fusion of time-domain and frequency-domain features enhances the comprehensive representation capability of node states, the reliability weighting mechanism reduces the negative impact of data noise on model training, multi-scale feature extraction realizes hierarchical expression from local details to global situation, and the construction of multi-modal edge relationships breaks through the limitations of single spatial topology, characterizing the complex interaction relationship caused by geological structure and mining activities. For example, in the scenario of roof collapse risk identification, this method can simultaneously capture abnormal vibration of support equipment, frequency-domain energy change of roof stress, and the propagation characteristics of mining disturbances in adjacent areas, improving the accuracy and timeliness of risk warning.

[0120] This application further proposes spatiotemporal feature extraction and cross-modal fusion of multi-scale node feature tensors and multi-modal edge relation matrices to obtain multi-level feature representations. This includes spatial feature propagation and aggregation of multi-scale node feature tensors based on spatial topological relationships and geological structural correlation edges in the multi-modal edge relation matrix, resulting in local feature patterns and regional feature distributions in the spatial dimension; temporal feature modeling and evolution analysis of multi-scale node feature tensors based on data flow correlation edges and mining disturbance propagation edges in the multi-modal edge relation matrix, resulting in short-term fluctuation features and long-term trend features in the time dimension; and spatiotemporal feature interaction and multi-scale fusion through a cross-modal attention mechanism based on local feature patterns and regional feature distributions in the spatial dimension, as well as short-term fluctuation features and long-term trend features in the time dimension, to generate a unified multi-level feature representation.

[0121] The multimodal edge relation matrix refers to a dynamic relational network encompassing four edge types: spatial topological relationships, geological structural associations, mining disturbance propagation, and data flow correlations. Specifically, it can be implemented using an adjacency matrix extension form from graph convolutional networks, modeling the interactions of geological structural constraints, mining process influences, and data flow correlations through different edge types. The multi-scale node feature tensor refers to a multi-level data representation fusing node-level micro-features, subgraph-level regional features, and graph-level macro-features. This can be implemented using graph pooling operations and feature pyramid structures to capture risk feature patterns at different spatial scales. The cross-modal attention mechanism is a fusion method that dynamically assigns weights to spatiotemporal features. Specifically, it can be implemented using a multi-head self-attention mechanism combined with spatiotemporal position encoding, achieving adaptive fusion by calculating the correlation degree of features across different spatiotemporal dimensions.

[0122] Specifically, in the spatial dimension processing stage, spatial topological relationship edges are used to construct physical connections between nodes, while geological structure association edges establish geological constraints based on strata strike and fault distribution. Node features are propagated along these two edge types through graph convolution operations, forming a feature distribution reflecting local stress concentration and regional tectonic stability. In the temporal dimension processing stage, data flow association edges are used to model the temporal correlation of sensor data, and mining disturbance propagation edges characterize the dynamic impact of mining activities on the surrounding area. Anomalous signals in short-term data fluctuations are extracted using a temporal graph neural network, and long short-term memory networks are combined to capture long-term trend changes resulting from mining progress. The spatiotemporal interaction stage employs a cross-modal attention mechanism to correlate the regional feature distribution in the spatial dimension with the trend features in the temporal dimension, generating node embedding vectors that integrate spatiotemporal dynamics. Ultimately, this results in a multi-level feature representation encompassing microscopic anomalies, regional propagation paths, and global evolutionary trends.

[0123] Compared to related technologies, traditional methods typically employ separate spatial feature extraction and time series analysis modules, resulting in insufficient spatiotemporal feature interaction. For example, spatial feature modeling in related technologies only considers physical adjacency relationships while ignoring geological structural constraints, and time feature analysis fails to distinguish the impact of mining disturbances and data noise. This scheme constructs a multimodal edge relationship matrix, incorporating geological structural associations and mining disturbance propagation into the graph structure. This allows spatial feature propagation to reflect rock strata mechanical constraints, and time series modeling to distinguish between process dynamics and random fluctuations. Furthermore, the cross-modal attention mechanism breaks through the traditional weighted average fusion method, achieving adaptive interaction by dynamically calculating the correlation strength of spatiotemporal features, thus solving to some extent the problem that static fusion weights are difficult to adapt to dynamic changes in the mine.

[0124] Through the above technical solution, this application achieves spatiotemporal collaborative modeling of mine risk characteristics, enabling risk identification to simultaneously capture abnormal fluctuations at local nodes, risk propagation paths within a region, and trend changes caused by global mining activities. Specifically, in the spatial dimension, the accuracy of identifying regional risk propagation paths is enhanced through geological structural correlation edges; in the temporal dimension, the ability to characterize the impact of mining technology dynamics on risk evolution is improved through mining disturbance propagation edges; and at the spatiotemporal interaction level, a cross-modal attention mechanism is used to establish dynamic correlations of risk characteristics at multiple scales, thereby solving to some extent the problem of lack of dynamism caused by the fragmented analysis of spatiotemporal features in traditional methods.

[0125] This application further proposes the following technical solutions, see [link / reference] Figure 5 Taking the server as the executing entity as an example, the following steps are included.

[0126] 501. Conduct risk propagation mechanism analysis and dynamic evolution prediction on multi-level feature representation and data quality assessment matrix to obtain risk level distribution map and risk evolution prediction trajectory; 502. Based on the risk level distribution map, risk evolution prediction trajectory, multi-level feature representation and real-time mining status, combined with the data quality assessment matrix, risk identification and confidence assessment are carried out to obtain risk identification results.

[0127] The risk identification results represent the identification information for four risk types: roof fall risk, water inrush risk, gas outburst risk, and rock burst risk. Risk propagation mechanism analysis refers to establishing a risk propagation model through spatiotemporal correlation characteristics and data quality parameters, specifically using dynamic equations or graph neural networks, to quantify the diffusion path and intensity of risk in the mining space. Dynamic evolution prediction refers to constructing a prediction model based on temporal characteristics and mining disturbance factors, specifically using time series prediction algorithms or recurrent neural networks, to simulate the trend of risk status changes over time. The risk level distribution map is a visual representation that maps node-level risk indicators to geographic space, specifically using interpolation algorithms or heatmap generation technology, to intuitively display the risk level in different areas. Credibility assessment refers to correcting the reliability of the risk identification results by combining data quality parameters, specifically using weighted scoring mechanisms or probability calibration methods, to eliminate the interference of low-quality data on the judgment results.

[0128] Specifically, the system first captures local anomaly signals through node-level micro-features, such as sudden stress changes or a sharp increase in gas concentration at a monitoring point. Simultaneously, it analyzes the correlation changes between adjacent nodes using sub-graph-level regional features, such as the spatial expansion trend of rock fracture zones. Next, it assesses the overall risk situation based on graph-level macro-features, such as the matching degree between the stress field distribution across the mine and historical accident patterns. During the dynamic evolution prediction process, the location and advance speed of the mining face in real-time mining conditions are input into the prediction model as boundary conditions. For example, when the mining face approaches an aquifer, the propagation coefficient of water permeability risk is automatically adjusted. A data quality assessment matrix continuously influences risk calculation throughout the process; for example, nodes in sensor failure areas are assigned lower weights to avoid abnormal data affecting the prediction results. Finally, the confidence assessment module outputs a set of risk types with probability values, such as an 85% probability of roof collapse and a 72% probability of rockburst, while also generating vector boundary data of the risk impact range.

[0129] Compared to related technologies, traditional methods typically analyze static monitoring data and mining plans independently, lacking dynamic modeling of risk propagation paths. For example, they may assess risk solely based on current gas concentration while ignoring gas diffusion changes caused by mining progress. This solution, by integrating multi-level features and real-time mining status, constructs a spatiotemporally continuous risk evolution model capable of predicting chain reactions triggered by mining disturbances. Related technologies often rely on single indicator thresholds for risk classification, such as judging roof collapse risk solely based on roof displacement. However, this solution, through correlation analysis of sub-map-level regional features, can identify complex risk causes, such as water permeability risk resulting from the combined effects of rock fracture and groundwater seepage.

[0130] Through the above technical solutions, this application achieves accurate differentiation of compound risk types, such as simultaneously identifying the superposition of roof fall risk and rockburst risk during the advance of the mining face; dynamically corrects the risk impact range, such as adjusting the diffusion radius of gas outburst risk in real time according to equipment distribution; improves the credibility of risk identification results, such as automatically reducing the confidence level of risk prediction results for areas with insufficient sensor coverage; and establishes a correlation model between risk evolution and mining activities, such as predicting the probability change curve of rockburst risk under different advance speeds.

[0131] This application further proposes a method for modeling and dynamically predicting mine risk propagation, including: extracting spatial correlations and risk propagation paths between regions based on subgraph-level regional features in multi-level feature representation, wherein the propagation path is determined by analyzing the spatial continuity and evolution synchronicity of features of adjacent subgraphs; determining the dynamic trend and intensity of global risk propagation based on graph-level macro features in multi-level feature representation; initializing the risk status of each node in the dynamic spatiotemporal graph based on node-level micro features in multi-level feature representation, and using a data quality assessment matrix to weighted correct the credibility of the node status; constructing a risk propagation dynamic equation based on the risk propagation path, global dynamic trend, and weighted corrected node risk status, and obtaining a risk level distribution map and risk evolution prediction trajectory by numerically simulating the risk propagation dynamic equation.

[0132] The subgraph-level regional features refer to the feature information of sub-regions composed of adjacent nodes in a dynamic spatiotemporal graph. Specifically, this can be achieved by aggregating and calculating the features of adjacent nodes using graph convolutional networks, used to characterize the risk transmission patterns within local regions. Spatial continuity and evolutionary synchronicity refer to the coherence of spatial distribution and the synergy in temporal evolution of adjacent subgraphs. This can be achieved by calculating the cosine similarity and dynamic temporal warping distance of feature vectors between subgraphs, used to identify potential risk propagation paths. The data quality assessment matrix is ​​a quality assessment parameter matrix containing four dimensions: equipment operating status, spatiotemporal consistency, data completeness, and abnormal fluctuations. Specifically, this can be achieved by using the analytic hierarchy process (AHP) to weight and score multi-source data, used to correct the credibility weights of node risk states. The risk propagation dynamics equation is a multivariate differential equation that integrates spatial propagation paths, global dynamic trends, and node states. Specifically, this can be achieved by using partial differential equations to model spatial propagation effects and temporal evolution patterns, used to quantify the intensity and speed of risk propagation.

[0133] Specifically, within the dynamic spatiotemporal graph framework, subgraph-level regional features identify highly correlated risk transmission paths by analyzing the topological connections and temporal evolution trends of adjacent subgraphs in spatial distribution. For example, when the stress field changes of adjacent subgraphs exhibit synchronous fluctuations, it can be determined that a risk diffusion channel exists in that area. Graph-level macro features establish a dynamic model of global risk propagation by statistically analyzing the stress field distribution patterns and mining disturbance propagation directions across the entire graph. Node-level micro features initialize the risk state values ​​of each monitoring point using sensor monitoring data, and deweight the states of low-confidence nodes using a data quality assessment matrix. The final constructed dynamic equation uses the spatial propagation path as the diffusion coefficient, the global dynamic trend as the driving term, and the weighted node states as initial conditions. It is numerically solved using the finite difference method to generate a risk distribution heatmap and propagation trajectory prediction curve for future time periods.

[0134] Compared to related technologies, traditional methods often employ single-dimensional statistical analysis or static risk assessment models, making it difficult to capture the dynamic transmission effects between geologically connected regions. For example, related technologies typically analyze risk indicators at each monitoring point independently, neglecting the chain reaction of mining disturbances in the spatial network. This proposed solution, however, constructs a dynamic spatiotemporal map model that integrates multi-scale features, enabling simultaneous characterization of transmission paths in local areas and dynamic trends in the global system. While related technologies often handle the impact of data quality differences on risk assessment through simple threshold filtering, this solution innovatively introduces a data quality assessment matrix to dynamically weight and correct node states, effectively improving the characteristics of risk state initialization.

[0135] Through the above technical solutions, this application achieves accurate modeling and dynamic evolution prediction of mine risk propagation paths. Specifically, it is reflected in: risk transmission path identification based on spatial continuity and evolution synchronicity analysis, which can capture the risk diffusion patterns of geological fault zones or stress concentration areas; the construction of dynamic equations that integrate global dynamic trends and local node states, which realizes the quantification of risk propagation intensity under the coupling effect of mining disturbance and geological structure; and the combination of a dynamic weighting mechanism for data quality assessment, which reduces the interference of abnormal sensor data on the risk prediction model to a certain extent and improves the reliability of the risk level distribution map.

[0136] This application further proposes a technical solution for risk identification and confidence assessment based on risk level distribution maps, risk evolution prediction trajectories, multi-level feature representations, and real-time mining status, combined with a data quality assessment matrix, to obtain risk identification results. Specifically, it includes the following steps: generating preliminary risk classification and level division, local feature verification, risk impact range simulation, dynamic correction, confidence scoring, and result generation.

[0137] The preliminary risk classification and grading process involves integrating the spatial distribution characteristics of the risk grade distribution map with the dynamic trends of the risk evolution prediction trajectory to generate preliminary risk types and grading results. This can be achieved using spatiotemporal clustering algorithms or probabilistic graphical models to overcome the limitations of static risk assessment. Local feature verification utilizes node-level micro-features to perform fine-grained verification of the preliminary classification results. This can be achieved using feature matching algorithms or anomaly detection models to eliminate false risk signals caused by data noise. Risk impact range simulation constructs a risk propagation path model based on sub-map-level regional features. This can be achieved using spatial interpolation algorithms or diffusion equation modeling to quantify the impact of geological structural associations and mining disturbances on risk propagation. Dynamic correction incorporates real-time mining status parameters into the impact range calculation. This can be achieved using dynamic weight adjustment algorithms or real-time parameter fitting methods to eliminate assessment biases caused by dynamic changes in mining activities. Confidence scoring quantifies the risk type and impact range based on a data quality assessment matrix. This can be achieved using weighted scoring algorithms or confidence propagation models to establish interpretability indicators for risk identification results.

[0138] Specifically, this technical solution first generates preliminary classification results using a risk level distribution map and evolutionary prediction trajectory. This is then validated locally using node-level micro-features to screen candidate risk types that match high-resolution monitoring data. Subsequently, a risk propagation model is constructed based on sub-map-level regional features to simulate the impact boundaries under geological structural associations and mining disturbances, and the impact range is dynamically adjusted using real-time mining status parameters. Finally, a data quality assessment matrix is ​​used to weight the validated risk types and corrected impact ranges with confidence, generating risk identification results that include probability distribution and confidence parameters. Each step achieves cross-level verification through data quality indicators and dynamic mining parameters, forming a technical closed loop from global prediction to local verification.

[0139] Compared to related technologies, traditional methods rely solely on a single data source or static model for risk classification, failing to handle the spatiotemporal correlation of multi-source data and lacking adaptability to dynamic changes in mining operations. This solution integrates multi-level feature representations with real-time mining status to construct a dynamic correction mechanism, enabling risk impact range assessment to respond in real-time to changes in the mining face location. Simultaneously, a data quality assessment matrix is ​​introduced to quantify the reliability of the identification results, addressing to some extent the lack of data reliability support in traditional methods for calculating risk probability.

[0140] Through the above technical solutions, this application effectively improves the characteristics of mine risk classification, eliminating misjudgments caused by low-quality data through a local feature verification mechanism. The dynamic correction mechanism makes the risk impact range assessment more closely aligned with actual mining conditions, reducing boundary deviations caused by changes in advance speed or equipment distribution. The credibility scoring transforms data quality parameters into quantifiable confidence indicators, providing an interpretable basis for risk decision-making and, to some extent, addressing the technical deficiency of traditional methods in assessing the credibility of identification results.

[0141] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0142] Figure 6 This is a schematic diagram of the structure of a risk identification system based on mine data provided in an embodiment of this application. See also... Figure 6 The system includes: The acquisition module 601 is used to acquire multi-source heterogeneous monitoring data of the target mine in response to the mine risk monitoring instruction. The multi-source heterogeneous monitoring data includes sensor monitoring data, geological structure data, mining process data and environmental monitoring data. The processing module 602 is used to process multi-source heterogeneous monitoring data to obtain a data quality assessment matrix, a standardized data stream, and feature-enhanced data. The data quality assessment matrix includes quality assessment parameters in four dimensions: equipment operating status, spatiotemporal consistency, data completeness, and abnormal fluctuations. The feature-enhanced data includes time-domain statistical features, frequency-domain energy features, and spatiotemporal correlation features. The feature representation determination module 603 is used to determine the multi-level feature representation of the dynamic spatiotemporal map of the target mine based on the data quality assessment matrix and feature enhancement data. The multi-level feature representation includes node-level micro features, sub-graph-level regional features and graph-level macro features. The identification result determination module 604 is used to determine the risk identification result of the target mine based on multi-level feature representation, data quality assessment matrix and real-time mining status of the target mine. The real-time mining status includes the location of the mining face, advance speed and equipment distribution information.

[0143] It should be noted that the risk identification system based on mine data provided in the above embodiments is only illustrated by the division of the above functional modules when identifying risks. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer equipment can be divided into different functional modules to complete all or part of the functions described above. In addition, the risk identification system based on mine data provided in the above embodiments and the risk identification method embodiments based on mine data belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0144] Figure 7 This is a schematic diagram of a server structure provided in an embodiment of this application. The server 700 can vary significantly due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 701 and one or more memories 702. The one or more memories 702 store at least one computer program, which is loaded and executed by the one or more processors 701 to implement the methods provided in the various method embodiments described above. Of course, the server 700 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server 700 may also include other components for implementing device functions, which will not be elaborated upon here.

[0145] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including a computer program that can be executed by a processor to perform the risk identification method based on mine data in the above embodiments. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0146] In an exemplary embodiment, a computer program product or computer program is also provided, which includes program code stored in a computer-readable storage medium. The processor of a computer device reads the program code from the computer-readable storage medium and executes the program code, causing the computer device to perform the aforementioned risk identification method based on mining data.

[0147] In some embodiments, the computer program involved in the present application embodiments may be deployed and executed on a computer device, or executed on multiple computer devices located in one location, or executed on multiple computer devices distributed in multiple locations and interconnected through a communication network. Multiple computer devices distributed in multiple locations and interconnected through a communication network may constitute a blockchain system.

[0148] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0149] The above are merely optional embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A risk identification method based on mine data, characterized in that, The method comprises: in response to a mine risk monitoring instruction of a target mine, acquiring multi-source heterogeneous monitoring data of the target mine, the multi-source heterogeneous monitoring data comprising sensor monitoring data, geological structure data, mining process data and environmental monitoring data; processing the multi-source heterogeneous monitoring data to obtain a data quality evaluation matrix of the multi-source heterogeneous monitoring data, standardized data flow and feature enhancement data, the data quality evaluation matrix comprising quality evaluation parameters of four dimensions of equipment operation state, spatio-temporal consistency, data completeness and abnormal fluctuation, and the feature enhancement data comprising time domain statistical features, frequency domain energy features and spatio-temporal correlation features; based on the data quality evaluation matrix and the feature enhancement data, determining a multi-level feature representation of a dynamic spatio-temporal graph of the target mine, the multi-level feature representation comprising node-level microscopic features, subgraph-level regional features and graph-level macroscopic features; based on the multi-level feature representation, the data quality evaluation matrix and a real-time mining state of the target mine, determining a risk identification result of the target mine, the real-time mining state comprising mining working face position, advancing speed and equipment distribution information.

2. The method of claim 1, wherein, The processing of the multi-source heterogeneous monitoring data to obtain the data quality evaluation matrix of the multi-source heterogeneous monitoring data, the standardized data flow and the feature enhancement data comprises: performing quality evaluation and reliability analysis on the multi-source heterogeneous monitoring data to obtain the data quality evaluation matrix and reliable data flow; based on the geological structure data and the mining process data, performing spatio-temporal alignment and multi-scale feature enhancement on the reliable data flow to obtain the standardized data flow and the feature enhancement data.

3. The method of claim 2, wherein, The performing of the quality evaluation and reliability analysis on the multi-source heterogeneous monitoring data to obtain the data quality evaluation matrix and the reliable data flow comprises: based on equipment operation parameters in the sensor monitoring data, determining working state scores of each sensor data; based on spatial correlation of the geological structure data and the sensor monitoring data, determining a spatial consistency index; based on time series consistency of the environmental monitoring data and the sensor monitoring data, determining a time consistency index; based on the working state scores, the spatial consistency index and the time consistency index, generating the data quality evaluation matrix; based on the data quality evaluation matrix, performing screening and repair on the multi-source heterogeneous monitoring data to obtain the reliable data flow.

4. The method of claim 2, wherein, The performing of the spatio-temporal alignment and multi-scale feature enhancement on the reliable data flow based on the geological structure data and the mining process data to obtain the standardized data flow and the feature enhancement data comprises: performing spatio-temporal alignment processing on the reliable data flow to obtain data sequences under a unified spatio-temporal reference; extracting time domain statistical features and frequency domain energy features from the data sequences under the unified spatio-temporal reference to obtain time domain feature components and frequency domain feature components; determine a spatio-temporal correlation feature component based on the geological structure data and the mining process data, the spatio-temporal correlation feature component being used to represent a spatio-temporal interaction intensity between monitoring data points based on a geological structure relationship and a mining process dynamics, a spatial decay model and a time sequence influence factor being used to quantify a spatial constraint of the geological structure on the monitoring points and a time sequence influence of the mining process on the monitoring point states, the spatial decay model being an exponential decay model or a Gaussian model, the time sequence influence factor being determined based on historical mining process data; fuse the time domain feature component, the frequency domain feature component and the spatio-temporal correlation feature component to obtain the feature enhanced data; perform standardization processing on the credible data stream based on the feature enhanced data to obtain the standardized data stream.

5. The method of claim 1, wherein, determine a multi-level feature representation of a dynamic spatio-temporal graph of the target mine based on the data quality evaluation matrix and the feature enhanced data, including: construct a multi-scale node feature tensor and a multi-modal edge relationship matrix of the dynamic spatio-temporal graph of the target mine based on the data quality evaluation matrix and the feature enhanced data, the multi-modal edge relationship matrix including five edge types of spatial topological relationship, geological structure correlation, stress field coupling relationship, mining disturbance propagation and data stream correlation; perform spatio-temporal feature extraction and cross-modal fusion on the multi-scale node feature tensor and the multi-modal edge relationship matrix to obtain the multi-level feature representation, the node-level microscopic feature being used to represent feature information of a single node in the dynamic spatio-temporal graph, the sub-graph-level regional feature being used to represent sub-regional feature information of adjacent nodes in the dynamic spatio-temporal graph, and the graph-level macroscopic feature being used to represent overall feature information of the dynamic spatio-temporal graph.

6. The method of claim 5, wherein, construct a multi-scale node feature tensor and a multi-modal edge relationship matrix of the dynamic spatio-temporal graph of the target mine based on the data quality evaluation matrix and the feature enhanced data, including: construct initial node features based on the feature enhanced data, wherein each initial node feature vector is formed by fusing a time domain statistical feature, a frequency domain energy feature and a spatio-temporal correlation feature at a corresponding position; perform reliability weighting processing on the initial node features based on the data quality evaluation matrix to obtain weighted node features; perform multi-scale feature extraction on the weighted node features to obtain the multi-scale node feature tensor, the node-level feature directly using the weighted node features, the regional-level feature being obtained by aggregating adjacent node features, and the global-level feature being obtained by statistical analysis of the whole graph features; construct a multi-modal edge relationship matrix based on the spatio-temporal correlation feature component in the feature enhanced data.

7. The method of claim 5, wherein, perform spatio-temporal feature extraction and cross-modal fusion on the multi-scale node feature tensor and the multi-modal edge relationship matrix to obtain the multi-level feature representation, including: perform spatial feature propagation and aggregation on the multi-scale node feature tensor based on spatial topological relationship and geological structure correlation edges in the multi-modal edge relationship matrix to obtain local feature patterns and regional feature distributions in the spatial dimension; Based on the data flow association edges and the mining disturbance propagation edges in the multi-modal edge relationship matrix, time sequence feature modeling and evolution analysis are performed on the multi-scale node feature tensor to obtain short-term fluctuation features and long-term trend features in the time dimension; Based on the local feature patterns and regional feature distributions in the spatial dimension, and the short-term fluctuation features and long-term trend features in the time dimension, spatio-temporal feature interaction and multi-scale fusion are performed through a cross-modal attention mechanism to generate a unified multi-level feature representation.

8. The method of claim 1, wherein, Based on the multi-level feature representation, the data quality evaluation matrix, and the real-time mining state of the target mine, a risk identification result of the target mine is determined, including: Risk propagation mechanism analysis and dynamic evolution prediction are performed on the multi-level feature representation and the data quality evaluation matrix to obtain a risk level distribution map and a risk evolution prediction trajectory; Based on the risk level distribution map, the risk evolution prediction trajectory, the multi-level feature representation, and the real-time mining state, risk identification and confidence assessment are performed in combination with the data quality evaluation matrix to obtain a risk identification result, which represents the identification information of four risk types: roof fall risk, water inrush risk, gas outburst risk, and rock burst risk.

9. The method of claim 8, wherein, The risk propagation mechanism analysis and dynamic evolution prediction on the multi-level feature representation and the data quality evaluation matrix to obtain the risk level distribution map and the risk evolution prediction trajectory include: Based on the subgraph-level regional features in the multi-level feature representation, spatial correlation and risk propagation paths between regions are extracted, and the propagation paths are determined by analyzing the spatial continuity and evolution synchronicity of adjacent subgraph features; Based on the graph-level macro features in the multi-level feature representation, the dynamic trend and intensity of global risk propagation are determined; Based on the node-level micro features in the multi-level feature representation, the risk states of each node in the dynamic spatio-temporal graph are initialized, and the credibility of the node states is weighted and corrected using the data quality evaluation matrix; Based on the risk propagation path, the global dynamic trend, and the weighted and corrected node risk state, a risk propagation dynamics equation is constructed, and numerical simulation is performed on the risk propagation dynamics equation to obtain the risk level distribution map and the risk evolution prediction trajectory.

10. A risk identification system based on mine data, characterized by, The system includes: An acquisition module for acquiring multi-source heterogeneous monitoring data of a target mine in response to a mine risk monitoring instruction of the target mine, the multi-source heterogeneous monitoring data including sensor monitoring data, geological structure data, mining process data, and environmental monitoring data; A processing module for processing the multi-source heterogeneous monitoring data to obtain a data quality evaluation matrix of the multi-source heterogeneous monitoring data, standardized data flow, and feature enhancement data, the data quality evaluation matrix including quality evaluation parameters in four dimensions: device operation state, spatio-temporal consistency, data completeness, and abnormal fluctuation, and the feature enhancement data including time domain statistical features, frequency energy features, and spatio-temporal correlation features; The feature representation determination module is configured to determine a multi-level feature representation of a dynamic space-time graph of the target mine based on the data quality evaluation matrix and feature enhancement data, the multi-level feature representation including node-level microscopic features, subgraph-level regional features, and graph-level macroscopic features. The identification result determination module is configured to determine a risk identification result of the target mine based on the multi-level feature representation, the data quality evaluation matrix, and a real-time mining state of the target mine, the real-time mining state including a mining and excavation working face position, a pushing speed, and equipment distribution information.

Images