Mine geological risk early warning method, device and equipment based on multi-source heterogeneous data and storage medium

Abstract

The application discloses a mine geological risk early warning method and device based on multi-source heterogeneous data, equipment and a storage medium, relates to the technical field of intelligent mines, and comprises the following steps: collecting multi-source heterogeneous data of a mine, then performing standardization processing to construct an initial dynamic graph of mine geological entities, updating the dynamic graph by using a space-time graph attention network, predicting a risk grade and a propagation path through a dynamic risk correlation network, finally outputting and generating early warning information in a visual format, and improving the accuracy and early warning timeliness of mine geological risk monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent mine, in particular to a mine geological risk early warning method and device based on multi-source heterogeneous data, equipment and storage medium. BACKGROUND

[0002] At present, the mine geological risk identification technology mainly depends on geological exploration data, sensor network (such as microseismic monitoring, stress meter, osmometer) and three-dimensional laser scanning and other means. These data are processed through statistical methods or shallow machine learning models (such as support vector machine, random forest) for risk early warning. However, these methods usually face the problems of data islandization, lack of spatial correlation and insufficient dynamic adaptability. In addition, although the existing three-dimensional geological modeling technology can statically present the rock layer distribution, it is difficult to dynamically express the geological structure change under mining disturbance, especially in complex structure area. The traditional method relies on empirical formula for modeling the spatial correlation, which cannot adaptively depict the risk propagation path. Secondly, the current mine geological risk identification mainly relies on artificial patrol and limited sensor data. These methods have limitations in real-time, accuracy and comprehensiveness, and are difficult to meet the needs of modern mine safety production. Therefore, there is an urgent need for a mine geological risk early warning method based on multi-source heterogeneous data to improve the evaluation of mine geological risk and timely warning. SUMMARY

[0003] The main purpose of the present application is to provide a mine geological risk early warning method, device, equipment and storage medium based on multi-source heterogeneous data, which aims to solve the technical problem of how to improve the evaluation of mine geological risk and timely warning.

[0004] To achieve the above purpose, the present application provides a mine geological risk early warning method based on multi-source heterogeneous data, comprising:

[0005] Collecting multi-source heterogeneous data of the mine, wherein the multi-source heterogeneous data includes sensor data, geological exploration structure data, mining activity data and unstructured data, and the unstructured data includes three-dimensional laser scanning point cloud, geological profile and drilling video;

[0006] Standardizing the multi-source heterogeneous data to obtain a multi-modal data set;

[0007] Constructing an initial dynamic graph of the geological entity of the mine based on the multi-modal data set, wherein the initial dynamic graph includes node feature vectors and initial adjacency relationships, and the geological entity includes roadway, rock stratum, fault and goaf;

[0008] Updating the initial dynamic graph through a spatio-temporal graph attention network to obtain updated node features and dynamic adjacency weights;

[0009] performing prediction based on the updated node features and dynamic adjacency weights through a dynamic risk association network to obtain a risk prediction result, wherein the dynamic risk association network comprises a collapse risk prediction head, a water inrush risk prediction head and a graph neural network;

[0010] outputting the risk prediction result in a visual format and generating early warning information.

[0011] In an embodiment, the step of standardizing the multi-source heterogeneous data to obtain a multi-modal data set comprises:

[0012] segmenting sensor data in the multi-source heterogeneous data by using a sliding window method and performing Z-Score standardization processing to obtain sensor time series data;

[0013] performing missing value filling and normalization processing on geological exploration structure data in the multi-source heterogeneous data to obtain structured parameters;

[0014] extracting rock boundary features in unstructured data in the multi-source heterogeneous data by using a semantic segmentation network to obtain image features;

[0015] obtaining text features by analyzing the mining activity data through natural language processing technology;

[0016] establishing a spatio-temporal index table based on the association of the sensor time series data, the structured parameters, the image features and the text features to corresponding geological entities;

[0017] generating a multi-modal database according to the spatio-temporal index table.

[0018] In an embodiment, the step of constructing an initial dynamic graph of geological entities of the mine based on the multi-modal data set, wherein the initial dynamic graph comprises node feature vectors and initial adjacency relationships, comprises:

[0019] extracting geological entities from the multi-modal data set to obtain an entity node set, which is specifically represented as ;

[0020] generating a node feature vector for each entity according to the entity set, wherein the node feature vector comprises a spatial coordinate, a lithology parameter vector, sensor data and a node type, and the node type is divided according to the area of the node, wherein a static node is an unexplored area, a dynamic node is a mining work area, and a risk node is a historical disaster area, and is specifically represented as:

[0021]

[0022]

[0023] wherein, is a node feature vector, represents a spatial coordinate, is a node lithology parameter vector, is sensor data, is a node type;

[0024] According to the spatial coordinates in the node feature vector, the Euclidean distance between any two nodes is calculated, and when the Euclidean distance is less than a preset distance threshold, a spatial adjacent edge is established, which is specifically represented as:

[0025]

[0026]

[0027] wherein, is the Euclidean distance between node and node , is a preset distance threshold, is a spatial coordinate of node , is a spatial coordinate of node , represents a spatial adjacent edge;

[0028] According to the lithology parameter vector in the node feature matrix, the cosine similarity between any two nodes is calculated, and when the cosine similarity is greater than a preset similarity threshold, a geological attribute similar edge is established, which is specifically represented as:

[0029]

[0030]

[0031] wherein, represents a cosine similarity, represents a lithology parameter vector of node , represents a lithology parameter vector of node , represents a geological attribute similar edge;

[0032] The spatial adjacent edge and the geological attribute similar edge are merged into a merged edge set to obtain an initial adjacency matrix.

[0033] The entity node set, the node feature vector, and the initial adjacency matrix are combined into an initial dynamic graph.

[0034] In an embodiment, the step of updating the initial dynamic graph by the spatio-temporal graph attention network to obtain updated node features and dynamic adjacency weights comprises:

[0035] The node feature vector and the initial adjacency matrix in the initial dynamic graph are input into a spatial attention module of the spatio-temporal graph attention network to obtain local spatial features of each node by aggregating neighbor node feature vectors, and are specifically represented as:

[0036]

[0037] wherein, is a neighbor set of node , is a node feature vector of a neighbor node , represents a local spatial feature vector of node at time , that is, a representation obtained by aggregating node features of the neighbor set of the node, represents the number of elements of the neighbor set;

[0038] The local spatial features are spliced with the node feature vector at the corresponding time step to obtain a fusion feature vector, and are specifically represented as:

[0039]

[0040] wherein, represents a fusion feature vector, represents an activation function, represents a learnable weight matrix, represents a node feature vector at a corresponding time step;

[0041] The fusion feature vector and the sensor data are input into a gated recurrent unit of the spatio-temporal graph attention network for updating to obtain updated node features and dynamic adjacency weights.

[0042] In an embodiment, the step of inputting the fusion feature vector and the sensor data into the gated recurrent unit of the spatio-temporal graph attention network for updating to obtain updated node features and dynamic adjacency weights comprises:

[0043] According to the fusion feature vector, the gated recurrent unit is input for optimization to obtain an optimized feature vector;

[0044] According to the sensor data, a water pressure difference, a crack permeability, and a stress divergence are obtained;

[0045] According to the water pressure difference and the crack permeability, a weight of a seepage path between nodes is calculated, and is specifically represented as:

[0046]

[0047] wherein, represents the fracture permeability, represents the water cross-sectional area, represents the fluid dynamic viscosity, is a node and a node The water pressure difference between time , is the seepage path weight between node and node at time step , is the Euclidean distance between node and node ;

[0048] The stress redistribution weight is calculated according to the projection of the stress divergence in the direction between nodes, which is specifically represented as:

[0049]

[0050] wherein, is the stress divergence at node , is the unit vector of node pointing to node , is the stress redistribution weight between node and node at time step ;

[0051] The seepage path weight and the stress redistribution weight are fused to obtain a physical constraint weight, which is specifically represented as:

[0052]

[0053] wherein, and are learnable parameters, is the physical constraint weight between node and node at time step ;

[0054] The dynamic adjacency weight is generated based on the physical constraint weight and the optimization feature vector, which is specifically represented as:

[0055]

[0056] wherein, is a nonlinear function, is a learnable weight matrix, is a learnable parameter vector, is a node is an optimized feature vector of the node is a neighbor set of the node is an index of all neighbor nodes of the node is an optimized feature vector of the node at a time step is a dynamic adjacency weight between the node and the node at a time step

[0057] updating the optimized feature vector according to the dynamic adjacency weight by using an attention mechanism to obtain an updated node feature, and a specific formula is as follows:

[0058]

[0059] wherein, is an activation function.

[0060] In an embodiment, the step of predicting by using a dynamic risk association network based on the updated node feature and the dynamic adjacency weight to obtain a risk prediction result, and the step of predicting by using a dynamic risk association network based on the updated node feature and the dynamic adjacency weight to obtain a risk prediction result, includes:

[0061] splicing the updated node feature and sensor data of a corresponding node to obtain a fused physical monitoring feature, and a specific expression is as follows:

[0062]

[0063] wherein, is an updated node feature, indicates sensor data, and the sensor data includes a stress value, a displacement value, and a seepage pressure value, is a learnable weight matrix, is a bias term, is a fused physical monitoring feature of the node at a time step is a stress divergence of the node at a time step

[0064] inputting the fused physical monitoring feature into a landslide risk prediction head to obtain a landslide probability of each node, and a specific expression is as follows:

[0065]

[0066] wherein,​​​​​​ is a collapse prediction weight vector, is a bias term, is a node collapse probability;

[0067] inputting the fused physical monitoring features into a water inrush risk prediction head to obtain a water inrush risk index of each node, and the water inrush risk index is specifically represented as:

[0068]

[0069] wherein, is a water inrush prediction weight vector, is a bias term, is a node water inrush risk probability;

[0070] allocating a risk level to each node according to the collapse probability and the water inrush risk index to obtain a risk level set;

[0071] diffusing and simulating in a graph neural network along a maximum weight direction according to the dynamic adjacency weight to obtain a risk propagation path;

[0072] combining the risk level set and the risk propagation path to obtain a risk prediction result.

[0073] In an embodiment, the step of outputting the risk prediction result in a visual format and generating an early warning information includes:

[0074] parsing the risk prediction result to extract a risk level and a spatiotemporal coordinate of a risk propagation path;

[0075] generating an early warning information based on the risk level;

[0076] rendering a color gradient heat map on a three-dimensional geological model based on the spatiotemporal coordinate to obtain a dynamic risk heat map;

[0077] superimposing the dynamic risk heat map and the early warning information to generate a visual scene;

[0078] sending the visual scene to a monitoring terminal to make the monitoring terminal display an evacuation route and a maintenance suggestion.

[0079] In addition, to achieve the above-mentioned purpose, the present application also provides a mine geological risk early warning device based on multi-source heterogeneous data, which comprises:

[0080] The acquisition module is configured to collect multi-source heterogeneous data of the mine, wherein the multi-source heterogeneous data comprises sensor data, geological exploration structure data, mining activity data and unstructured data, and the unstructured data comprises three-dimensional laser scanning point cloud, geological profile and drilling video.

[0081] The processing module is configured to perform standardization processing on the multi-source heterogeneous data to obtain a multi-modal data set.

[0082] The construction module is configured to construct an initial dynamic graph of a geological entity of the mine based on the multi-modal data set, wherein the initial dynamic graph comprises a node feature vector and an initial adjacency relationship, and the geological entity comprises a roadway, a rock stratum, a fault and a goaf.

[0083] The updating module is configured to update the initial dynamic graph by a spatio-temporal graph attention network to obtain updated node features and dynamic adjacency weights.

[0084] The result module is configured to perform prediction based on the updated node features and dynamic adjacency weights by a dynamic risk correlation network to obtain a risk prediction result, wherein the dynamic risk correlation network comprises a collapse risk prediction head, a water inrush risk prediction head and a graph neural network.

[0085] The execution module is configured to output the risk prediction result in a visual format and generate an early warning information.

[0086] In addition, to achieve the above-mentioned purpose, the present application further provides a storage medium, which is a computer readable storage medium, and a computer program is stored on the storage medium, and the computer program is executed by a processor to implement the steps of the mine geological risk early warning method based on multi-source heterogeneous data as described above.

[0087] In addition, to achieve the above-mentioned purpose, the present application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the mine geological risk early warning method based on multi-source heterogeneous data as described above.

[0088] The present application collects multi-source heterogeneous data of the mine, then performs standardization processing to construct an initial dynamic graph of a mine geological entity, updates the dynamic graph by a spatio-temporal graph attention network, predicts the risk level and propagation path by a dynamic risk correlation network, finally outputs in a visual format and generates an early warning information, which improves the accuracy and early warning timeliness of the mine geological risk monitoring, and provides a strong guarantee for the mine safety production. BRIEF DESCRIPTION OF DRAWINGS

[0089] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, for those of ordinary skill in the art, the other drawings can also be obtained based on these drawings without any creative work.

[0090] Figure 1 The flowchart of the first embodiment of the mine geological risk early warning method based on multi-source heterogeneous data of the present application;

[0091] Figure 2 The flowchart of the second embodiment of the mine geological risk early warning method based on multi-source heterogeneous data of the present application;

[0092] Figure 3 The flowchart of the third embodiment of the mine geological risk early warning method based on multi-source heterogeneous data of the present application;

[0093] Figure 4 The module structure diagram of the mine geological risk early warning device based on multi-source heterogeneous data of the first embodiment of the mine geological risk early warning method based on multi-source heterogeneous data of the present application;

[0094] Figure 5 The device structure diagram of the hardware running environment involved in the mine geological risk early warning method based on multi-source heterogeneous data in the embodiments of the present application.

[0095] The purpose implementation, functional characteristics and advantages of the present application will be further described with reference to the accompanying drawings in combination with the embodiments. DETAILED DESCRIPTION

[0096] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application, and are not used to limit the present application.

[0097] In order to better understand the technical solutions of the present application, the following will be described in detail in combination with the drawings in the specification and specific embodiments.

[0098] The traditional mine safety monitoring method mainly relies on manual patrol and limited sensor data. These methods have limitations in real-time, accuracy and comprehensiveness, and are difficult to meet the needs of modern mine safety production. In addition, the existing monitoring technology has deficiencies in processing multi-source heterogeneous data, dynamically expressing three-dimensional spatial topological relationship and real-time updating risk correlation network, which leads to insufficient accuracy of risk assessment and timeliness of real-time warning.

[0099] Therefore, the present application proposes a mine geological risk early warning method based on multi-source heterogeneous data to solve the above problems. The main solution of the embodiment of the present application is: collecting multi-source heterogeneous data of the mine, wherein the multi-source heterogeneous data includes sensor data, geological exploration structure data, mining activity data and unstructured data, and the unstructured data includes three-dimensional laser scanning point cloud, geological profile and drilling video; standardizing the multi-source heterogeneous data to obtain a multi-modal data set; constructing an initial dynamic graph of a geological entity of the mine based on the multi-modal data set, wherein the initial dynamic graph includes a node feature vector and an initial adjacency relationship, and the geological entity includes a roadway, a rock stratum, a fault and a goaf; updating the initial dynamic graph through a spatio-temporal graph attention network to obtain updated node features and dynamic adjacency weights; predicting based on the updated node features and dynamic adjacency weights through a dynamic risk correlation network to obtain a risk prediction result, wherein the dynamic risk correlation network includes a collapse risk prediction head, a water inrush risk prediction head and a graph neural network; outputting the risk prediction result in a visual format and generating early warning information.

[0100] Based on the above, the embodiment of the present application also provides a mine geological risk early warning method based on multi-source heterogeneous data, which is described with reference to Figure 1 , Figure 1 The flowchart of the first embodiment of the mine geological risk early warning method based on multi-source heterogeneous data of the present application. In this embodiment, the mine geological risk early warning method based on multi-source heterogeneous data includes steps S10-S60:

[0101] Step S10, collecting multi-source heterogeneous data of the mine.

[0102] It should be noted that the above multi-source heterogeneous data includes sensor data, geological exploration structure data, mining activity data and unstructured data. Specifically, the sensor data mainly includes time series data collected by devices such as microseismic monitors, stress meters, osmotic pressure meters and displacement meters deployed in the mine, such as vibration frequency, stress value, seepage pressure and deformation amount. These data can reflect the dynamic changes inside the mine in real time, providing key information for risk early warning. Secondly, the geological exploration structure data covers rock mechanics parameters (such as compressive strength, elastic modulus), fault distribution and hydrogeological map information. This kind of data is crucial for understanding the physical properties of ore bodies and their surrounding environment, and helps to build accurate three-dimensional geological models, thereby laying the foundation for risk assessment. Mining activity data includes working face position, advance speed and blasting records. These data reflect the impact of human activities on the geological conditions of the mine, especially the stress redistribution and potential risk point changes during mining. By analyzing mining activity data, the evolution trend of geological risks caused by mining can be more accurately predicted. Unstructured data further enriches the data dimension of mine geological risk identification, including three-dimensional laser scanning point cloud, geological profile and drilling video. Three-dimensional laser scanning point cloud can provide high-precision surface shape information for monitoring surface subsidence and deformation; geological profile and drilling video provide an intuitive view of underground geological structure, which helps to identify geological anomalies such as fault zones and karst caves.

[0103] Step S20, standardizing the multi-source heterogeneous data to obtain a multi-modal data set.

[0104] It should be noted that due to the diversity of mine multi-source heterogeneous data, including sensor data, geological exploration structure data, mining activity data and unstructured data, these data have different dimensions and formats, so they need to be standardized to ensure that they can be effectively integrated and used for subsequent analysis.

[0105] Further, step S20 further comprises: firstly, segmenting the sensor data in the multi-source heterogeneous data using a sliding window method, and obtaining sensor time series data through Z-Score standardization processing. Specifically, for sensor data, the collected vibration frequency, stress value, displacement, and other time series data are segmented and processed using a sliding window method to extract local time characteristics, and then the Z-Score standardization method is used to eliminate the dimensional differences between different sensors, thereby obtaining sensor time series data in a unified scale. This data can reflect the dynamic processes of internal stress changes, microseismic activities, etc. in the mine. Then, the geological exploration structure data in the multi-source heterogeneous data is subjected to missing value filling and normalization processing to obtain structured parameters. Specifically, for geological exploration structure data such as rock thickness, compressive strength, elastic modulus, etc., there are usually missing values or measurement errors, so interpolation methods such as KNN interpolation or linear interpolation are used for missing value filling, and the Min-Max normalization method is used to scale the data to form structured parameters, which facilitates subsequent graph model construction and feature fusion. Then, the semantic segmentation network is used to extract the rock boundary features in the unstructured data in the multi-source heterogeneous data to obtain image features. Specifically, the semantic segmentation network (such as U-Net, DeepLab, etc.) is used to perform pixel-level classification on the image to extract rock boundary features such as rock boundary, fault strike, and fracture distribution, forming image features to represent the spatial morphology of geological entities. Subsequently, the mining activity data is analyzed by natural language processing technology to obtain text features. Specifically, the text logs in the mining activity data record key events such as working face advancement, blasting operation, and support construction. The text is analyzed by natural language processing technology (such as BERT, LSTM, etc.) to extract structured semantic features such as event type, timestamp, and location information, forming text features to provide behavioral context for risk prediction. Finally, based on the association of sensor time series data, structured parameters, image features, and text features to corresponding geological entities, a spatio-temporal index table is established, and then a multi-modal database is generated according to the spatio-temporal index table. Specifically, the above-mentioned sensor time series data, structured parameters, image features, and text features are associated to corresponding geological entities (such as roadway, rock, fault, and goaf) through a spatio-temporal indexing mechanism to construct a spatio-temporal index table containing timestamps, spatial coordinates, and attribute features. This index table realizes the unified mapping of multi-source heterogeneous data in the spatio-temporal dimension. Finally, based on the spatio-temporal index table, a multi-modal database is constructed in combination with a graph database or a relational database system. This database not only supports efficient storage and query of structured and unstructured data, but also provides a unified data interface for subsequent dynamic risk modeling based on graph neural networks.

[0106] Step S30, constructing an initial dynamic graph of the geological entities of the mine based on the multi-modal data set.

[0107] It's important to note that this process involves abstracting key geological entities in the mine, such as tunnels, strata, faults, and goafs, into nodes in a graph neural network. Each node has a defined feature vector, which integrates multi-source information from sensors, geological exploration, image analysis, and text parsing. The node feature vector includes not only the entity's physical and chemical properties but also its behavioral characteristics during mining operations, such as stress changes and displacement rates. The construction of the initial dynamic graph also includes defining adjacency relationships between nodes, based on actual spatial distance, geological similarity, or other correlation indicators. For example, adjacent tunnels and strata may be connected by spatial proximity edges, while nodes with similar lithological parameters are connected by geological property similarity edges. These adjacency relationships form the skeleton of the graph neural network, providing the foundation for subsequent dynamic updates and risk assessment.

[0108] Furthermore, step S30 also includes: extracting geological entities from the multimodal dataset to obtain a set of entity nodes, specifically represented as follows: .

[0109] Then, a node feature vector is generated for each entity based on the entity set. This node feature vector includes spatial coordinates, lithological parameter vectors, sensor data, and node type. The node type is categorized according to its region: static nodes represent unmined areas, dynamic nodes represent mining areas, and risk nodes represent areas with historical disasters. Specifically:

[0110]

[0111]

[0112] in, For node feature vectors, Represents spatial coordinates, For nodes lithological parameter vector, For sensor data, For node types, spatial coordinates provide the location information of the entity, lithological parameter vectors reflect its physical properties, sensor data captures dynamic changes, and node types specify the specific area where the entity is located.

[0113] The Euclidean distance between any two nodes is calculated based on the spatial coordinates in the node feature vector. When the Euclidean distance is less than a preset distance threshold, the two nodes are considered to have a spatial proximity relationship, and a spatial proximity edge can be established. This relationship helps to capture the direct physical interactions between geological bodies, specifically expressed as follows:

[0114]

[0115]

[0116] in, For nodes and nodes The Euclidean distance between them For the preset distance threshold, For nodes spatial coordinates, For nodes spatial coordinates, Indicates spatially adjacent edges;

[0117] The cosine similarity between any two nodes is calculated based on the lithology parameter vector in the node feature matrix. When the cosine similarity is greater than a preset similarity threshold, it indicates that the two nodes have high similarity in geological attributes, and a geological attribute similarity edge can be established, specifically represented as follows:

[0118]

[0119]

[0120] in, Represents cosine similarity. Represents a node lithological parameter vector, Represents a node lithological parameter vector, This indicates edges with similar geological properties.

[0121] The initial adjacency matrix is ​​obtained by merging spatially adjacent edges and edges with similar geological attributes. Specifically, this process involves traversing all possible node pairs and determining whether a connection is established between them according to preset rules, and then merging them. After traversing all nodes, the initial adjacency matrix is ​​obtained. This adjacency matrix reflects not only the spatial distance-based interactions between nodes but also the association information based on geological attributes. Finally, the set of entity nodes, node feature vectors, and the initial adjacency matrix are combined into an initial dynamic graph, specifically represented as follows:

[0122]

[0123] in, For the set of entity nodes ( (Geological entities) For time The set of edges, The node feature matrix is ​​derived from the node eigenvectors and the initial adjacency matrix. This initial dynamic graph effectively captures the complexity and dynamism of the geological structure within the mine, providing strong support for in-depth research on mine stability and optimization of mining design. Furthermore, as new data is continuously added, this dynamic graph can be continuously updated and improved to adapt to various changes during the mine development process.

[0124] Step S40: Update the initial dynamic graph using a spatiotemporal graph attention network to obtain updated node features and dynamic adjacency weights.

[0125] It should be noted that updating the initial dynamic graph using the Spatiotemporal Graph Attention Network (ST-GAT) can more accurately reflect the dynamic changes in geological conditions and the real-time characteristics of risk evolution. Specifically, the node feature vectors and the initial adjacency matrix in the initial dynamic graph are input into the spatial attention module of the ST-GAT network. By aggregating the feature vectors of neighboring nodes and using an average aggregation method, the feature vectors of all neighboring nodes are weighted and summed to obtain the local spatial features of each node, specifically represented as follows:

[0126]

[0127] in, It is a node The neighborhood group, For neighboring nodes The node feature vectors, Represents a node At any moment The local spatial feature vector is a representation obtained by aggregating the node features from the set of the node's neighbors. Indicates the number of elements in the neighbor set;

[0128] To further integrate information from both temporal and spatial dimensions, local spatial features are concatenated with the node feature vectors at the corresponding time steps to obtain a fused feature vector, specifically represented as follows:

[0129]

[0130] in, Represents the fused feature vector. This represents the activation function. Represents the learnable weight matrix. This represents the node feature vector at the corresponding time step;

[0131] The fused feature vector and sensor data are input into the gated recurrent unit (GRU) of the spatiotemporal graph attention network for updating, resulting in updated node features and dynamic adjacency weights. The GRU is a neural network structure specifically designed for processing time-series data; it effectively captures long-term dependencies in time series and dynamically adjusts node features and adjacency weights based on new observation data. Specifically, the fused feature vector is input into the gated recurrent unit for optimization, resulting in an optimized feature vector. Then, water pressure difference, fracture permeability, and stress divergence are obtained from sensor data. Fracture permeability is obtained from geological exploration data, while water pressure difference is measured by a piezometer installed in the mine. Stress divergence is determined by analyzing stress changes at different locations within the mine, which helps to understand the stress redistribution caused by mining activities.

[0132] Next, the seepage path weights between nodes are calculated based on the water pressure difference and crack permeability, specifically as follows:

[0133]

[0134] in, Indicates crack permeability. Indicates the cross-sectional area of ​​the water passage. Indicates fluid dynamic viscosity, For nodes and nodes In time Water pressure difference, For time step Time node and nodes The seepage path weights between them For nodes and nodes The Euclidean distance between them;

[0135] Next, the stress redistribution weights are calculated based on the projection of the stress divergence along the inter-node direction, specifically expressed as follows:

[0136]

[0137] in, For nodes Stress divergence at the point, For nodes Pointing to node unit vector, For time step Time node and nodes Stress redistribution weights between;

[0138] Then, the seepage path weight and stress redistribution weight are fused to obtain the physical constraint weight, which comprehensively considers the physical interactions between geological entities, and is specifically expressed as follows:

[0139]

[0140] in, and For learnable parameters, For time step Time node and nodes Physical constraint weights between them;

[0141] Subsequently, a dynamic adjacency weight is generated based on the physical constraint weights and the optimized feature vectors. This weight not only considers the similarity of geological attributes but also incorporates the physical constraints provided by real-time monitoring data, specifically expressed as follows:

[0142]

[0143] in, It is a nonlinear function. For learnable weight matrix, For learnable parameter vectors, For nodes Optimized feature vectors, It is a node The neighborhood group, For traversing nodes The index of all neighboring nodes, For nodes At time step Optimized feature vectors For time step Time node and nodes Dynamic adjacency weights between them;

[0144] Finally, the attention mechanism is used to update and optimize the feature vector based on the dynamic adjacency weights, resulting in the updated node features. The specific formula is as follows:

[0145]

[0146] in, This is the activation function. This process allows the model to adaptively adjust the connection strength between each node and its neighbors, more accurately simulating risk propagation paths under actual geological conditions. Through such an iterative update mechanism, the risk evolution patterns of geological hazards can be captured more accurately, providing strong support for mine safety.

[0147] Step S50, based on the updated node features and dynamic adjacency weights, the prediction is made through the dynamic risk association network to obtain the risk prediction result.

[0148] It should be noted that the dynamic risk association network is a deep learning model specially designed for mine geological risk assessment. It integrates the updated node features and dynamic adjacency weights to make accurate risk prediction. The dynamic risk association network includes a landslide risk prediction head, a water inrush risk prediction head, and a graph neural network. The landslide risk prediction head and the water inrush risk prediction head are two independent branches, which focus on assessing the risk of two specific types of geological disasters in mines, namely landslides and water inrush. Each prediction head consists of a fully connected layer and an activation function, which can calculate the risk probability of the corresponding disaster based on the input node features and adjacency weights. The graph neural network (GNN) is the core part of the dynamic risk association network, which can process graph structure data and learn the complex relationships between nodes. GNN uses a message passing mechanism to aggregate the information of neighboring nodes for each node's feature vector, capturing the topological features of the entire graph structure. This mechanism enables GNN to dynamically update node features and adjacency weights to reflect real-time changes in the mine geological environment.

[0149] Through the coordinated work of these three parts, the dynamic risk association network can comprehensively assess various geological risks in mines and output prediction results such as risk levels and propagation paths.

[0150] Step S60, the risk prediction result is output in a visual format and generates an early warning information.

[0151] It should be noted that the risk prediction result is converted into a visual heat map or three-dimensional model display. For example, color gradients are used to represent the risk levels in different regions: green represents low-risk areas, and yellow to red corresponds to medium to high-risk areas. This visualization method allows management personnel to quickly identify potential risk points and their impact range. In addition, it can also generate a warning animation to show the spatio-temporal evolution process of the risk propagation path. For example, how does the influence radius of rock burst expand over time, or how does groundwater seepage spread along the fracture network. These dynamic demonstrations greatly enhance the understanding of disaster development trends and help develop more accurate emergency response strategies. Finally, based on the risk prediction result, an early warning information is automatically generated and updated in real time on the visualization platform. When the risk exceeds the preset threshold, the early warning mechanism is automatically triggered, and relevant personnel are notified to guide them to take necessary safety measures such as evacuating personnel, strengthening support, etc. This ensures that in emergency situations, a quick response can be made to maximize the safety and efficiency of mine operations.

[0152] The embodiment collects multi-source heterogeneous data of a mine, then performs standardization processing to construct an initial dynamic graph of mine geological entities, updates the graph structure by using a spatio-temporal graph attention network, predicts a risk level and a propagation path through a dynamic risk correlation network, finally outputs and generates early warning information in a visual format, improves the accuracy and early warning timeliness of mine geological risk monitoring, and provides a strong guarantee for mine safety production.

[0153] Based on the first embodiment of the present application, the same or similar contents as the above embodiment one can be referred to the above introduction, and will not be described in detail. On this basis, please refer to Figure 2 , the mine geological risk early warning method based on multi-source heterogeneous data step S50 further includes steps S201-S206:

[0154] Step S201, the updated node feature is spliced with the sensor data of the corresponding node to obtain a fused physical monitoring feature.

[0155] It should be noted that the updated node feature contains a vector reflecting the spatial correlation and time evolution between geological entities obtained by processing the spatio-temporal graph attention network. These feature vectors not only contain the inherent properties of geological entities such as lithology, structure, etc., but also incorporate topological relationship information learned through graph neural networks. The sensor data of the corresponding node provides real-time monitoring of physical quantities such as stress, displacement, and osmotic pressure, which directly reflect the current physical state and change trend of the geological entity. Splicing these sensor data with node feature vectors means combining real-time dynamic information with static geological information of geological entities to form a rich and real-time fused physical monitoring feature, which is specifically represented as:

[0156]

[0157] wherein, is the updated node feature, represents a real-time monitoring value, which includes a stress value, a displacement value, and an osmotic pressure value, is a learnable weight matrix, is a bias term, is the fused physical monitoring feature of node at time step , and is the stress divergence of node at time step .

[0158] Step S202, input the fused physical monitoring feature into the landslide risk prediction head to obtain the landslide probability of each node.

[0159] It is worth noting that the collapse risk prediction head is usually composed of a set of fully connected layers that can learn the complex relationship between the fused features and the collapse risk. Through training, this prediction head can identify key factors that affect collapse, such as stress concentration, rock structure weaknesses, etc.

[0160] Specifically, after inputting the fused physical monitoring features, the collapse risk prediction head calculates the collapse probability of each node through forward propagation. This process involves adjusting weights and applying activation functions to ensure that the output probability values are both statistically sound and reflect actual conditions. The resulting probability values represent the likelihood of each node collapsing within a given monitoring condition within a certain period of time, specifically represented as:

[0161] Specifically represented as:

[0162]

[0163] where, is the collapse prediction weight vector, is the bias term, is the collapse probability of node .

[0164] Step S203, input the fused physical monitoring features into the water inrush risk prediction head to obtain the water inrush risk index of each node.

[0165] It is worth noting that the collapse risk prediction head is usually composed of a set of fully connected layers that can analyze inputs that combine real-time sensor data and geological features to predict the likelihood of water inrush at each node. These fused features include key geological information and dynamic monitoring data, such as groundwater level, water pressure, rock permeability, and historical water inrush events, etc.

[0166] Specifically, the water inrush risk prediction head can identify complex patterns and correlations related to water inrush events, thereby calculating a risk index for each node. This index reflects the risk level of water inrush at the node under current monitoring conditions. The higher the risk index, the greater the likelihood of water inrush at the node, specifically represented as:

[0167]

[0168] where, is the water inrush prediction weight vector, is the bias term, is the water inrush risk probability of node . After obtaining the water inrush risk index of each node, mine management personnel can more accurately identify potential high-risk areas and take appropriate preventive measures, such as strengthening the drainage system, implementing underground water monitoring and early warning, adjusting mining strategies, etc.

[0169] Step S204, assign risk levels to each node according to the collapse probability and the water inrush risk index, and obtain a risk level set.

[0170] It should be noted that a process usually involves setting a series of predefined risk thresholds, according to which the probability and index are mapped to different risk levels, such as low risk, medium risk and high risk.

[0171] The assignment of risk levels enables mine managers to intuitively identify and prioritize areas where disasters are most likely to occur. For example, nodes with high risk levels may require immediate emergency measures such as evacuating personnel, reinforcing support or suspending mining activities. Medium-risk areas may require enhanced monitoring and regular assessment, while low-risk areas can continue regular operations but still require basic monitoring and preventive measures. In addition, the risk level set can also be used for long-term planning and management of the mine, helping decision-makers optimize resource allocation and improve the safety and efficiency of the mine.

[0172] Step S205, according to the dynamic adjacency weight, simulate the diffusion of risk propagation path in the graph neural network along the direction of maximum weight, and obtain the risk propagation path.

[0173] It should be noted that simulating the risk propagation path in the graph neural network according to the dynamic adjacency weight is an effective method to predict and visualize how the risk spreads in the mine geological structure. This process involves using the connection strength between nodes in the graph network, i.e. adjacency weight, to simulate the possibility of risk propagation from one node to another. The greater the weight, the stronger the connection between two nodes, and the higher the possibility of risk propagation from one node to another.

[0174] In this way, the key path of risk propagation can be identified, i.e. the nodes and connections that are most likely to cause risk spread. This is crucial for mine safety management, as it can help managers identify areas that need to be closely monitored and prevented.

[0175] Step S206, combine the risk level set with the risk propagation path to obtain the risk prediction result.

[0176] It is worth noting that the risk level set provides the risk assessment of each node, while the risk propagation path demonstrates how the risk spreads in the mine geological structure. Combining these two pieces of information, a comprehensive risk prediction result can be generated, which not only identifies high-risk areas, but also predicts the possible propagation direction and impact range of these risks. This combination allows mine managers to have a clearer understanding of the spatio-temporal distribution of risks, so as to make more accurate and targeted risk response strategies. For example, if a certain area is predicted to be high-risk, and the risk propagation path shows that the risk in this area has the potential to spread to adjacent areas, the manager can take preventive measures in these potentially affected areas in advance, such as strengthening monitoring, reinforcing structures, or adjusting operation plans. In addition, this comprehensive risk prediction result can also be used for emergency response planning in mines, ensuring that actions can be taken quickly and effectively when risks occur, reducing possible losses.

[0177] The embodiment forms a fusion of physical monitoring features by splicing the updated node features with real-time monitoring values, and then inputs them into the collapse and water inrush risk prediction head to calculate the collapse probability and water inrush risk index of each node respectively. According to this, the risk level is assigned, and the risk propagation path is simulated, and finally the comprehensive risk prediction result is obtained by combining the risk level and the propagation path, realizing the accurate assessment and early warning of the mine geological risk, improving the intelligent level of mine safety management, and effectively preventing geological disasters.

[0178] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar contents as the above embodiment one can refer to the above introduction, and the subsequent will not be repeated. On this basis, please refer to Figure 3 , the mine geological risk early warning step S60 based on multi-source heterogeneous data further includes steps S301-S305:

[0179] Step S301, the risk prediction result is analyzed, and the spatio-temporal coordinates of the risk level and the risk propagation path are extracted.

[0180] It is worth noting that through the multi-task prediction head from the ST-GAT, each node will have a corresponding risk score or probability value. According to the preset risk threshold (for example, three levels of low, medium and high), these values can be converted into intuitive risk levels. Then, for the analysis of the risk propagation path, it involves extracting the spatio-temporal coordinates of the risk propagation path, which refers to the process of how the disaster spreads from a starting point to other areas within a certain time period.

[0181] Step S302, generating early warning information based on the risk level.

[0182] It should be noted that when the risk level of a certain area reaches the preset alert line, the system automatically generates a corresponding early warning signal. These early warning signals are usually divided into three levels: yellow (prompt attention), orange (warning) and red (emergency action). The above yellow corresponds to low risk level, orange corresponds to medium risk level, and red corresponds to high risk level, so as to quickly identify the severity of the risk. Then, the system will automatically generate detailed early warning information for different levels of risk. For yellow warning, it may only need to increase the monitoring frequency and remind the relevant personnel; orange warning needs to develop more specific response strategies, such as increasing support strength or limiting activities in some areas; and red warning means that emergency measures must be taken immediately, such as evacuating personnel, suspending work, etc., to prevent disasters.

[0183] Step S303, rendering a color gradient heat map on the three-dimensional geological model based on the space-time coordinates to obtain a dynamic risk heat map.

[0184] It should be noted that different colors in the heat map represent different risk levels. In this embodiment, yellow is usually used to represent low-risk areas, orange to represent medium-risk areas, and red to identify high-risk areas. This color gradient design allows managers to quickly identify which areas need special attention. At the same time, in order to reflect the time dimension change of risk, the system will dynamically update the heat map according to the latest real-time monitoring data. This means that as new data continues to flow in, the color distribution on the heat map will also be adjusted accordingly, accurately reflecting the changing trend of risk. For example, if the risk of a certain area increases due to increased groundwater pressure or stress redistribution within a certain time period, the color of that area will gradually change from a cooler tone (such as yellow) to a warmer tone (such as red).

[0185] Step S304, superimposing the dynamic risk heat map with early warning information to generate a visual scene.

[0186] It should be noted that the dynamic risk heat map serves as the base layer, presenting the risk levels of different areas in the form of color gradient. At the same time, early warning information is displayed through specific identifiers (such as flashing icons, text prompts, etc.) in the corresponding position, so as to quickly attract the attention of managers. For example, when the risk level of a certain area reaches the preset threshold, not only will the position display a prominent red color on the heat map, but also a warning sign with the words "high risk" will appear, and the corresponding early warning level (such as yellow / orange / red alarm) will be triggered. In addition, emergency response recommendations can also be automatically generated, such as recommending personnel evacuation routes or identifying key areas that need to strengthen support.

[0187] To further enhance the user experience, the visualization platform also provides multi-perspective zooming, historical data backtracking, and other functions. Management personnel can freely adjust the view to examine the risk situation of the entire mine area from a macro perspective, or focus on specific locations to analyze the historical record of risk changes in detail.

[0188] In step S305, the visualization scene is sent to the monitoring terminal to make the monitoring terminal display the evacuation route and maintenance recommendations.

[0189] It should be noted that once a geological disaster is predicted to occur, the generated visualization scene will be immediately sent to various monitoring terminals. These terminals can include large-screen displays in the control room, mobile devices such as tablets and smartphones, etc., ensuring that information can be conveyed to all relevant personnel in the first instance. On the monitoring terminal, in addition to displaying the color gradient risk heat map, the recommended evacuation route is highlighted, usually marked in bright colors (such as blue arrows) for easy identification. For different levels of risk areas, the system automatically generates and displays corresponding maintenance recommendations, such as specific locations and methods for strengthening support, monitoring points that require special attention, etc. These recommendations are generated based on real-time data analysis and are highly targeted, helping to quickly take effective protective measures.

[0190] In addition, the monitoring terminal can also provide interactive functions, allowing management personnel to adjust the evacuation route or maintenance strategy according to actual conditions and update the display content in real time. This flexibility ensures that optimal decisions can be made even in emergency situations. In this way, not only can the ability to respond to sudden geological disasters be significantly improved, but the safety of mine workers can also be maximized, reducing property losses.

[0191] The embodiment extracts the risk level and the spatiotemporal coordinates of the risk propagation path by analyzing the risk prediction results, generates early warning information based on this, and renders a dynamic risk heat map on the three-dimensional geological model. The dynamic risk heat map and the early warning information are superimposed to generate a visualization scene, which is finally sent to the monitoring terminal to display the evacuation route and maintenance recommendations, providing intuitive risk visualization and immediate early warning, enhancing emergency response efficiency and accuracy, effectively ensuring mine operation safety, and reducing disaster risk and possible losses.

[0192] Based on the first embodiment of the present application, the present application also provides a mine geological risk early warning device based on multi-source heterogeneous data, please refer to Figure 4 , the device comprises:

[0193] The acquisition module 10 is used to collect multi-source heterogeneous data of the mine, wherein the multi-source heterogeneous data includes sensor data, geological exploration structure data, mining activity data, and unstructured data, and the unstructured data further includes three-dimensional laser scanning point cloud, geological profile, and drilling video.

[0194] The processing module 20 is configured to perform standardization processing on the multi-source heterogeneous data to obtain a multi-modal data set.

[0195] The construction module 30 is configured to construct an initial dynamic graph of a geological entity of the mine based on the multi-modal data set, wherein the initial dynamic graph comprises a node feature vector and an initial adjacency relationship, and the geological entity comprises a roadway, a rock stratum, a fault and a goaf.

[0196] The updating module 40 is configured to update the initial dynamic graph by a spatio-temporal graph attention network to obtain updated node features and dynamic adjacency weights.

[0197] The result module 50 is configured to perform prediction based on the updated node features and the dynamic adjacency weights by a dynamic risk association network to obtain a risk prediction result, wherein the dynamic risk association network comprises a collapse risk prediction head, a water inrush risk prediction head and a graph neural network.

[0198] The execution module 60 is configured to output the risk prediction result in a visual format and generate an early warning information.

[0199] The mine geological risk early warning device based on multi-source heterogeneous data provided in the present application adopts the mine geological risk early warning method based on multi-source heterogeneous data in the above embodiment, and can solve the technical problem of how to improve the mine geological risk assessment and thus perform timely early warning. Compared with the prior art, the mine geological risk early warning device based on multi-source heterogeneous data provided in the present application has the same beneficial effects as the mine geological risk early warning method based on multi-source heterogeneous data provided in the above embodiment, and other technical features in the mine geological risk early warning device based on multi-source heterogeneous data are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0200] In an embodiment, the processing module 20 is further configured to segment sensor data in the multi-source heterogeneous data by using a sliding window method, and perform Z-Score standardization processing to obtain sensor time series data; perform missing value filling and normalization processing on geological exploration structure data in the multi-source heterogeneous data to obtain structured parameters; extract rock stratum boundary features in unstructured data in the multi-source heterogeneous data by using a semantic segmentation network to obtain image features; analyze mining activity data by using a natural language processing technology to obtain text features; establish a spatio-temporal index table based on the association of the sensor time series data, the structured parameters, the image features and the text features to corresponding geological entities; and generate a multi-modal database according to the spatio-temporal index table.

[0201] In an embodiment, the construction module 30 is further configured to extract geological entities from the multi-modal data set to obtain an entity node set; generate a node feature vector for each entity according to the entity set, wherein the node feature vector comprises a spatial coordinate, a lithology parameter vector, sensor data, and a node type, the node type is divided according to a region of the node, a static node is an unexplored region, a dynamic node is a mining working region, and a risk node is a historical disaster region; calculate a Euclidean distance between any two nodes according to the spatial coordinate in the node feature vector, and establish a spatial proximity edge when the Euclidean distance is less than a preset distance threshold; calculate a cosine similarity between any two nodes according to the lithology parameter vector in the node feature matrix, and establish a geological attribute similarity edge when the cosine similarity is greater than a preset similarity threshold; merge the spatial proximity edge and the geological attribute similarity edge to obtain an initial adjacency matrix; and combine the entity node set, the node feature vector, and the initial adjacency matrix to obtain an initial dynamic graph.

[0202] In an embodiment, the update module 40 is further configured to input the node feature vector and the initial adjacency matrix in the initial dynamic graph into a spatial attention module of a spatio-temporal graph attention network to obtain a local spatial feature of each node by aggregating neighbor node feature vectors; splice the local spatial feature with a node feature vector at a corresponding time step to obtain a fusion feature vector; and input the fusion feature vector and the sensor data into a gated recurrent unit of the spatio-temporal graph attention network for updating to obtain an updated node feature and a dynamic adjacency weight.

[0203] In an embodiment, the update module 40 is further configured to input the fusion feature vector into the gated recurrent unit for optimization to obtain an optimized feature vector; obtain a water pressure difference, a crack permeability, and a stress divergence from the sensor data; calculate a seepage path weight between nodes according to the water pressure difference and the crack permeability; calculate a stress redistribution weight according to a projection of the stress divergence in a direction between the nodes; fuse the seepage path weight and the stress redistribution weight to obtain a physical constraint weight; generate the dynamic adjacency weight based on the physical constraint weight and the optimized feature vector; and update the optimized feature vector according to the dynamic adjacency weight by using an attention mechanism to obtain an updated node feature.

[0204] In an embodiment, the result module 50 is further configured to splice the updated node feature with the sensor data of a corresponding node to obtain a fusion physical monitoring feature; input the fusion physical monitoring feature into a collapse risk prediction head to obtain a collapse probability of each node; input the fusion physical monitoring feature into a water inrush risk prediction head to obtain a water inrush risk index of each node; assign a risk level to each node according to the collapse probability and the water inrush risk index to obtain a risk level set; simulate diffusion according to the dynamic adjacency weight in the graph neural network in a direction of a maximum weight to obtain a risk propagation path; and combine the risk level set and the risk propagation path to obtain a risk prediction result.

[0205] In an embodiment, the execution module 60 is further configured to parse the risk prediction result, extract a risk level and a time-space coordinate of a risk propagation path; generate early warning information based on the risk level; render a color gradient heat map on a three-dimensional geological model based on the time-space coordinate, to obtain a dynamic risk heat map; superimpose the dynamic risk heat map and the early warning information to generate a visual scene; and send the visual scene to a monitoring terminal, so that the monitoring terminal displays an evacuation route and maintenance suggestions.

[0206] The present application provides a mine geological risk early warning device based on multi-source heterogeneous data. The mine geological risk early warning device based on multi-source heterogeneous data comprises at least one processor and a memory in communication connection with the at least one processor. The memory stores instructions executable by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the mine geological risk early warning method based on multi-source heterogeneous data in the above embodiment one.

[0207] Reference will be made to the following description Figure 5 which shows a structural schematic diagram of the mine geological risk early warning device based on multi-source heterogeneous data suitable for use to implement the embodiments of the present application. The mine geological risk early warning device based on multi-source heterogeneous data in the embodiments of the present application can include, but is not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal Digital Assistant), PADs (Portable Application Description), PMPs (Portable Media Player), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), and the like, and fixed terminals such as digital TVs, desktop computers, and the like. Figure 5 The mine geological risk early warning device based on multi-source heterogeneous data shown is merely an example and should not bring any limitation to the functions and use range of the embodiments of the present application.

[0208] As Figure 5As shown, the mine geological risk early warning device based on multi-source heterogeneous data can include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 1002 or loaded from a storage device 1003 into a random access memory (RAM) 1004. In the RAM 1004, various programs and data required for operation of the mine geological risk early warning device based on multi-source heterogeneous data are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; the storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the mine geological risk early warning device based on multi-source heterogeneous data to communicate wirelessly or wired with other devices to exchange data. Although various mine geological risk early warning devices based on multi-source heterogeneous data are shown in the figure, it should be understood that all of the shown are not required to be implemented or possessed. More or less can be alternatively implemented or possessed.

[0209] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program carrying on a computer readable storage medium, the computer program containing program codes for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network through a communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the methods of the embodiments of the present disclosure are performed.

[0210] The mine geological risk early warning device based on multi-source heterogeneous data provided in the application adopts the mine geological risk early warning method based on multi-source heterogeneous data in the above embodiment, and can solve the technical problem of how to improve the mine geological risk assessment and thus perform timely early warning. Compared with the prior art, the beneficial effects of the mine geological risk early warning device based on multi-source heterogeneous data provided in the application are the same as those of the mine geological risk early warning method based on multi-source heterogeneous data provided in the above embodiment, and other technical features of the mine geological risk early warning device based on multi-source heterogeneous data are the same as those disclosed in the above embodiment method, and thus are not described herein.

[0211] It should be understood that parts of the present application can be realized by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0212] The above is merely specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0213] The present application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e. computer programs) for executing the mine geological risk early warning method based on multi-source heterogeneous data in the above embodiment.

[0214] The computer readable storage medium provided in the present application may, for example, be a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor, or device, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electric connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read only memory (ROM), an erasable programmable read only memory (EPROM or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium can be any tangible storage medium containing or storing a program that can be executed by or used in conjunction with an instruction execution device. The program code contained on the computer readable storage medium can be transmitted in any suitable storage medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency), and the like, or any suitable combination of the above.

[0215] The above computer readable storage medium can be contained in the mine geological risk early warning device based on multi-source heterogeneous data; or can exist independently without being assembled into the mine geological risk early warning device based on multi-source heterogeneous data.

[0216] The above computer readable storage medium carries one or more programs, which, when executed by the mine geological risk early warning device based on multi-source heterogeneous data, enable the mine geological risk early warning device based on multi-source heterogeneous data to write computer program codes in one or more programming languages or combinations thereof for executing the operations of the present application. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" language or similar programming languages. The program code can be executed entirely on a user computer, partially on a user computer, as a separate software package, partially on a user computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, through the Internet using an Internet service provider).

[0217] The flowcharts and block diagrams in the drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a segment, or a portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks noted in succession can in fact be executed substantially concurrently or can sometimes be executed in reverse order, depending on the functionality involved. It will also be noted that each block of the block diagrams and / or flowcharts, and combinations thereof, can be implemented by dedicated hardware-based systems which perform the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0218] The modules involved in the embodiments of the present application can be implemented in the form of software or in the form of hardware. In some cases, the name of the module does not constitute a limitation on the unit itself.

[0219] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e. computer programs) for executing the above-mentioned mine geological risk early warning method based on multi-source heterogeneous data, and can solve the technical problem of how to improve the evaluation of mine geological risk and thus perform timely early warning. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the mine geological risk early warning method based on multi-source heterogeneous data provided by the above-mentioned embodiments, and will not be described here.

[0220] The present application also provides a computer program product comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned mine geological risk early warning method based on multi-source heterogeneous data.

[0221] The computer program product provided by the present application can solve the technical problem of how to improve the evaluation of mine geological risk and thus perform timely early warning. Compared with the prior art, the computer program product provided by the present application has the same beneficial effects as the mine geological risk early warning method based on multi-source heterogeneous data provided by the above-mentioned embodiments, and will not be described here.

[0222] The above only describes some embodiments of the present application, and does not limit the patent scope of the present application, and any equivalent structural transformation made by using the content of the present application specification and drawings, or direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. A method for early warning of geological risks in mines based on multi-source heterogeneous data, characterized in that, include: Collect multi-source heterogeneous data from the mine, including sensor data, geological exploration structure data, mining activity data, and unstructured data. The unstructured data includes three-dimensional laser scanning point clouds, geological profile maps, and borehole videos. The multi-source heterogeneous data is standardized to obtain a multimodal dataset; An initial dynamic graph of the geological entities of the mine is constructed based on the multimodal dataset, wherein the initial dynamic graph includes node feature vectors and initial adjacency relationships, and the geological entities include roadways, rock strata, faults and goafs; The initial dynamic graph is updated using a spatiotemporal graph attention network to obtain updated node features and dynamic adjacency weights. Based on the updated node features and dynamic adjacency weights, a risk prediction result is obtained by using a dynamic risk association network, wherein the dynamic risk association network includes a landslide risk prediction head, a water inrush risk prediction head, and a graph neural network. The risk prediction results are output in a visual format and early warning information is generated. The step of predicting risk based on the updated node features and dynamic adjacency weights through a dynamic risk association network to obtain risk prediction results includes: The updated node features are concatenated with the sensor data of the corresponding nodes to obtain fused physical monitoring features, specifically represented as follows: in, For the updated node features, This represents sensor data, including stress values, displacement values, and seepage pressure values. For learnable weight matrix, For bias terms, For nodes At time step The integration of physical monitoring characteristics For nodes At time step Stress divergence; The fused physical monitoring features are input into the landslide risk prediction head to obtain the landslide probability of each node, specifically expressed as follows: in, For the collapse prediction weight vector, For bias terms, For nodes The probability of a landslide; The fused physical monitoring features are input into the water inrush risk prediction head to obtain the water inrush risk index for each node, specifically expressed as: in, This is the weight vector for predicting sudden water inrush. For bias terms, For nodes The probability of sudden water inrush risk; Each node is assigned a risk level based on the collapse probability and the water inrush risk index, thus obtaining a risk level set. Based on the dynamic adjacency weights, a diffusion simulation is performed in the graph neural network along the direction of maximum weight to obtain the risk propagation path; By combining the set of risk levels with the risk propagation path, a risk prediction result is obtained.

2. The method as described in claim 1, characterized in that, The step of standardizing the multi-source heterogeneous data to obtain a multimodal dataset includes: The sensor data in the multi-source heterogeneous data is segmented using the sliding window method and then normalized using Z-Score to obtain the sensor time series data. Missing values ​​are filled and normalized in the geological exploration structure data of the multi-source heterogeneous data to obtain structured parameters. The rock strata boundary features in the unstructured data of the multi-source heterogeneous data are extracted using a semantic segmentation network to obtain image features; The mining activity data was analyzed using natural language processing techniques to obtain text features; A spatiotemporal index table is established by associating the sensor time-series data, structured parameters, image features, and text features with the corresponding geological entities; A multimodal database is generated based on the spatiotemporal index table.

3. The method as described in claim 1, characterized in that, The step of constructing an initial dynamic graph of the geological entities of the mine based on the multimodal dataset, wherein the initial dynamic graph includes node feature vectors and initial adjacency relationships, includes: Geological entities are extracted from the multimodal dataset to obtain a set of entity nodes, specifically represented as follows: ; A node feature vector is generated for each entity based on the entity set. This node feature vector includes spatial coordinates, lithological parameter vectors, sensor data, and node type. The node type is categorized according to the node's region: static nodes represent unmined areas, dynamic nodes represent mining areas, and risk nodes represent areas with historical disasters. Specifically: in, For node feature vectors, Represents spatial coordinates, For nodes lithological parameter vector, For sensor data, For node type; The Euclidean distance between any two nodes is calculated based on the spatial coordinates in the node feature vector. When the Euclidean distance is less than a preset distance threshold, a spatial proximity edge is established, specifically as follows: in, For nodes and nodes The Euclidean distance between them For the preset distance threshold, For nodes spatial coordinates, For nodes spatial coordinates, Indicates spatially adjacent edges; The cosine similarity between any two nodes is calculated based on the lithological parameter vector in the node feature matrix. When the cosine similarity is greater than a preset similarity threshold, a geological attribute similarity edge is established, specifically as follows: in, Represents cosine similarity. Represents a node lithological parameter vector, Represents a node lithological parameter vector, Indicates edges with similar geological properties; The spatially adjacent edges and the edges with similar geological attributes are merged into an edge set to obtain an initial adjacency matrix; The set of entity nodes, the node feature vectors, and the initial adjacency matrix are combined to form an initial dynamic graph.

4. The method as described in claim 1, characterized in that, The step of updating the initial dynamic graph using a spatiotemporal graph attention network to obtain updated node features and dynamic adjacency weights includes: The node feature vectors and initial adjacency matrix in the initial dynamic graph are input into the spatial attention module of the spatiotemporal graph attention network. By aggregating the feature vectors of neighboring nodes, the local spatial features of each node are obtained, specifically as follows: in, It is a node The neighborhood group, Neighboring nodes The node feature vectors, Represents a node At any moment The local spatial feature vector is a representation obtained by aggregating the node features from the set of the node's neighbors. Indicates the number of elements in the neighbor set; The local spatial features are concatenated with the node feature vectors at the corresponding time steps to obtain a fused feature vector, specifically represented as follows: in, Represents the fused feature vector. This represents the activation function. Represents the learnable weight matrix. This represents the node feature vector at the corresponding time step; The fused feature vector and sensor data are input into the gated recurrent unit of the spatiotemporal graph attention network for updating, resulting in updated node features and dynamic adjacency weights.

5. The method as described in claim 4, characterized in that, The step of inputting the fused feature vector and sensor data into the gated recurrent unit of the spatiotemporal graph attention network for updating, to obtain the updated node features and dynamic adjacency weights, includes: The fused feature vector is input into the gated loop unit for optimization to obtain the optimized feature vector. The water pressure difference, crack permeability, and stress divergence are obtained based on the sensor data. The seepage path weights between nodes are calculated based on the water pressure difference and crack permeability, specifically as follows: in, Indicates crack permeability. Indicates the cross-sectional area of ​​the water passage. Indicates fluid dynamic viscosity, For nodes and nodes In time Water pressure difference, For time steps Time node and nodes The seepage path weights between them For nodes and nodes The Euclidean distance between them; The stress redistribution weight is calculated based on the projection of the stress divergence onto the inter-node direction, specifically expressed as follows: in, For nodes Stress divergence at the point, For nodes Pointing to node unit vector, For time steps Time node and nodes Stress redistribution weights between; The seepage path weight and the stress redistribution weight are fused to obtain the physical constraint weight, which is specifically expressed as follows: in, and For learnable parameters, For time steps Time node and nodes Physical constraint weights between them; Dynamic adjacency weights are generated based on the physical constraint weights and the optimized feature vectors, specifically as follows: in, It is a nonlinear function. For learnable weight matrix, For learnable parameter vectors, For nodes Optimized feature vectors It is a node The neighborhood group, For traversing nodes The index of all neighboring nodes, For nodes At time step Optimized feature vectors For time steps Time node and nodes Dynamic adjacency weights between them; The optimized feature vector is updated using an attention mechanism based on the dynamic adjacency weights to obtain the updated node features, as shown in the specific formula: in, This is the activation function.

6. The method as described in claim 1, characterized in that, The step of outputting the risk prediction results in a visual format and generating early warning information includes: The risk prediction results are analyzed to extract the spatiotemporal coordinates of the risk level and risk propagation path; Early warning information is generated based on the aforementioned risk level; A color gradient heatmap is rendered on a three-dimensional geological model based on the aforementioned spatiotemporal coordinates to obtain a dynamic risk heatmap. The dynamic risk heatmap is overlaid with the early warning information to generate a visual scene; The visualized scene is sent to the monitoring terminal so that the monitoring terminal can display the evacuation route and maintenance suggestions.

7. A mine geological risk early warning device based on multi-source heterogeneous data, characterized in that, The device includes: The acquisition module is used to collect multi-source heterogeneous data from the mine, including sensor data, geological exploration structure data, mining activity data and unstructured data. The unstructured data includes three-dimensional laser scanning point clouds, geological profile maps and borehole videos. The processing module is used to standardize the multi-source heterogeneous data to obtain a multimodal dataset; The construction module is used to construct an initial dynamic graph of the geological entities of the mine based on the multimodal dataset, wherein the initial dynamic graph includes node feature vectors and initial adjacency relationships, and the geological entities include roadways, rock strata, faults and goafs; The update module is used to update the initial dynamic graph through a spatiotemporal graph attention network to obtain updated node features and dynamic adjacency weights. The result module is used to predict risks based on the updated node features and dynamic adjacency weights through a dynamic risk association network, obtaining risk prediction results. The dynamic risk association network includes a landslide risk prediction head, a water inrush risk prediction head, and a graph neural network. It is also used to concatenate the updated node features with the sensor data of the corresponding nodes to obtain fused physical monitoring features; input the fused physical monitoring features into the landslide risk prediction head to obtain the landslide probability of each node; input the fused physical monitoring features into the water inrush risk prediction head to obtain the water inrush risk index of each node; assign a risk level to each node according to the landslide probability and the water inrush risk index to obtain a risk level set; perform diffusion simulation along the direction of maximum weight in the graph neural network based on the dynamic adjacency weights to obtain a risk propagation path; and combine the risk level set with the risk propagation path to obtain the risk prediction results. The execution module is used to output the risk prediction results in a visual format and generate early warning information.

8. A mine geological risk early warning device based on multi-source heterogeneous data, characterized in that, The device includes: a memory, a processor, and a mine geological risk early warning program based on multi-source heterogeneous data stored in the memory and running on the processor, wherein the mine geological risk early warning program based on multi-source heterogeneous data is configured to implement the steps of the mine geological risk early warning method based on multi-source heterogeneous data as described in any one of claims 1-6.

9. A storage medium, characterized in that, The storage medium stores a mine geological risk early warning program based on multi-source heterogeneous data. When the processor executes the mine geological risk early warning program based on multi-source heterogeneous data, it implements the steps of the mine geological risk early warning method based on multi-source heterogeneous data as described in any one of claims 1-6.

Images