Mine safety decision deduction method based on digital twinning

Abstract

The invention discloses a mine safety decision deduction method based on digital twinning, and relates to the technical field of mineral engineering, and the method comprises the steps: building a three-dimensional geometric mine structure model based on a multi-source fusion data set, carrying out the attribute injection and discretization processing of the three-dimensional geometric mine structure model, and generating a real-time mine digital twinning body; based on the multi-source fusion data set, mine element nodes and relation edges are defined, and a security-process-resource knowledge graph is constructed; obtaining a mine state field from the safety-process-resource knowledge graph and the real-time mine digital twinborn body, and constructing a mine safety decision state space and an action space; and identifying a decision disturbance event in the multi-source monitoring data according to the trigger threshold, and constructing a mine safety decision reward function in combination with the mine safety decision state space and the action space. The semantic reasoning capability of the decision making process is improved, and the effects of improving the decision making accuracy, real-time performance and intelligent predictability are achieved.

Description

Technical Field

[0001] This invention relates to the field of mining engineering technology, and in particular to a method for mine safety decision-making simulation based on digital twins. Background Technology

[0002] In the field of mine safety production, conventional safety decision-making methods are typically based on multi-source data acquisition and processing systems, integrating geological exploration data, engineering drawings, and real-time sensor monitoring information, and improving data quality through anomaly detection and data cleaning techniques. These methods rely on coal mine safety regulations and production process documents, employing rule engines or manual analysis for risk identification and response decisions. With the development of information technology, some systems have introduced data fusion and visualization tools to monitor and alarm environmental parameters to support daily safety management. These methods have formed standardized processes through long-term application, providing a fundamental guarantee for mine operation safety, and are gradually integrating into the trend of digital management, emphasizing data reliability and the timeliness of decision-making.

[0003] However, conventional methods for 3D mine structure modeling often rely on 2D drawings or simplified geometric representations, which are insufficient to accurately express the complex spatial topological relationships of entities such as roadways and working faces, thus limiting the accuracy of hazard location. At the same time, in terms of decision support, existing technologies focus on real-time alarms and static rule applications, lacking inference mechanisms that integrate knowledge graphs and dynamic physical fields, and are unable to simulate the interactive effects of process sequences and resource scheduling, thereby restricting the foresight and adaptability of decision-making. These shortcomings make the safety decision-making process often lag behind changes in operating conditions and make it difficult to cope with risks caused by the coupling of multiple factors. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a mine safety decision-making simulation method based on digital twins to solve the problems of low model integration and insufficient dynamic simulation capabilities.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a mine safety decision-making simulation method based on digital twins, which includes:

[0008] Collect raw data from multiple sources in the mine and preprocess it to generate a multi-source fusion dataset;

[0009] Based on a multi-source fusion dataset, a three-dimensional geometric mine structure model is constructed, and attribute injection and discretization processing are performed on the three-dimensional geometric mine structure model to generate a real-time mine digital twin.

[0010] Based on a multi-source fusion dataset, we define mine element nodes and relationship edges to construct a safety-process-resource knowledge graph.

[0011] Mine status fields are obtained from the safety-process-resource knowledge graph and the real-time mine digital twin, and a mine safety decision-making state space and action space are constructed.

[0012] Based on the trigger threshold, decision disturbance events in multi-source monitoring data are identified, and a mine safety decision reward function is constructed by combining the mine safety decision state space and action space.

[0013] Multi-source monitoring data is injected into a real-time mine digital twin to screen out executable sequences for mine safety decisions. The reward value of the executable sequences is calculated using a mine safety decision reward function, and then rearranged into recommended sequences for mine safety decisions.

[0014] A reconstruction operation is performed on the recommended sequence for mine safety decisions, the deviation value of the action effect is obtained and the parameters of the real-time mine digital twin are updated, and a mine safety decision simulation report is generated.

[0015] As a preferred embodiment of the mine safety decision-making simulation method based on digital twins described in this invention, the preprocessing includes anomaly identification, anomaly correction, and spatiotemporal alignment processing.

[0016] The multi-source fusion dataset includes process specification data, mine foundation engineering data, geological datasets, and multi-source monitoring data.

[0017] As a preferred embodiment of the mine safety decision-making simulation method based on digital twins described in this invention, the steps for constructing a three-dimensional geometric mine structure model based on a multi-source fusion dataset are as follows:

[0018] Perform coordinate transformation on the mine foundation engineering and geological datasets to generate a mine foundation dataset with unified coordinates;

[0019] Extract three-dimensional geometric features from a coordinate-unified mine basic dataset, obtain spatial topological relationships, and generate entity-connected spatial structures;

[0020] Based on the entity-connected spatial structure, structured roadway entities, working face entities, and shaft entities are constructed, and attribute parameters are bound to all entities to form a three-dimensional geometric mine structure model.

[0021] As a preferred embodiment of the mine safety decision-making simulation method based on digital twins described in this invention, the steps for injecting attributes and discretizing the three-dimensional geometric mine structure model to generate a real-time mine digital twin are as follows:

[0022] Perform attribute injection operations on the three-dimensional geometric mine structure model to generate a three-dimensional static digital twin skeleton of the mine;

[0023] Discretize the three-dimensional static digital twin skeleton of the mine to generate a real-time digital twin of the mine.

[0024] As a preferred embodiment of the mine safety decision-making simulation method based on digital twins described in this invention, the steps for defining mine element nodes and relationship edges based on a multi-source fusion dataset and constructing a safety-process-resource knowledge graph are as follows:

[0025] The process specification data is structured to generate structured data of processes and resources;

[0026] Extract and align fields from structured data of processes and resources to generate risk-triggered correlation data;

[0027] Based on risk-triggered correlation data and the three-dimensional static digital twin skeleton of the mine, process nodes, resource nodes, risk nodes and spatial location nodes are defined, and the relationship edges between nodes are defined to construct a safety-process-resource knowledge graph.

[0028] As a preferred embodiment of the mine safety decision-making simulation method based on digital twins described in this invention, the steps of obtaining mine status fields from the safety-process-resource knowledge graph and the real-time mine digital twin, and constructing the mine safety decision-making state space and action space are as follows.

[0029] Physical field state discretization is performed on the real-time mine digital twin to obtain the physical field state field;

[0030] Map the mine safety-process-resource knowledge graph to obtain the process status field and resource status field;

[0031] Multi-source monitoring data is mapped to a real-time mine digital twin by spatial location mapping. A multi-level state expression structure is established by combining physical field state fields, process state fields and resource state fields, and uniformly encoded into a mine safety decision state space.

[0032] The knowledge rules in the screening process and resource structured data are examined for feasibility and a space for mine safety decision-making actions is generated.

[0033] As a preferred embodiment of the mine safety decision-making simulation method based on digital twins described in this invention, the steps of identifying decision disturbance events in multi-source monitoring data according to trigger thresholds and constructing a mine safety decision-making reward function by combining the mine safety decision-making state space and action space are as follows:

[0034] Based on the trigger condition fields in the process specification data, identify the fields in the multi-source monitoring data that match the trigger threshold, and integrate them into a set of decision disturbance events;

[0035] The corresponding benefit and loss fields are extracted from the mine safety decision state space, mine safety decision action space and process procedure data, and combined with the set of decision disturbance events to construct the mine safety decision reward function.

[0036] As a preferred embodiment of the mine safety decision-making simulation method based on digital twins described in this invention, the steps of injecting multi-source monitoring data into a real-time mine digital twin and selecting executable sequences for mine safety decisions are as follows:

[0037] The monitoring fields from the multi-source monitoring data are inserted into the real-time mine digital twin to update the physical field and generate the current mine multi-physics field state.

[0038] The current multiphysics state of the mine and the mine safety decision-making action space are matched to select executable candidate actions and combine them into an executable sequence for mine safety decisions.

[0039] As a preferred embodiment of the mine safety decision-making simulation method based on digital twins described in this invention, the steps of calculating the reward value of the executable sequence of safety decisions through the mine safety decision reward function and rearranging it into a recommended sequence of mine safety decisions are as follows:

[0040] Based on the operational constraint field in the process specification data, the executable sequence of mine safety decision is checked for constraints, candidate actions that meet the constraints are identified, and combined into a candidate sequence of mine safety decision.

[0041] The candidate sequence for mine safety decision is input into the mine safety decision reward function, the positive and negative values ​​are accumulated and calculated, and it is matched with the mine safety decision state space to output the reward value. The result is then sorted according to the reward value to generate the recommended sequence for mine safety decision.

[0042] As a preferred embodiment of the mine safety decision-making simulation method based on digital twins described in this invention, the steps of performing a reconstruction operation on the mine safety decision recommendation sequence, obtaining the action effect deviation value and updating the parameters of the real-time mine digital twin, and generating a mine safety decision-making simulation report are as follows:

[0043] The mine safety decision recommendation sequence is mapped to a real-time mine digital twin to capture changes in the physical field and record them as decision execution feedback data.

[0044] Obtain the action effect deviation value of the decision execution feedback data, update the parameters of the real-time mine digital twin, and summarize them into a mine safety decision simulation report.

[0045] The beneficial effects of this invention are as follows: by constructing a three-dimensional geometric mine structure model and a real-time mine digital twin, a high-precision dynamic digital mapping of the mine physical environment is achieved, providing a virtual space foundation for safety decision-making; by constructing a safety-process-resource knowledge graph, a structured association between processes, resources and risk factors is realized, improving the semantic reasoning ability of the decision-making process, and achieving the effect of improving the accuracy, real-time performance and intelligent predictability of decision-making. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of a mine safety decision-making simulation method based on digital twins.

[0048] Figure 2 This is a flowchart of data acquisition and preprocessing.

[0049] Figure 3 A flowchart for building 3D models and knowledge graphs.

[0050] Figure 4 This is a flowchart for decision-making simulation. Detailed Implementation

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0052] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0054] Reference Figures 1-4As one embodiment of the present invention, this embodiment provides a mine safety decision-making simulation method based on digital twins, comprising the following steps:

[0055] S1. Collect raw data from multiple sources in the mine and preprocess it to generate a multi-source fusion dataset.

[0056] Furthermore, multi-source raw data from the mine is collected. Specifically, raw process specification data is obtained by scanning coal mine safety regulations and production process documents, and then broken down into process step fields, resource usage fields, operating condition fields, trigger condition fields, and operating restriction condition fields; geological surveys are conducted on exploration boreholes to obtain raw geological datasets, which are then broken down into stratigraphic structure fields, fault parameter fields, aquifer parameter fields, gas occurrence fields, and rock mass physical property fields; raw mine foundation engineering data, including roadway entity, working face entity, and shaft entity data, is obtained from engineering drawings and design reports; and raw multi-source monitoring data (including gas concentration, temperature, pressure, personnel location, and equipment status) is obtained through real-time monitoring of continuous data by sensors, and then broken down into monitoring fields, timestamp fields, and spatial location fields.

[0057] It should be noted that the coal mine safety regulations document is a mandatory standard document promulgated by the Ministry of Emergency Management, which includes coal mine geology, underground coal mines, open-pit coal mines, safety management and supervision, and supplementary provisions; the production process document refers to a standardized engineering design document compiled based on the original geological dataset during the mine construction phase, used to guide specific production operation processes.

[0058] Preprocessing includes anomaly identification, anomaly correction, and spatiotemporal alignment.

[0059] Furthermore, anomaly identification and correction are performed on the multi-source raw data from the mine. Specifically, all fields in the original process specification data are scanned to detect anomalies that do not conform to the coal mine safety regulations, and these anomalies are replaced according to the standard process specifications. For the original mine foundation engineering, the rationality of the coordinates of the roadway entities is checked through geometric verification, anomalies exceeding the preset anomaly range are identified, and the coordinate values ​​are corrected with reference to the engineering drawings. For the original geological dataset and original multi-source monitoring data, the mean and standard of each field are obtained from the original geological dataset and original multi-source monitoring data using Z-score, and linear interpolation is used to apply positive values ​​around the anomalies. The coordinates of constant data points are used to obtain the equations of straight lines and correct outliers. Spatiotemporal alignment processing is performed on the multi-source raw data of the mine for anomaly identification and correction. Specifically, based on the global coordinate system in the original mine foundation engineering and the minimum timestamp of the original multi-source monitoring data, a spatial grid and standard time point are defined to establish a unified spatiotemporal reference frame. The timestamp field of each mine multi-source raw data (such as the timestamp field of monitoring data) is aligned to the nearest standard time point. The spatial location field of the geological dataset and the coordinate field of the mine foundation engineering are mapped to the unified spatiotemporal reference frame through the nearest neighbor interpolation algorithm to ensure spatial overlap and generate a multi-source fusion dataset.

[0060] It should be noted that the preset range is defined by the minimum and maximum coordinate values ​​extracted from the engineering drawings, for example, the x-coordinate range is 0 to 1000 meters.

[0061] The multi-source fusion dataset includes process specification data, mine foundation engineering data, geological datasets, and multi-source monitoring data.

[0062] S2. Based on the multi-source fusion dataset, a three-dimensional geometric mine structure model is constructed, and attribute injection and discretization processing are performed on the three-dimensional geometric mine structure model to generate a real-time mine digital twin.

[0063] Perform coordinate transformation on the mine foundation engineering and geological datasets to generate a mine foundation dataset with unified coordinates.

[0064] Furthermore, homogeneous coordinate equations are constructed based on the endpoint coordinates of the roadway entities in the mine foundation engineering as control points. The transformation matrix is ​​solved by least-squares fitting, and the local coordinate system of the geological dataset is mapped to the global coordinate system of the mine foundation engineering according to the transformation matrix to achieve coordinate alignment. The endpoints of the entities in the mine foundation engineering are triangulated using the Delaunay triangulation algorithm to generate triangular meshes. Based on the adjacency relationship of triangles in the triangular mesh, overlapping areas with triangle areas less than the overlap threshold are identified. The overlapping areas are corrected by adjusting the coordinate points and inserting new coordinate points. Based on a predefined attribute dictionary, the data fields are traversed and replaced with unified codes to generate a mine foundation dataset with unified coordinates.

[0065] It should be noted that the overlap threshold is obtained by traversing all roadway entities, obtaining the projected area of ​​each entity, and using a specific percentile of the area value (e.g., the 10th percentile) as the overlap threshold, for example, an area of ​​0.1 square meters; the predefined attribute dictionary is set based on the coal mine safety regulations document standard and geological dataset, and is replaced with a unified code by traversing the data fields of the geological dataset, for example, mapping the stratigraphic structure field to an integer code, where 1 represents "coal seam" and 2 represents "rock strata", with a value range of 1-10.

[0066] Three-dimensional geometric features are extracted from a coordinate-unified mine basic dataset, and spatial topological relationships are obtained to generate a spatial structure connecting entities.

[0067] Furthermore, three-dimensional geometric features are extracted from the coordinate-unified mine basic dataset. Based on the vertex coordinates and boundary line data in the three-dimensional geometric features, the geometric contours of roadway entities, working face entities, and shaft entities are identified. The geometric contours are then triangulated again using the Delaunay triangulation algorithm to generate triangular mesh surfaces. The edge connectivity of triangles in the triangular mesh is identified to obtain the spatial topological relationships between entities, such as adjacency and containment relationships. Based on the spatial topological relationships, a graph structure is constructed, where nodes represent entity geometric elements and edges represent connection relationships, generating a spatial structure connecting entities.

[0068] Based on the entity-connected spatial structure, structured roadway entities, working face entities, and shaft entities are constructed, and attribute parameters are bound to all entities to form a three-dimensional geometric mine structure model.

[0069] Furthermore, based on the spatial structure of the entity connection, the geometric boundary node sets of the roadway entity, working face entity, and shaft entity are identified, and a three-dimensional voxel mesh is constructed. The coordinates of the geometric boundary node set are mapped to the mesh vertices, and the Euclidean distance from the vertex of each voxel (a volume element in three-dimensional space, similar to a two-dimensional pixel) to the nearest entity surface is calculated as the distance field value. Through the traveling cube algorithm, each voxel is traversed, and a sequence of triangular facets is generated according to the positive or negative sign of the distance field value at the voxel vertex to complete surface reconstruction and generate a triangular mesh model. Data fields (such as stratigraphic structure fields and fault parameter fields) are extracted from the coordinate-unified mine basic dataset and mapped to the triangular mesh model to generate a three-dimensional geometric mine structure model.

[0070] It should be noted that the moving cube algorithm is a three-dimensional mesh reconstruction method that generates a sequence of triangular faces in a three-dimensional voxel mesh based on the distance field value of the voxel vertices and completes the surface reconstruction.

[0071] Perform attribute injection operations on the three-dimensional geometric mine structure model to generate a three-dimensional static digital twin skeleton of the mine.

[0072] Furthermore, spatial coordinate data of vertices and faces are extracted from the three-dimensional geometric mine structure model. The nearest neighbor search algorithm is used to extract data fields (such as process step fields in process specification data) from the multi-source fusion dataset, and these fields are bound to the corresponding vertices and faces through key-value pair mapping. Vertices and faces of all injected data fields are integrated to generate a three-dimensional static digital twin skeleton of the mine.

[0073] Discretize the three-dimensional static digital twin skeleton of the mine to generate a real-time digital twin of the mine.

[0074] Furthermore, vertex normals are obtained from the 3D static digital twin skeleton of the mine, and the Euclidean distance difference between the normals of the vertex and the normals of adjacent vertices is calculated to obtain the local curvature value. Based on the local curvature value, the voxel size is dynamically adjusted to generate a non-uniform 3D voxel mesh. The physical field (the distribution of multi-physical data in the mine), such as the temperature field, pressure field, and gas concentration field, is extracted from the 3D static digital twin skeleton of the mine using the bilinear interpolation method and assigned to each voxel. The time discretization parameter is set (e.g., the time step is defined as 1 second), and different physical equations are selected according to the physical field and discretized into finite difference form (e.g., the heat conduction equation is selected for the temperature field) to construct the state transition function. The state value of the next time step (including the values ​​of temperature, pressure, and gas concentration) is obtained through the state transition function, and the corresponding voxels are updated to generate a real-time digital twin of the mine.

[0075] S3. Based on the multi-source fusion dataset, define mine element nodes and relationship edges to construct a safety-process-resource knowledge graph.

[0076] The process specification data is structured to generate structured data of processes and resources.

[0077] Furthermore, process procedure data is extracted from the multi-source fusion dataset. Based on the rule base of coal mine safety procedure documents, regular expression algorithms are used to match keywords (such as "tunneling" and "recovery") in the process step fields, and the equipment names and quantities in the resource occupancy field are obtained. According to the text position relationship between the process step fields and the operation condition fields and trigger condition fields, such as line number differences, paragraph positions, and the order and proximity of keywords in the fields, the operation condition fields and trigger condition fields are bound to the corresponding process step fields to form a process sequence. The keywords in the process step fields, the equipment names and quantities in the resource occupancy field, and the process sequence are organized into a structured table format to generate structured process and resource data.

[0078] It should be noted that regular expression algorithms are text parsing methods that match keywords (such as "tunneling" or "recovery") from process step fields according to character patterns to identify the semantic content of the process.

[0079] Extract and align fields from structured data of processes and resources to generate risk-triggered correlation data.

[0080] Furthermore, process step fields, resource occupancy fields, and trigger condition fields are extracted from the structured data of processes and resources. Based on the risk rule base of the coal mine safety regulations document (a set of rules composed of risk keywords extracted from the coal mine safety regulations document and corresponding trigger conditions), risk keywords (such as "gas exceeding the limit" or "roof pressure") in the trigger condition fields are scanned, and the trigger condition fields are aligned with the process step fields. For example, when the process step field contains "tunneling", it is associated with "gas exceeding the limit" in the trigger condition field. The aligned fields are then combined into risk trigger association data.

[0081] Based on risk-triggered correlation data and the three-dimensional static digital twin skeleton of the mine, process nodes, resource nodes, risk nodes and spatial location nodes are defined, and the relationship edges between nodes are defined to construct a safety-process-resource knowledge graph.

[0082] Furthermore, process step fields, resource occupancy fields, and risk trigger fields are extracted from the risk trigger association data to define process nodes, resource nodes, and risk nodes. Spatial coordinate data is extracted from the three-dimensional static digital twin skeleton of the mine to define spatial location nodes. Based on the risk trigger association data (e.g., the correspondence between process step fields and risk trigger fields), a rule matching algorithm is used to scan the field correspondences and define the relationship edges between nodes. For example, when the process node "tunneling" is associated with the risk node "gas exceeding the limit", a "trigger" edge is added, or when the resource node "coal mining machine" is associated with the process node "recovery", a "use" edge is added. All nodes and relationship edges are integrated into a graph data structure to generate a safety-process-resource knowledge graph.

[0083] S4. Obtain mine status fields from the safety-process-resource knowledge graph and the real-time mine digital twin, and construct the mine safety decision-making state space and action space.

[0084] Physical field state discretization is performed on the real-time mine digital twin to obtain the physical field state field.

[0085] Furthermore, based on the voxel grid of the real-time mine digital twin, the partial derivatives of each physical field (e.g., temperature field, pressure field, and gas concentration field) in three-dimensional space are obtained as gradient values. Regions where the gradient values ​​exceed the gradient threshold are identified as high-variable regions using the finite difference method. The sampling point density is increased in the high-variable regions, and the physical fields of the sampling points are extracted from the voxel grid using a bilinear interpolation algorithm. The physical fields of the sampling points are organized into a structured array to form the physical field state field.

[0086] It should be noted that the gradient threshold is set based on the gradient distribution statistics of historical multi-source monitoring data. It is set by calculating a specific percentile (e.g., the 90th percentile) of the gradient values ​​in historical data. For example, the temperature gradient threshold is set to 2°C / meter. The value range is usually the interval between the minimum and maximum values ​​of the historical physical field gradient data, such as 0.5°C / meter to 10°C / meter.

[0087] Map the mine safety-process-resource knowledge graph to obtain the process status field and resource status field.

[0088] Furthermore, the mine safety-process-resource knowledge graph is mapped. Specifically, a breadth-first search algorithm is used to traverse all process nodes and resource nodes in the knowledge graph. The status value of each process node is extracted from each process node. For example, the status value of a process node includes "in progress" or "completed", generating a process status field. The status value of a resource node is extracted from the resource node. For example, the status value of a resource node includes "available" or "occupied", generating a resource status field.

[0089] Multi-source monitoring data is mapped to a real-time mine digital twin through spatial location mapping. A multi-level state expression structure is established by combining physical field state fields, process state fields, and resource state fields, and uniformly encoded into a mine safety decision state space.

[0090] Furthermore, a KD-tree spatial index is constructed based on the voxel grid coordinates of the real-time mine digital twin. The KD-tree insertion algorithm is used to insert the coordinates of each voxel's center point as a node into the tree structure. For each spatial location point of the multi-source monitoring data, the Euclidean distance is calculated using the KD-tree nearest neighbor search algorithm to find the voxel grid coordinates with the smallest distance, which are then used as the nearest neighbor coordinates. Monitoring fields (e.g., gas concentration and temperature values) are mapped to their corresponding nearest neighbor coordinates. The physical field status field, process status field, and resource status field are weighted and averaged, and the field values ​​are integrated according to a spatial hierarchy (e.g., roadway level and working face level) to generate integrated status data. Through hierarchical encoding, the integrated status data is mapped into a three-dimensional array according to spatial hierarchy, physical field type, and status category. The first dimension represents the spatial grid index, the second dimension represents the physical field type, and the third dimension represents the process and resource status, establishing a multi-level status expression structure. Finally, normalization processing is used to uniformly scale the three-dimensional array to a fixed range, such as [0, 1], to generate the mine safety decision-making status space.

[0091] The knowledge rules in the screening process and resource structured data are examined for feasibility and a space for mine safety decision-making actions is generated.

[0092] Furthermore, process step fields, resource occupancy fields, and operation constraint fields are extracted from the structured data of processes and resources. Operation sequence rules are read from the process step field (e.g., in coal mine production, the process step field contains the specific sequence that "tunneling" must be performed before "mining"), equipment allocation rules are read from the resource occupancy field (e.g., in the "tunneling" process, the resource occupancy field specifies the use of "tunneling machine" equipment and a quantity of two units), and safety constraint rules are read from the operation constraint field (e.g., the gas concentration must not exceed 1.0% or the operation time is limited to 8 hours). The process involves: forming a set of candidate actions; traversing each candidate action and checking whether the operation sequence rules conflict with the process status field in the mine safety decision state space, such as verifying whether the "tunneling" process is executed before the "recovery" process; checking whether the equipment allocation rules match the resource status field, such as confirming whether the number of "tunneling machines" is available and not occupied; checking whether the safety constraint rules meet the real-time monitoring data, such as comparing whether the gas concentration value is lower than 1.0%; eliminating infeasible rules, encoding the verified candidate actions into a structured action list, and generating the mine safety decision action space.

[0093] S5. Identify decision disturbance events in multi-source monitoring data based on trigger thresholds, and construct a mine safety decision reward function by combining the mine safety decision state space and action space.

[0094] Based on the trigger condition fields in the process specification data, fields that match the trigger thresholds in multi-source monitoring data are identified and integrated into a set of decision disturbance events.

[0095] Furthermore, based on the triggering condition fields in the process specification data, the triggering thresholds of the triggering condition fields are extracted from the multi-source fusion dataset, including the gas concentration threshold (e.g., 1.0%), temperature threshold (e.g., 30°C), and pressure threshold (e.g., 101.3 kPa). The monitoring fields in the multi-source monitoring data are traversed, and the monitoring fields that reach the corresponding triggering thresholds are identified point by point (e.g., when the gas concentration monitoring field reaches 1.5%, it is marked as a matching event), and the corresponding timestamp field and spatial location field are recorded to form event instances. All event instances are aggregated into a structured list to generate a set of decision disturbance events.

[0096] It should be noted that the gas concentration threshold, temperature threshold, and pressure threshold are set based on the safety standards in the coal mine safety regulations document and are directly extracted from the process regulations data. For example, the gas concentration threshold is usually in the range of 0.5% to 2.0%, the temperature threshold is usually in the range of 20°C to 40°C, and the pressure threshold is usually in the range of 90 kPa to 110 kPa.

[0097] The corresponding benefit and loss fields are extracted from the mine safety decision state space, mine safety decision action space and process procedure data, and combined with the set of decision disturbance events to construct the mine safety decision reward function.

[0098] Furthermore, based on the physical field state fields (such as temperature and pressure field values) in the mine safety decision state space and the trigger thresholds (such as a gas concentration threshold of 1.0%) in the decision disturbance event set, the physical field state fields are compared with the trigger thresholds. Physical field state fields below the trigger threshold are defined as benefit fields, and the total number of benefit fields is counted. Physical field state fields above the trigger threshold are defined as loss fields, and the total number of loss fields is counted. Verified candidate actions are extracted from the mine safety decision action space. The equipment utilization efficiency in the resource occupancy field is read. Candidate actions with equipment utilization efficiency higher than the efficiency threshold are judged as high-efficiency actions and marked as benefit fields, with weights assigned to the benefit fields. Candidate actions with equipment utilization efficiency not higher than the efficiency threshold are judged as inefficient actions and marked as loss fields, with weights assigned to the loss fields. Operational constraint fields (such as maximum) are extracted from the process specification data. The system compares monitoring fields (such as actual operation time and real-time ventilation volume) in the mine safety decision-making state space with operation constraint fields, based on the 8-hour duration of major operations and the minimum ventilation volume standard. Monitoring fields that meet the constraints (e.g., operation time ≤ 8 hours, ventilation volume ≥ standard value) are integrated into benefit fields; those that do not meet the constraints (e.g., operation time > 8 hours, ventilation volume < standard value) are integrated into loss fields. Based on the set of decision disturbance events, the number of events is obtained, a weight is assigned to each event, and the severity value of the decision disturbance event is obtained. A reward function expression is constructed. The total number of benefit fields, the weight of benefit fields, the specific value of benefit fields, the total number of loss fields, the weight of loss fields, the specific value of loss fields, the number of events in the decision disturbance event set, the weight of decision disturbance events, and the severity value of decision disturbance events are substituted into the reward function expression to generate a complete mine safety decision-making reward function.

[0099] It should be noted that the efficiency threshold is set based on coal mine safety regulations documents and process regulations data. It is directly set by extracting the equipment's output per unit time index (such as the amount of coal mined per unit time by a tunneling machine) from the process regulations data as a benchmark value. The range of values ​​depends on the type of equipment. For example, the benchmark value range for the amount of coal mined per unit time may be 10-100 tons / hour.

[0100] The expression for the reward function for mine safety decisions is:

[0101] ;

[0102] in, This is the reward value; This represents the total number of revenue fields. For the first The numerical weight of the physical field state field for each benefit field; For the first The value of the physical field state field for each revenue field; The index is the ordinal number of the revenue field; This represents the total number of loss fields; For the first The numerical weights of the physical state field for each loss field; For the first The value of the physical state field of each loss field; The index is the ordinal number of the loss field; The number of events in the set of decision disturbance events; To indicate the first The severity value weight of each decision disturbance event; For the first The severity value of each decision disturbance event; This is the sequence index for decision-making disturbance events.

[0103] S6. Inject multi-source monitoring data into the real-time mine digital twin, screen out executable sequences for mine safety decisions, calculate the reward value of the executable sequences for safety decisions using the mine safety decision reward function, and rearrange them into recommended sequences for mine safety decisions.

[0104] The monitoring fields from the multi-source monitoring data are inserted into the real-time mine digital twin to update the physical field and generate the current multi-physics state of the mine.

[0105] Furthermore, based on the nearest neighbor coordinates in the voxel grid of the real-time mine digital twin, the monitoring fields (such as gas concentration and temperature value) are mapped to the voxels with the corresponding nearest neighbor coordinates. The bilinear interpolation method is used to update the physical field properties in the voxels according to the monitoring fields, such as temperature field and pressure field, to ensure that the data is synchronized with the current timestamp. All updated voxel physical field states are integrated to generate the mine multi-physics field state at the current moment.

[0106] The current multiphysics state of the mine and the mine safety decision-making action space are matched to select executable candidate actions and combine them into an executable sequence for mine safety decisions.

[0107] Furthermore, each candidate action in the mine safety decision-making action space is traversed, and the execution prerequisite fields of the candidate actions are extracted. For example, the equipment status must be "available" or the monitoring parameters must be lower than the trigger threshold. The execution prerequisite fields of the candidate actions are compared in real time with the corresponding fields in the mine multiphysics field status at the current moment. For example, it is checked whether the equipment status field matches "available" and whether the gas concentration field is lower than 1.0%. If all prerequisites are met, the candidate action is marked as executable, and all executable candidate actions are combined into an executable sequence of mine safety decisions according to the process sequence.

[0108] Based on the operational constraint fields in the process specification data, the executable sequences for mine safety decisions are checked for constraints, candidate actions that meet the constraints are identified, and combined into candidate sequences for mine safety decisions.

[0109] Furthermore, operational constraint fields are extracted from the process specification data, such as a gas concentration threshold of 1.0% and an operation time limit of 8 hours. Each executable candidate action in the executable sequence of mine safety decisions is traversed, and the monitoring fields of the executable candidate actions (such as real-time gas concentration value and process time) are compared with the operational constraint fields. A condition judgment algorithm is used for checking, such as a gas concentration value below 1.0% and a process time not exceeding 8 hours. If all constraints are met, the current candidate action is marked as a candidate action that meets the constraints. All marked candidate actions are combined without following the order of the executable sequence of mine safety decisions to generate a candidate sequence of mine safety decisions.

[0110] The candidate sequence for mine safety decision is input into the mine safety decision reward function, the positive and negative values ​​are accumulated and calculated, and it is matched with the mine safety decision state space to output the reward value. The result is then sorted according to the reward value to generate the recommended sequence for mine safety decision.

[0111] Furthermore, each candidate action sequence in the mine safety decision candidate sequence is traversed. Based on the index of each candidate action, the state field of the current candidate action is extracted from the mine safety decision state space, including the physical field state field, process state field, and resource state field. The state field is then input into the mine safety decision reward function to calculate the sum of the positive values ​​of the benefit field, the sum of the negative values ​​of the loss field, and the sum of the negative values ​​of the decision disturbance event penalty. The reward value of each candidate action is then calculated by accumulating these values. The reward values ​​of all candidate action sequences are sorted in descending order to generate the mine safety decision recommendation sequence.

[0112] S7. Perform a reconstruction operation on the recommended sequence for mine safety decisions, obtain the deviation value of the action effect, update the parameters of the real-time mine digital twin, and generate a mine safety decision simulation report.

[0113] The recommended sequence of mine safety decisions is mapped onto a real-time digital twin of the mine to capture changes in the physical field and record them as decision execution feedback data.

[0114] Furthermore, each candidate action in the mine safety decision recommendation sequence is traversed, and the current timestamp field, process step field, and associated spatial location field of the candidate action are extracted. The KD-tree nearest neighbor search algorithm is used to find the coordinates of the voxel center point that matches the spatial location field. The updated values ​​of the temperature field and pressure field in the voxel grid are obtained through the state transition function to simulate the physical field changes after the process step is executed. The physical field state fields before and after the update are compared to obtain the physical field change of each voxel, including the absolute value difference and gradient change. The physical field change, process step field, current timestamp field, and spatial location field are integrated into a key-value pair structure to generate decision execution feedback data.

[0115] Obtain the action effect deviation value of the decision execution feedback data, update the parameters of the real-time mine digital twin, and summarize them into a mine safety decision simulation report.

[0116] Furthermore, the physical field change, process step field, and timestamp field are extracted from the decision execution feedback data and compared with the expected state value in the mine safety decision state space. The absolute deviation between the physical field state field of each voxel and the expected value (i.e., the state value of the next time step in the real-time mine digital twin) is calculated. Using the least squares optimization method, with the goal of minimizing the deviation, the parameters of the state transition function in the real-time mine digital twin are adjusted. The action effect deviation value, the updated parameters of the state transition function, the process step field, and the timestamp field are integrated into a structured log to generate a mine safety decision simulation report.

[0117] In summary, this invention achieves high-precision dynamic digital mapping of the mine's physical environment by constructing a three-dimensional geometric mine structure model and a real-time mine digital twin, providing a simulateable virtual space foundation for safety decision-making. By constructing a safety-process-resource knowledge graph, it realizes the structured association between processes, resources, and risk factors, improving the semantic reasoning ability of the decision-making process and achieving the effect of enhancing the accuracy, real-time performance, and intelligent predictability of decision-making.

[0118] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A mine safety decision-making simulation method based on digital twins, characterized in that: include, Collect raw data from multiple sources in the mine and preprocess it to generate a multi-source fusion dataset; Based on a multi-source fusion dataset, a three-dimensional geometric mine structure model is constructed, and attribute injection and discretization processing are performed on the three-dimensional geometric mine structure model to generate a real-time mine digital twin. Based on a multi-source fusion dataset, we define mine element nodes and relationship edges to construct a safety-process-resource knowledge graph. Mine status fields are obtained from the safety-process-resource knowledge graph and the real-time mine digital twin, and a mine safety decision-making state space and action space are constructed. Based on the trigger threshold, decision disturbance events in multi-source monitoring data are identified, and a mine safety decision reward function is constructed by combining the mine safety decision state space and action space. Multi-source monitoring data is injected into a real-time mine digital twin to screen out executable sequences for mine safety decisions. The reward value of the executable sequences is calculated using a mine safety decision reward function, and then rearranged into recommended sequences for mine safety decisions. A reconstruction operation is performed on the recommended sequence for mine safety decisions, the deviation value of the action effect is obtained and the parameters of the real-time mine digital twin are updated, and a mine safety decision simulation report is generated.

2. The mine safety decision-making simulation method based on digital twins as described in claim 1, characterized in that: The preprocessing includes anomaly identification, anomaly correction, and spatiotemporal alignment processing; The multi-source fusion dataset includes process specification data, mine foundation engineering data, geological datasets, and multi-source monitoring data.

3. The mine safety decision-making simulation method based on digital twins as described in claim 2, characterized in that: The steps for constructing a three-dimensional geometric mine structure model based on a multi-source fusion dataset are as follows. Perform coordinate transformation on the mine foundation engineering and geological datasets to generate a mine foundation dataset with unified coordinates; Extract three-dimensional geometric features from a coordinate-unified mine basic dataset, obtain spatial topological relationships, and generate entity-connected spatial structures; Based on the entity-connected spatial structure, structured roadway entities, working face entities, and shaft entities are constructed, and attribute parameters are bound to all entities to form a three-dimensional geometric mine structure model.

4. The mine safety decision-making simulation method based on digital twins as described in claim 3, characterized in that: The steps for performing attribute injection and discretization on the three-dimensional geometric mine structure model to generate a real-time digital twin of the mine are as follows. Perform attribute injection operations on the three-dimensional geometric mine structure model to generate a three-dimensional static digital twin skeleton of the mine; Discretize the three-dimensional static digital twin skeleton of the mine to generate a real-time digital twin of the mine.

5. The mine safety decision-making simulation method based on digital twins as described in claim 4, characterized in that: The steps for constructing a safety-process-resource knowledge graph based on a multi-source fusion dataset, defining mine element nodes and relationship edges, are as follows. The process specification data is structured to generate structured data of processes and resources; Extract and align fields from structured data of processes and resources to generate risk-triggered correlation data; Based on risk-triggered correlation data and the three-dimensional static digital twin skeleton of the mine, process nodes, resource nodes, risk nodes and spatial location nodes are defined, and the relationship edges between nodes are defined to construct a safety-process-resource knowledge graph.

6. The mine safety decision-making simulation method based on digital twins as described in claim 5, characterized in that: The steps for obtaining mine status fields from the safety-process-resource knowledge graph and the real-time mine digital twin, and constructing the mine safety decision-making state space and action space are as follows: Physical field state discretization is performed on the real-time mine digital twin to obtain the physical field state field; Map the mine safety-process-resource knowledge graph to obtain the process status field and resource status field; Multi-source monitoring data is mapped to a real-time mine digital twin by spatial location mapping. A multi-level state expression structure is established by combining physical field state fields, process state fields and resource state fields, and uniformly encoded into a mine safety decision state space. The knowledge rules in the screening process and resource structured data are examined for feasibility and a space for mine safety decision-making actions is generated.

7. The mine safety decision-making simulation method based on digital twins as described in claim 6, characterized in that: The steps for identifying decision-making disturbance events in multi-source monitoring data based on trigger thresholds and constructing a mine safety decision-making reward function by combining the mine safety decision-making state space and action space are as follows. Based on the trigger condition fields in the process specification data, identify the fields in the multi-source monitoring data that match the trigger threshold, and integrate them into a set of decision disturbance events; The corresponding benefit and loss fields are extracted from the mine safety decision state space, mine safety decision action space and process procedure data, and combined with the set of decision disturbance events to construct the mine safety decision reward function.

8. The mine safety decision-making simulation method based on digital twins as described in claim 7, characterized in that: The steps for injecting multi-source monitoring data into a real-time mine digital twin and selecting executable sequences for mine safety decisions are as follows. The monitoring fields from the multi-source monitoring data are inserted into the real-time mine digital twin to update the physical field and generate the current mine multi-physics field state. The current multiphysics state of the mine and the mine safety decision-making action space are matched to select executable candidate actions and combine them into an executable sequence for mine safety decisions.

9. The mine safety decision-making simulation method based on digital twins as described in claim 8, characterized in that: The steps for calculating the reward value of the executable sequence of safety decisions using the mine safety decision reward function and rearranging it into a recommended sequence for mine safety decisions are as follows: Based on the operational constraint field in the process specification data, the executable sequence of mine safety decision is checked for constraints, candidate actions that meet the constraints are identified, and combined into a candidate sequence of mine safety decision. The candidate sequence for mine safety decision is input into the mine safety decision reward function, the positive and negative values ​​are accumulated and calculated, and it is matched with the mine safety decision state space to output the reward value. The result is then sorted according to the reward value to generate the recommended sequence for mine safety decision.

10. The mine safety decision-making simulation method based on digital twins as described in claim 9, characterized in that: The steps for performing a reconstruction operation on the recommended sequence for mine safety decisions, obtaining the action effect deviation value, updating the parameters of the real-time mine digital twin, and generating a mine safety decision simulation report are as follows. The mine safety decision recommendation sequence is mapped to a real-time mine digital twin to capture changes in the physical field and record them as decision execution feedback data. Obtain the action effect deviation value of the decision execution feedback data, update the parameters of the real-time mine digital twin, and summarize them into a mine safety decision simulation report.

Images