Digital employee-based safety management and control system for dangerous sources in underground coal mine

Abstract

The present application relates to the field of information technology and intelligent management and control of safety production in underground coal mines, in particular to a dangerous source safety management and control system for underground coal mines based on digital employees, which arranges work by process employees and sends field operation instructions; pose employees establish a mobile coordinate system of the advancing support by fusing pose observation, construct a sparse graph to solve work node coordinates and generate a mismatch index from an information matrix for adaptive updating, and trigger closed-loop re-calibration when the threshold is exceeded; verification employees collect evidence to infer node attribution potential and output access control instructions, process employees arrange work node sequences and send field operation instructions; pose employees establish a mobile coordinate system of the advancing support, a sparse graph constraint set generates and updates work node coordinates, and closed-loop re-calibration is triggered when the pose fusion mismatch index exceeds the threshold; verification employees collect evidence information to calculate multi-modal attribution potential values, output access control instructions and dynamically adjust them according to the mismatch index.

Description

Technical Field

[0001] This invention relates to the field of information technology and intelligent control technology for safe production in underground coal mines, and more specifically, to a safety control system for hazardous sources in underground coal mines based on digital employees. Background Technology

[0002] Safety management of hazards in underground coal mines typically involves work plan preparation, on-site command execution, process verification and certification, and access control for key areas. In existing technologies, work plans are often issued to the site based on work procedures or job tasks, and risk identification and handling are achieved through personnel positioning, equipment positioning, video monitoring, and on-site inspections. For access control of key locations and processes, methods such as electronic fences, rule-based threshold judgments, or manual verification are often used to reduce risks such as accidentally entering hazardous areas, operating outside of designated procedures, or failing to obtain required certifications.

[0003] However, underground coal mining operations exhibit significant dynamic evolution characteristics: as the mining face advances continuously, hydraulic supports move in a queue-like pattern, and the spatial structure and access boundaries of the roadway may undergo considerable changes in local time periods due to factors such as support, transportation, and accumulation. Positioning and attitude perception typically rely on multi-source data, such as advance displacement and travel distance, equipment operating attitude, short-range positioning observations, roadway topology, and on-site images. In the underground environment, these data are easily affected by dust obstruction, multipath and obstruction of wireless signals, equipment vibration and shock, sensor zero-bias drift, and local obstruction and lock-off, leading to inconsistencies in attitude observations from different sources in time and space. This, in turn, causes problems such as coordinate drift, jumps, or crossing of roadway boundaries at work nodes.

[0004] While some solutions employ multi-source fusion or filtering estimation to improve stability, they often lack a comprehensive design in areas such as the sequential representation of nodes in the workflow, the synchronous evolution of reference coordinates as the work surface advances, the quantification of reliability after constraint fusion, and the linkage between closed-loop re-evidence and recalibration when fusion mismatch is significant. On the one hand, coordinate reliability lacks interpretable quantitative indicators, making it difficult to distinguish between short-term noise and structural mismatch. On the other hand, evidence collection information is often not tightly bound to work nodes and instruction windows, resulting in insufficient traceability of evidence collection and an inability to effectively feed back into coordinate correction. Furthermore, access control strategies are mostly static rules that fail to dynamically adjust with changes in pose consistency, which can easily lead to security risks such as allowing access when reliability decreases or excessive interception when reliability is normal, resulting in a loss of work efficiency.

[0005] Therefore, establishing a consistent, verifiable, and adaptive closed-loop mechanism between process instruction issuance, work node coordinate generation and updating, mismatch assessment, closed-loop recalibration, and multimodal verification and access control has become a key direction for further improving the safety management and control of existing hazardous sources. Summary of the Invention

[0006] In view of the shortcomings of existing technologies, the purpose of this invention is to provide a safety management and control system for hazardous sources in underground coal mines based on digital employees.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The underground coal mine hazard source safety management system based on digital employees includes: The process employee module is used to orchestrate tasks for digital employees and generate work node sequences, and issue on-site operation instructions to the corresponding work node sequences. The pose observation module is used to acquire pose observation data sets, establish a moving coordinate system for the propulsion support that evolves synchronously with the advance of the mining face, construct a sparse graph constraint set, generate coordinate transformation parameter sets and work node coordinates based on the sparse graph constraint set, construct an information matrix and obtain the uncertainty of the information matrix, generate a pose fusion mismatch index, adjust the weights of the sparse graph constraint set based on the pose fusion mismatch index to update the work node coordinates, and trigger closed-loop recalibration when the pose fusion mismatch index is greater than the recalibration threshold to correct the work node coordinates. The employee verification module is used to trigger the collection and uploading of evidence information based on the coordinates of the work node after the on-site operation instruction is detected in the process employee module; it calculates the multimodal attribution potential value and outputs access control instructions for implementing access control on the work node sequence.

[0008] In one embodiment, the process employee module receives access control instructions and controls the sequence of work nodes to advance, restrict, or pause. The access control instructions include allow instructions, restrict advance instructions, and pause instructions.

[0009] In one embodiment, the pose observation data set includes mining face advance displacement data, hydraulic support push stroke data, hydraulic support axial displacement data, mining equipment operating posture data, and positioning observation data.

[0010] In one embodiment, the sparse graph constraint set is constructed as follows: the observation constraints, roadway topology constraints, and evidence anchor point constraints formed by the pose observation data set are uniformly constructed into a sparse graph constraint set.

[0011] In one embodiment, the coordinate transformation parameter set and the coordinates of the working node are generated as follows: taking the moving coordinate system of the propulsion support as the reference, the rotation parameters and translation parameters are determined according to the minimum residual sum of squares criterion based on the sparse graph constraint set, thus forming the coordinate transformation parameter set; the coordinates of the working node in the moving coordinate system of the propulsion support are transformed by the coordinate transformation parameter set to obtain the coordinates of the working node.

[0012] In one embodiment, the pose fusion mismatch index is generated as follows: an information matrix is ​​constructed and the covariance matrix is ​​obtained by inverting the information matrix; the uncertainty of the information matrix is ​​obtained based on the trace of the covariance matrix; the propulsion Mahalanobis residual index and the lateral drift spectral energy index are calculated; the pose fusion mismatch index is obtained by fusing the information matrix uncertainty, the propulsion Mahalanobis residual index, and the lateral drift spectral energy index.

[0013] In one embodiment, closed-loop recalibration includes issuing a recalibration command to collect recalibration evidence information, and using the recalibration evidence information as evidence anchor point constraints to update the sparse graph constraint set to correct the coordinates of the operation node.

[0014] In one embodiment, the evidence information includes at least on-site images, spatial structure information, and operational scenario parameters.

[0015] In one embodiment, a candidate set of nodes is generated based on the coordinates of the job nodes and multimodal spatiotemporal attribution inference is performed to obtain the posterior distribution of node attribution, and the multimodal attribution potential value is calculated accordingly.

[0016] In one embodiment, access control commands are output based on the multimodal attribution potential value to implement access control for the sequence of work nodes. The generation rules of the access control commands are dynamically adjusted according to the magnitude of the pose fusion mismatch index, and when the access control command is a pause command, the pose employee module is triggered to perform closed-loop recalibration.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention obtains the coordinates of the work node by constructing a sparse graph constraint set and generating a coordinate transformation parameter set. At the same time, it constructs an information matrix and obtains the uncertainty of the information matrix, and then generates a pose fusion mismatch index and adjusts the weight of the sparse graph constraint set accordingly to update the coordinates of the work node. This maintains the continuity, interpretability and stability of the coordinates of the work node under dynamic conditions such as continuous advancement of the working face and movement of the support queue. When the pose fusion mismatch index is greater than the recalibration threshold, closed-loop recalibration is triggered to correct the coordinates of the work node. The coordinate update is gated by the mismatch index, which can promptly enter the correction process when the fusion inconsistency is significant. This avoids long-term cumulative drift from causing structural deviations in the coordinates of the work node, and makes the correction action and threshold judgment form a closed-loop linkage, improving the timeliness and certainty of coordinate reliability recovery under abnormal working conditions. After verification staff receive on-site operation instructions from process staff, evidence collection and uploading are triggered based on the work node coordinates. Furthermore, multimodal attribution potential values ​​are calculated and output as access control commands to implement access control over the work node sequence. This ensures a consistent mapping relationship between access control and work node coordinates, evidence collection information, and node sequence. The generation rules for access control commands are dynamically adjusted according to the pose fusion mismatch index. When the access control command is a pause command, it triggers the pose staff to perform closed-loop recalibration, thereby achieving a collaborative closed loop between verification decision-making and coordinate correction. This ensures access security at key nodes while also considering operational continuity and processing efficiency. Attached Figure Description

[0018] Figure 1 This is a structural block diagram of a safety management and control system for hazardous sources in underground coal mines based on digital employees. Figure 2 The flowchart for generating the pose fusion mismatch index, updating the coordinates of the task nodes, and entering closed-loop recalibration; Figure 3 Flowchart for closed-loop recalibration; Figure 4 This is a flowchart of an access control system designed to calculate multimodal attribution potential energy values ​​and output access control commands. Detailed Implementation

[0019] Reference Figures 1 to 4 A safety management system for hazardous sources in underground coal mines based on digital employees includes: The Process Employee module is used to orchestrate tasks for digital employees and generate work node sequences. It breaks down the hazard source safety management process in underground coal mines into executable and traceable task flows, solidifying these tasks as work node sequences. This transforms the original process organization, which relied on human experience, into a repeatable and auditable node-based process. It provides the subsequent Pose Employee module with the node targets that need to be located / aligned, and also provides the Verification Employee module with the node objects that need to be verified. On-site operation instructions are issued to the corresponding sites in the work node sequence; a correspondence is established between each work node in the work node sequence and the on-site execution action, and the on-site execution is reached through on-site operation instructions, thereby ensuring that the on-site execution is consistent with the system orchestration. The on-site operation instructions are not only the start signal for the on-site action, but also the timing basis for the subsequent employee verification module to trigger the collection and uploading of evidence information, ensuring that the closed-loop logic of instruction first—execution then verification is established.

[0020] In one specific implementation, when process employees schedule tasks for digital employees, they break down mining operations into several work nodes according to the granularity of safety control, and configure entry conditions, completion conditions, and on-site verification points for each work node, thereby generating a sequence of work nodes arranged in time and space order. Based on the current node in the work node sequence, process employees issue on-site operation instructions to the corresponding site. The on-site operation instructions at least include the execution object, execution action, execution sequence, and feedback requirements upon completion, which are used to map the on-site operation actions to the work node sequence one by one. For example, when the work node is the confirmation node for the hydraulic support to be pushed into place, the on-site operation instructions issued by the process employees clearly specify the requirements for pushing stroke verification, confirmation of the placement mark, and feedback evidence, thereby ensuring that the node can be consistently judged by subsequent access control. After issuing on-site operation instructions, process staff continuously receive access control instructions and control the sequence of work nodes based on these instructions: When a release instruction is received, the process staff advances the sequence of work nodes to the next work node and triggers the issuance of the next on-site operation instruction; when a restricted advancement instruction is received, the process staff allows the sequence of work nodes to advance within a limited range, such as only allowing entry into auxiliary nodes or waiting nodes, while maintaining access restrictions on critical nodes; when a pause instruction is received, the process staff immediately pauses the advancement of the sequence of work nodes at the current node and maintains the locked state of the on-site operation instructions until a release instruction or a restricted advancement instruction is received again, at which point the corresponding restriction is lifted. This forms a closed-loop access control execution link centered on release instructions, restricted advancement instructions, and pause instructions, ensuring that the advancement of the sequence of work nodes is strictly consistent with the safety verification results.

[0021] The pose observation module is used to acquire pose observation data sets, unifying multi-source observations related to spatial position, attitude, and propulsion into a single pose observation data set, forming the basic observation input for the field status. Through the continuous acquisition of pose observation data sets, data support can be provided for the calculation and updating of work node coordinates, ensuring that subsequent coordinate system establishment, constraint construction, and coordinate calculation do not rely on a single sensor or a single source of information. Establish a moving coordinate system for the propulsion support that evolves synchronously with the advancement of the mining face. In scenarios where the mining face continuously advances and the on-site reference system dynamically changes, construct a moving coordinate system for the propulsion support that is updated synchronously with the advancement process, ensuring that the spatial representation always closely matches the actual on-site movement. The positional representation of operational nodes can be described and calculated within the moving coordinate system of the propulsion support, reducing coordinate drift and inconsistencies caused by the advancement of the working face, and ensuring that the coordinates of subsequent operational nodes have an interpretable on-site reference. A sparse graph constraint set is constructed to organize the key relationships affecting positioning / alignment in the form of constraints and express them through a sparse graph structure. This ensures that the computation reflects the relationships between observations, topologies, and anchor points, while maintaining computational feasibility as the scale increases. The sparse graph constraint set provides an optimizable set of constraints for subsequent generation of coordinate transformation parameter sets and job node coordinates, making the solution process reliable and convergent. Based on the sparse graph constraint set, a coordinate transformation parameter set is generated, and the coordinates of the work nodes are generated. On the one hand, the coordinate transformation parameter set unifies data from different sources and reference systems into a common expression framework; on the other hand, the spatial location of the nodes is defined as the work node coordinates, realizing a calculable representation of where the nodes are on site. The work node coordinates are the spatial basis for the subsequent employee verification module to trigger the collection and uploading of evidence information, and also the coordinate basis for access control commands to perform node-level management. An information matrix is ​​constructed and its uncertainty is obtained. The information matrix is ​​used to express the constraint strength and information content in the current solution state, and the uncertainty of the information matrix is ​​used to quantify the reliability / stability of the coordinate results of the current operation node. The uncertainty of the information matrix provides a key component for the subsequent pose fusion mismatch index, enabling the system to obtain not only coordinate results but also a quantitative characterization of the reliability of the results, thereby supporting subsequent weight adjustment and recalibration triggering. A pose fusion mismatch index is generated. This index integrates the pose observation data set, sparse graph constraint set, and information matrix uncertainty to form a unified indicator reflecting whether the observations and solutions are consistent and whether the fusion is mismatched. As a quality gating signal, the pose fusion mismatch index guides the system to take timely corrective measures when coordinate reliability decreases, observation conflicts occur, or drift intensifies, preventing erroneous coordinates from being passed down to downstream verification and access control. The pose fusion mismatch index is used to adjust the weights of the sparse graph constraint set to update the coordinates of the work node. When the pose fusion mismatch index reflects inconsistencies or changes in reliability between different constraints, the influence of each constraint in the solution is dynamically changed by adjusting the weights of the sparse graph constraint set, thereby achieving adaptive coordinate updates. This step transforms the discovery of mismatches into concrete actions to correct the solution, enabling the coordinates of the work node to be continuously corrected according to changes in the field conditions, reducing the cumulative deviation in long-term operation. When the pose fusion mismatch index exceeds the recalibration threshold, closed-loop recalibration is triggered to correct the coordinates of the work nodes. This sets a clear quality baseline for the system—when the pose fusion mismatch index exceeds the recalibration threshold, it indicates that conventional weight adjustments are insufficient to restore reliability. In this case, closed-loop recalibration is used to perform a more forceful and thorough correction of the work node coordinates. This mechanism ensures that the system can return to a reliable state even under extreme disturbances, sensor anomalies, or significant changes in the on-site structure, avoiding the continuation of control processes on unreliable coordinates.

[0022] In one specific implementation, when acquiring the pose observation dataset and merging it to establish a moving coordinate system for the propulsion support that evolves synchronously with the advance of the mining face, the displacement data of the mining face is used as the main scale input along the propulsion direction, the travel data of the hydraulic support is used as the incremental constraint of the support relative to the propulsion, the displacement data of the hydraulic support is used as the basis for lateral drift compensation, and the operating posture data of the mining equipment is introduced synchronously for posture alignment and direction consistency correction. At the same time, the positioning observation data provides global reference anchoring. The above data are aligned according to the timestamp, and the propulsion displacement is used as the driving force for the evolution of the coordinate system. In each propulsion cycle, the origin position and axial direction of the moving coordinate system of the propulsion support are updated. The origin is recursively shifted with the propulsion displacement, the longitudinal axis points to the propulsion direction, the lateral axis is rotated and corrected according to the operating posture data of the mining equipment, and the vertical axis maintains orthogonal constraints consistent with the roadway spatial structure, thereby forming a coordinate expression that evolves synchronously with the on-site propulsion process and can be continuously updated. For example, when the displacement data of the mining face during a certain propulsion cycle indicates an increase in the propulsion distance, the travel data of the hydraulic support indicates that the support has been pushed into place, and the displacement data of the hydraulic support indicates that there is lateral movement, then while updating the origin of the propulsion support moving coordinate system, a lateral movement compensation amount is introduced into the lateral axis, and the axial direction is finely adjusted in combination with the operating posture data of the mining equipment. Then, the cumulative drift is constrained by the positioning observation data, so that all subsequent pose expressions related to propulsion remain consistent and traceable under the same coordinate reference that evolves with propulsion.

[0023] In one specific implementation, when constructing the information matrix and obtaining its uncertainty, under the work node coordinate solution state determined by the sparse graph constraint set, the residuals of various constraints are linearized and summarized with respect to the Jacobian of the state variables to form a weighted normal equation consistent with the constraint weights. The coefficient matrix of the weighted normal equation is used as the information matrix. The covariance matrix is ​​obtained by inverting the information matrix, and the uncertainty of the information matrix is ​​obtained based on the trace of the covariance matrix. This uncertainty is used to quantify the reliability of the current work node coordinates under the overall constraints. To allow the sparse graph constraint set to be directly solved using the minimum residual sum of squares criterion, let the state variable to be solved be... It includes at least the coordinates of the working node in the moving coordinate system of the propulsion support, as well as the rotation and translation parameters in the coordinate transformation parameter set; for each type of constraint, a residual term is constructed and weights are introduced to obtain the weighted objective function:

[0024] Among them, the residuals of the observation constraints can be formed by the observations from the pose observation data set. With the current Predicted observations The difference is obtained, that is The preferred method for using topological constraints on roadways is to prevent work node coordinates from crossing roadway boundaries. Let the signed distance from the boundary of the reachable area of ​​the roadway to the work node coordinates be... (If it is within the reachable region and is non-negative), then it can be taken as... The residual is used to penalize out-of-bounds errors; the anchor point constraint for evidence collection preferably uses the anchor point coordinates obtained from the parsing of evidence collection information. As a stable reference, take ,in Indicates the relationship with the first The coordinate components of the operation node corresponding to each anchor point. Furthermore, the Jacobian of the residuals with respect to the state variables is used to form a weighted normal equation consistent with the constraint weights, and the coefficient matrix of this weighted normal equation is used as the information matrix; In one alternative implementation, the minimum residual sum of squares criterion is solved iteratively using the Gauss-Newton method or the Levenberg-Marquardt method: in the... In each iteration, a weighted normal equation is constructed based on the Jacobian of each constraint residual with respect to the state variable \(x\), and the increment is solved. ,renew ;when The iteration stops when the value falls below a preset threshold, the objective function decreases below a preset threshold, or the number of iterations reaches the upper limit. To improve the solution efficiency under large-scale nodes and constraints, sparse matrix factorization or the conjugate gradient method is preferred for solving the weighted normal equations, and a robust kernel function is introduced for outlier residuals in each iteration to suppress the impact of outlier observations on the solution. When generating the pose fusion mismatch index, based on the uncertainty of the information matrix, the propulsion Mahalanobis residual index and the lateral drift spectral energy index are calculated. The propulsion Mahalanobis residual index is obtained by combining the Mahalanobis distance obtained by combining the residual between the actual propulsion measurement formed by the pose observation data set and the propulsion amount predicted by the coordinate transformation parameter set with the covariance matrix. The lateral drift spectral energy index is obtained by constructing the lateral drift residual sequence by the lateral component sequence of the working node coordinates on the lateral axis of the propulsion support moving coordinate system and calculating the spectral energy ratio. The pose fusion mismatch index is obtained by fusing the uncertainty of the information matrix, the propulsion Mahalanobis residual index, and the lateral drift spectral energy index, thereby simultaneously characterizing the global uncertainty, propulsion direction consistency, and lateral drift anomaly. Let the first The horizontal component of the coordinates of the working nodes within each propulsion cycle on the horizontal axis of the propulsion support's moving coordinate system is: Preferably, the horizontal component sequence is constructed using the propulsion period as the discrete sampling period. And using a sliding window length of Mean filtering yields Based on this, an aberration residual sequence is constructed. For sequences within the window Performing Discrete Fourier Transform yields And the spectral energy index of lateral drift is defined by energy percentage: ;

[0025] Among them, the preset frequency band The preferred frequency setting is the one corresponding to the shifting cycle characterized by the hydraulic support shifting stroke data: Let the shifting cycle be... The main frequency is Then it is acceptable. The corresponding discrete frequency index set; when packet loss or uneven sampling occurs, interpolation or zero-padding is preferred to maintain the window length. and to Windowing is applied to reduce spectral leakage; Let the length of the sliding window be... The sampling period is Then the discrete frequency corresponding to the discrete Fourier transform is When the preset clock speed is... And preset frequency band When, the corresponding discrete frequency index set is taken as Based on this, when calculating the spectral energy index of lateral drift, Sum the spectral energy; Let the uncertainty of the information matrix be... To advance the Markov residual index as The spectral energy index of lateral drift is The pose fusion mismatch index is Historical samples were preferably collected under normal operating conditions to obtain... and And obtained by standardization and normalization. , , ,in To prevent constants with zero denominators, and to ensure the fusion result remains monotonic for each component and suppresses the influence of extrema, a monotonic compression function is preferred. ,get , , The pose fusion mismatch index is defined as follows: ,in and When offline calibration data is lacking, it is preferable to use [the appropriate data]. When calibrated samples exist, it is preferable to adjust according to the contribution of the three factors to the mismatch event. ; When adjusting the weights of the sparse graph constraint set based on the pose fusion mismatch index to update the coordinates of the work node, the pose fusion mismatch index is mapped as a weight adjustment factor. The weights of the observation constraints, roadway topology constraints and evidence anchor point constraints in the sparse graph constraint set are adjusted differently. The solution is then executed again with the goal of minimizing the sum of squared residuals to update the coordinates of the work node. In one specific implementation, let the observation constraints, tunnel topology constraints, and evidence anchor point constraints in the sparse graph constraint set be in the th... The weights at the next update are respectively , and And the pose is fused with the mismatch index Mapped to weight adjustment factor Preferred ,in This is a statistical reference value for the pose fusion mismatch index under normal operating conditions. For weighted differential updates, multiplicative updates are preferred, with upper and lower bound constraints set. , , ,in To update the step size, These are the lower and upper bounds of the weights, respectively; the updated selection is then normalized. To suppress overall weight drift, the solution is executed again with the goal of minimizing the sum of squared residuals to update the job node coordinates; For example, when the Mahalanobis residual index increases significantly and the uncertainty of the information matrix also increases, it indicates that there is a significant inconsistency between the observation of the advance direction and the current solution and the overall credibility decreases. At this time, the weights of the evidence anchor point constraint and the roadway topology constraint are increased, and the weights of the corresponding observation constraint are reduced, so that the coordinates of the work node are more constrained by stable references and reachable geometry and return to the reasonable area. When the energy ratio of the lateral drift spectrum index increases abnormally in the preset frequency band, it indicates that the lateral drift has periodic or sudden anomalies. In this case, the weights of the observation constraint related to the lateral component are suppressed and the topology boundary constraint is strengthened to suppress the abnormal swing of the coordinates of the work node in the lateral direction, thereby realizing the adaptive update and robust convergence of the coordinates of the work node as the field conditions change.

[0026] In one specific implementation, when the pose fusion mismatch index is greater than the recalibration threshold, closed-loop recalibration is triggered to correct the coordinates of the work node. A recalibration command is sent to the site corresponding to the work node sequence, guiding the site to collect and upload recalibration evidence information around the neighborhood of the current work node coordinates. The evidence information includes at least site images, spatial structure information, and work scene parameters. The site images preferably cover fixed reference objects and support number identifiers around the work node. The spatial structure information is used to characterize stable geometric features such as the roadway outline, the relative relationship of support components, or the passage boundary. The work scene parameters are used to characterize the scene context that is strongly related to the current evidence collection time, such as the advance stage of the mining face, the moving state of the hydraulic support, and the operating posture state of the mining equipment. The recalibrated evidence information is parsed into evidence anchor point constraints, and the evidence anchor point constraints are incorporated into the sparse graph constraint set to complete the update, so that the updated sparse graph constraint set simultaneously possesses observation consistency constraints, roadway topology consistency constraints, and anchor point stable reference constraints. When parsing recalibrated evidence information into evidence anchor point constraints, it is preferable to follow the process of anchor point identification—anchor point location—uncertainty estimation—constraint writing: identify fixed reference objects, support number markings, or other uniquely identifiable visual anchor points from the field images to obtain anchor point identifiers. With image measurement ; Combine the roadway outline, boundary lines, or geometric features of support components extracted from spatial structural information to mark the anchor points. Match the topological nodes or boundary segments corresponding to the roadway topological constraints, and obtain the anchor point coordinates in the moving coordinate system of the propulsion support. The anchor point coordinates can be obtained from image measurements. The relative pose obtained by back projection is fused with the global reference anchoring provided by the positioning observation data. The effectiveness of the anchor points is screened using operational scenario parameters to ensure that the anchor points are consistent with the current advancement stage, the hydraulic support movement status, and the operating attitude of the mining equipment. Then, an uncertainty description is given for the anchor point coordinates, preferably using the covariance matrix. Characterization, and from this, the anchor weight matrix is ​​obtained. Add anchor point constraints to the sparse graph constraint set: when the anchor point is a point reference, the preferred anchor point constraint residual is selected. When the anchor point is a boundary reference, it is preferable to set the residual as the shortest distance from the node to the boundary. and with As corresponding weights, they participate in the solution of the minimum residual sum of squares criterion, thereby enabling the recalibration and certification information to stably feed back into the coordinate correction of the operation node in a computable manner; Based on the updated sparse graph constraint set, the rotation and translation parameters are redefined and the work node coordinates are updated using the minimum residual sum of squares criterion, thereby achieving closed-loop correction of the work node coordinates. For example, when a fixed marker at a bend is clearly visible in the field image and the spatial structure information indicates that the width and boundary line of the roadway at that location are stable, but the original work node coordinates show a shift across the boundary, incorporating the evidence anchor point constraint into the sparse graph constraint set can force the work node coordinates to be pulled back to the reachable area consistent with the roadway topology, thereby completing the correction. The basis for setting the recalibration threshold can be determined by combining residual gating and statistical consistency discrimination methods in existing technologies. For example, a baseline distribution can be established using historical samples of pose fusion mismatch index under normal working conditions, and the mean plus multiple standard deviations can be used as the trigger boundary to control the probability of false triggering. Alternatively, the trigger threshold can be determined based on the chi-square distribution confidence limit corresponding to the advance Mahalanobis residual index. At the same time, joint gating can be performed by combining the information matrix uncertainty and the abnormal proportion requirements of the crosstalk spectral energy index, thereby ensuring sensitivity to significant mismatch while suppressing frequent recalibration caused by short-term noise. The recalibration threshold setting and joint gating optimization are performed as follows: Under normal operating conditions, continuous propulsion cycle samples are collected, and the pose fusion mismatch index for each cycle is calculated. Information matrix uncertainty Promote the development of Markov residual index With spectral energy index of lateral drift The baseline statistic is obtained. The preferred recalibration threshold is... ,in Determined by the expected false trigger probability; advancing the optimal selection of the Mahalanobis residual index threshold. ,in To advance the dimension of the measured residual vector and take the degrees of freedom , The significance level is set at [value missing]. The optimal threshold for the cross-drift spectral energy index is [value missing]. and in the sliding window length is The proportion of anomalies calculated in the most recent period ,in The indicator function; joint gating preferably satisfies and Time-triggered closed-loop recalibration, where , This serves as the lower limit for the proportion of abnormalities, thus balancing sensitivity to significant mismatches with suppression of short-term noise.

[0027] The employee verification module, upon receiving on-site operation instructions from the process employee module, triggers the collection and uploading of evidence information based on the coordinates of the work nodes. It binds the verification action to both the temporal and spatial sequence of on-site execution—ensuring verification follows the actual execution rhythm after receiving on-site operation instructions from the process employee module, and ensuring evidence collection occurs at or within the spatial location of the corresponding node based on the work node coordinates. The collection and uploading of evidence information provides the system with usable evidence input for verification and provides a multi-source data foundation for subsequent calculations of multimodal attribution potential values. The multimodal attribution potential value is calculated by performing a multimodal association assessment between the evidence information and the target node described by the coordinates of the work node, and quantifying whether the evidence information matches the target node and how much they match as a multimodal attribution potential value. The introduction of the multimodal attribution potential value transforms the verification result from a qualitative judgment into a calculable indicator, which facilitates consistent and interpretable decisions on allowing, restricting, or suspending access control commands. The system outputs access control commands for implementing access control over the sequence of work nodes. It transforms verification results into control actions for process advancement, enabling the system to manage access control over the sequence of work nodes. These access control commands are feedback outputs from the employee verification module to the process employee module, ensuring that the sequence of work nodes can proceed in an orderly manner when verification requirements are met, and be restricted or suspended when verification is insufficient or risk increases. This forms a closed-loop safety management chain of orchestration, execution, verification, and control.

[0028] In one specific implementation, after the process employee issues the on-site operation instruction, the verification employee triggers the collection and uploading of evidence information based on the coordinates of the current work node. The evidence information includes at least on-site images, spatial structure information and work scene parameters. The evidence information is then time-bound and coordinate-bound with the issued on-site operation instruction to ensure that the evidence collection occurs within the spatial neighborhood and instruction execution window of the corresponding work node. Verification staff generate a candidate set of nodes based on the coordinates of the work nodes and perform multimodal spatiotemporal attribution inference. The candidate set of nodes is preferably formed by setting adaptive search ranges in the advancing direction and the lateral direction, with the current work node coordinates as the center. The accessibility constraints of the roadway topology and the verifiability constraints of the evidence anchor points are superimposed to eliminate unreasonable candidates. The target identifiers and scene texture features extracted from the field images, the roadway contours and boundary geometric features extracted from the spatial structure information, and the advancing stage and equipment attitude state features extracted from the work scene parameters are jointly modeled to construct the observation likelihood related to the candidate nodes. Combined with the prior spatiotemporal continuity of the candidate nodes, the posterior distribution of node attribution is obtained, and the multimodal attribution potential value is calculated accordingly. The multimodal attribution potential value is preferably monotonically correlated with the confidence concentration of the posterior distribution of node attribution, so that the more obvious the attribution, the lower the potential energy, and the more dispersed the attribution, the higher the potential energy. Let the candidate set of nodes be The multimodal observations corresponding to the evidence information are For each candidate node Construction observation likelihood Preferred selection:

[0029] in These represent the difference metrics calculated from target identifiers and scene texture features, tunnel contours and boundary geometric features, and advancement phases and equipment attitude state features, respectively. This is the scale parameter. It is combined with the prior spatiotemporal continuity of the candidate nodes. ;

[0030] Obtain the posterior distribution of node affiliation ;

[0031] The optimal definition of the multimodal attribution potential value is the normalized entropy. ;

[0032] Therefore, the more concentrated the posterior distribution of node affiliation, the better. The lower and more dispersed The higher; Verification staff outputs access control commands for implementing access control on the sequence of work nodes based on the multimodal attribution potential energy value. The generation rules of the access control commands are dynamically adjusted according to the size of the pose fusion mismatch index. Specifically, when the pose fusion mismatch index is small, a stricter potential energy threshold is used to improve efficiency and priority is given to outputting the release command. When the pose fusion mismatch index is at a medium level, the potential energy threshold is relaxed and a restriction advance command is introduced to allow the sequence of work nodes to enter the transition node waiting for verification or auxiliary verification. When the pose fusion mismatch index is large and the multimodal attribution potential energy value exceeds the pause threshold, a pause command is output and the pose staff is simultaneously triggered to perform closed-loop recalibration to restore the reliability of the work node coordinates. Let the pose fusion mismatch index be... The multimodal attribution potential value is And set the pose fusion mismatch index segmentation threshold. Let the potential energy threshold used for determining access control commands be... and And satisfy The potential energy threshold is used to determine whether propulsion needs to be restricted or paused, and the higher the potential energy threshold, the more stringent the triggering condition. Segmented monotonic adjustment is preferred: when... Time to take , ;when Time to take , ;when Time to take , and satisfy and The preferred rule for determining access control commands is: if... If so, then output a release command; if If so, then output a command to restrict propulsion; if If the output is not true, a pause command will be output and the pose worker module will be triggered to perform closed-loop recalibration to restore the reliability of the work node coordinates. For example, at the node where the hydraulic support is moved into place, if the on-site image shows that the support number and the spatial structure information indicate that the location of the roadway boundary is highly consistent with the coordinates of the current work node, and the node's posterior distribution is highly concentrated near the current node with a low multimodal attribution potential value, then a release command is output to advance to the next node; if the on-site image is obscured by dust, causing the posterior distribution to be dispersed between two adjacent nodes and the pose fusion mismatch index to increase synchronously, then a restriction command is output to require supplementary evidence collection; if the posterior distribution deviates significantly from the current node and the multimodal attribution potential value continues to increase, then a pause command is output and a closed-loop recalibration is triggered to correct the work node coordinates before access control is restored.

[0033] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A safety management and control system for hazardous sources in underground coal mines based on digital employees, characterized in that: include: The process employee module is used to orchestrate tasks for digital employees and generate work node sequences, and issue on-site operation instructions to the corresponding work node sequences. The pose and employee module is used to acquire pose observation data sets and establish a moving coordinate system for the propulsion support that evolves synchronously with the advance of the mining face. Construct a sparse graph constraint set, generate a coordinate transformation parameter set based on the sparse graph constraint set, and generate the coordinates of the work node; construct an information matrix and obtain the uncertainty of the information matrix; generate a pose fusion mismatch index; adjust the weights of the sparse graph constraint set based on the pose fusion mismatch index to update the coordinates of the work node; when the pose fusion mismatch index is greater than the recalibration threshold, trigger closed-loop recalibration to correct the coordinates of the work node. The employee verification module is used to trigger the collection and uploading of evidence information based on the coordinates of the work node after an on-site operation instruction is found in the process employee module. Calculate the multimodal attribution potential value and output access control instructions for implementing access control on the sequence of work nodes.

2. The underground coal mine hazard source safety management and control system based on digital employees as described in claim 1, characterized in that, The process employee module receives access control instructions and controls the sequence of work nodes to advance, restrict, or pause. Access control instructions include grant instructions, restrict advance instructions, and pause instructions.

3. The underground coal mine hazard source safety management and control system based on digital employees as described in claim 1, characterized in that, The pose observation data set includes mining face advance displacement data, hydraulic support push stroke data, hydraulic support lateral displacement data, mining equipment operating posture data, and positioning observation data.

4. The underground coal mine hazard source safety management and control system based on digital employees as described in claim 1, characterized in that, The sparse graph constraint set is constructed as follows: the observation constraints, roadway topology constraints, and evidence anchor point constraints formed by the pose observation data set are uniformly constructed into a sparse graph constraint set.

5. The underground coal mine hazard source safety management and control system based on digital employees as described in claim 1, characterized in that, The coordinate transformation parameter set and the coordinates of the working node are generated as follows: Taking the moving coordinate system of the propulsion support as the reference, the rotation parameters and translation parameters are determined according to the minimum residual sum of squares criterion based on the sparse graph constraint set, thus forming the coordinate transformation parameter set; the coordinates of the working node in the moving coordinate system of the propulsion support are transformed by the coordinate transformation parameter set to obtain the coordinates of the working node.

6. The underground coal mine hazard source safety management and control system based on digital employees according to any one of claims 1-5, characterized in that, The pose fusion mismatch index is generated as follows: an information matrix is ​​constructed and the covariance matrix is ​​obtained by inverting the information matrix; the uncertainty of the information matrix is ​​obtained based on the trace of the covariance matrix; the propulsion Mahalanobis residual index and the lateral drift spectrum energy index are calculated. The pose fusion mismatch index is obtained by fusing information matrix uncertainty, advancing Mahalanobis residual index, and lateral drift spectrum energy index.

7. The underground coal mine hazard source safety management and control system based on digital employees according to claim 1, characterized in that, Closed-loop recalibration includes issuing recalibration commands, collecting recalibration certification information, and using the recalibration certification information as certification anchor point constraints to update the sparse graph constraint set to correct the coordinates of the operation nodes.

8. The underground coal mine hazard source safety management and control system based on digital employees according to claim 7, characterized in that, Evidence collection information should include at least on-site images, spatial structure information, and operational scenario parameters.

9. The underground coal mine hazard source safety management and control system based on digital employees according to claim 1, characterized in that, A candidate set of nodes is generated based on the coordinates of the task nodes, and multimodal spatiotemporal attribution inference is performed to obtain the posterior distribution of node attribution, and the multimodal attribution potential value is calculated accordingly.

10. The underground coal mine hazard source safety management and control system based on digital employees according to claim 9, characterized in that, Based on the multimodal attribution potential value, access control commands are output for access control of the work node sequence. The generation rules of the access control commands are dynamically adjusted according to the size of the pose fusion mismatch index. When the access control command is a pause command, the pose employee module is triggered to perform closed-loop recalibration.

Images