A mine team meeting whole-process standardized intelligent management system

Abstract

This invention discloses a standardized intelligent management system for the entire process of mine team meetings, specifically relating to the field of mine team meeting management. It addresses the difficulty in integrating pre-meeting status identification, node execution constraints, and process compliance judgment in mine team meetings. The system uses a hardware integration module to acquire multi-dimensional physiological characteristics and biometric identifiers of participants and extract abnormal state representation sets. A meeting process management module, combined with the current team's job attributes, generates a navigation framework containing node association attributes and a standard preset timestamp sequence. A database management module forms a corresponding safety warning data map around the navigation framework. The meeting management module then completes status interaction confirmation, node change log encapsulation, and process compliance judgment based on the navigation framework and safety warning data map, outputting the process processing results: freezing the status transition of the current meeting node or solidifying it in a data traceability pool.

Description

Technical Field

[0001] This invention relates to the field of mine team meeting management, and more specifically, to a standardized intelligent management system for the entire process of mine team meetings. Background Technology

[0002] Mining team meetings are a crucial part of pre-shift safety management and production organization in coal mining enterprises. Meeting content typically involves personnel attendance confirmation, safety oaths, risk pre-control, task briefings, and on-site confirmation. The meeting execution process is related to both the personnel's physiological state and whether meeting milestones are completed in the predetermined order and within the designated timeframe. Current team meeting management methods primarily rely on routine training broadcasts, equipment linkage, or meeting record keeping. While these methods can handle document presentation, audio and video capture, and basic interaction, they lack a unified processing chain that covers pre-meeting identification, in-meeting control, and post-meeting follow-up in high-risk mining operation scenarios, such as abnormal pre-meeting conditions of participants, rushed execution of meeting milestones, and perfunctory interaction during the interactive phase.

[0003] Represented by the existing invention patent document CN117156097A, "Intelligent Processing Method and System for Meeting Audio Data Based on IoT Sensing," related solutions have been able to conduct low-level sensing and audio processing around meeting scenarios. However, their focus remains on meeting audio data processing and device-side linkage. They lack corresponding processing structures for state control logic triggered by multi-dimensional physiological characteristics in mine team meetings, node execution relationships constrained by navigation frameworks, and process compliance judgments based on node transition logs. Therefore, in the context of mine team meetings, a standardized intelligent management solution covering the entire process, including abnormal state representation sets, navigation frameworks, safety warning data maps, and process compliance detection features, is still needed.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a standardized intelligent management system for the entire process of mine team meetings. This system uses a hardware integration module to acquire multidimensional physiological characteristics and biometric identifiers of participants and extract abnormal state representation sets. A meeting process management module, combined with the current team's job attributes, generates a navigation framework containing node association attributes and a standard preset timestamp sequence. A database management module forms a corresponding safety warning data map around the navigation framework. The meeting management module then completes status interaction confirmation, node change log encapsulation, and process compliance judgment based on the navigation framework and safety warning data map, outputting the process processing results of freezing the current meeting node's status transition or solidifying it in a data traceability pool, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A standardized intelligent management system for the entire process of mine team meetings includes: Hardware integration module, meeting workflow management module, database management module, and meeting management module; The hardware integration module is used to acquire the multidimensional physiological characteristics and biometric identifiers of the participants, organize the multidimensional physiological characteristics according to the biometric identifiers, use the support vector data description model to perform spatial mapping and fit the steady-state envelope surface, and extract the abnormal state representation set that deviates from the steady-state envelope surface. The meeting process management module is used to receive the abnormal status representation set, and configure the status control logic of each core meeting node in combination with the job operation attributes of the current shift, construct the time sequence diagram, and generate a navigation framework containing node association attributes and standard preset timestamp sequences. The database management module is used to parse the association attributes of each node in the navigation framework, construct a graph neural network composed of business nodes, aggregate and iterate the features of adjacent business nodes, and generate a security warning data map that is back to the navigation framework. The meeting management module is used to drive the audio-visual equipment to output safety warning data maps according to the navigation framework and complete the status interaction confirmation. It performs spatiotemporal alignment and encapsulation of the on-site audio and video streams and node transition logs, extracts process compliance detection features for node transition logs to determine process compliance, and freezes the status flow of the current meeting node when it crosses the preset blocking threshold.

[0007] Furthermore, the hardware integration module first generates a biometric identifier through the face recognition unit or fingerprint recognition unit, and then continuously reads the heart rate, blood oxygen saturation and body surface temperature of the participants under the same sampling clock. The detection values ​​of each dimension obtained at the same sampling time are combined into a multi-dimensional physiological sign vector in a fixed order, and the sampling points corresponding to the missing dimensions are removed before forming the multi-dimensional physiological sign vector sequence.

[0008] Furthermore, the hardware integration module retrieves historical normal attendee samples corresponding to biometric identifiers to form a steady-state sample sequence. It uses a Gaussian kernel function to complete spatial mapping and fit a steady-state envelope surface, retaining support vectors, corresponding Lagrange multipliers, envelope center, and envelope radius. It also merges multiple consecutive abnormal sampling points that deviate from the steady-state envelope surface into corresponding abnormal state representations.

[0009] Furthermore, the meeting process management module reads the job operation attributes from the current shift's scheduling records, and filters out the core meeting node set corresponding to the current shift based on the mine safety regulations node template. It then arranges each core meeting node in a fixed order to form a time sequence diagram, and generates a standard preset timestamp sequence based on the shortest effective dwell time of each core meeting node.

[0010] Furthermore, the state control logic includes at least four sequential states: pending entry, resident execution, state interaction confirmation, and allowed flow. When the biometric identifier in the abnormal state representation set corresponds to a core meeting node, the meeting process management module switches the state interaction confirmation method of the core meeting node to a single-person confirmation path and uses the completion of the single-person confirmation path as a condition for allowing flow.

[0011] Furthermore, the database management module organizes the successor-successor relationships, state control logic, and corresponding time points in the standard preset timestamp sequence of each graph node in the navigation framework into node association attribute vectors in a fixed order. Business nodes are constructed using node association attribute vectors, and adjacency matrices with self-connections are generated based on the successor-successor relationships in the navigation framework.

[0012] Furthermore, the database management module uses the navigation framework of historical completed meetings and the corresponding manually confirmed data retrieval results as training samples to update the trainable mapping matrix of the graph neural network offline. During the online execution phase, it calls the trained trainable mapping matrix to aggregate and iterate the business nodes. Subsequently, the security warning data in the data candidate set are uniformly encoded and sorted according to the matching score to form a security warning data map.

[0013] Furthermore, the meeting management module sequentially drives the audiovisual equipment to output corresponding safety warning information according to the order of the graph nodes in the navigation framework, and promotes the resident execution, status interaction confirmation and permission flow according to the status control logic of the corresponding graph nodes; when switching between states, the graph node identifier, status name and status switching time are recorded to form a node transition log, and the on-site audio and video streams are segmented and encapsulated according to the adjacent status switching time.

[0014] Furthermore, the meeting management module extracts the transition time of each graph node entering the allowed transition state from the node transition log, forms an actual execution timestamp sequence according to the graph node order, and reads the standard preset timestamp sequence corresponding to each graph node in the navigation framework for comparison; at the same time, a discrete time window is opened in the status interaction confirmation stage to statistically analyze the underlying concurrent operation events generated by the interactive terminal according to the operation type.

[0015] Furthermore, the meeting management module determines the node transition time difference based on the comparison between the actual execution timestamp sequence and the standard preset timestamp sequence, determines the instruction concurrency divergence based on the statistical results of underlying concurrent operation events within the discrete time window, and maps the node transition time difference and instruction concurrency divergence to the benchmark covariance matrix space generated by the historical compliant execution log for judgment. When the preset blocking limit is exceeded, a violation warning is output and the state transition of the current meeting node is frozen.

[0016] Technical effects and advantages of the present invention: a standardized intelligent management system for the entire process of mine team meetings This invention forwards the multidimensional physiological characteristics identification results of participants to the group meeting entry point and forms an abnormal state representation set. The abnormal state representation set is then written into the state control logic of the core meeting node, so that the meeting execution object is expanded from simple attendance and sign-in to a controlled object with pre-meeting state judgment results, making the basis for starting the meeting process clearer.

[0017] This invention uses the navigation framework generated by the meeting process management module as the main thread, and organizes the core meeting nodes, status control logic, standard preset timestamp sequences, and security warning data graphs generated by the database management module into a unified call chain, so that the node execution path, data push relationship and status interaction confirmation are consistent, reducing the disconnect between data playback, node flow and log recording.

[0018] This invention focuses on compliance detection features in the node transition log extraction process within the meeting management module. It combines the spatiotemporal alignment and encapsulation results of the on-site audio and video streams to determine process compliance. When a preset blocking boundary is crossed, the state transition of the current meeting node is frozen, enabling the meeting process to not only leave a record but also generate real-time judgment and traceability results for specific nodes and specific state intervals. Attached Figure Description

[0019] Figure 1 This is a diagram illustrating the overall architecture of a standardized intelligent management system for the entire process of mine team meetings, as described in this invention. Figure 2 This invention provides a flowchart of the pre-meeting status recognition process for a hardware integration module in a standardized intelligent management system for the entire process of mine team meetings. Figure 3 This is a schematic diagram of the navigation framework generated by the meeting process management module in a standardized intelligent management system for the entire process of mine team meetings according to the present invention. Figure 4 This is a schematic diagram illustrating the generation of safety warning data maps in the database management module of a standardized intelligent management system for the entire process of mine team meetings according to the present invention. Figure 5 This invention presents a flowchart of the meeting execution and process compliance determination process in the meeting management module of a standardized intelligent management system for the entire process of mine team meetings. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 - Figure 5 This invention provides a standardized intelligent management system for the entire process of mine team meetings, including: a hardware integration module, a meeting process management module, a database management module, and a meeting management module; The hardware integration module is used to acquire the multidimensional physiological characteristics and biometric identifiers of the participants, organize the multidimensional physiological characteristics according to the biometric identifiers, use the support vector data description model to perform spatial mapping and fit the steady-state envelope surface, and extract the abnormal state representation set that deviates from the steady-state envelope surface. The meeting process management module is used to receive the abnormal status representation set, and configure the status control logic of each core meeting node in combination with the job operation attributes of the current shift, construct the time sequence diagram, and generate a navigation framework containing node association attributes and standard preset timestamp sequences. The database management module is used to parse the association attributes of each node in the navigation framework, construct a graph neural network composed of business nodes, aggregate and iterate the features of adjacent business nodes, and generate a security warning data map that is back to the navigation framework. The meeting management module is used to drive the audio-visual equipment to output safety warning data maps according to the navigation framework and complete the status interaction confirmation. It performs spatiotemporal alignment and encapsulation of the on-site audio and video streams and node transition logs, extracts process compliance detection features for node transition logs to determine process compliance, and freezes the status flow of the current meeting node when it crosses the preset blocking threshold.

[0022] The present invention transforms the conventional execution method of mine team meetings, which mainly involves data playback and confirmation, into a closed-loop processing mechanism that covers pre-meeting status identification, in-meeting path control, and process compliance judgment. Specifically, it first models and identifies abnormal state characteristics based on the multidimensional physiological characteristics of the participants, and extracts abnormal state characterization results. Then, combined with the job attributes of the current team, the abnormal state characterization results are written into the state control logic of the core meeting nodes, forming a meeting execution path with sequential and interactive constraints. During the meeting execution, it further analyzes the macro-level progress pace and micro-level operational behavior based on the node transition log, judges whether there is compressed execution, malicious fast-forwarding, or disordered operation in the meeting process, and freezes the state transition of the current meeting node when the judgment result exceeds the preset boundary.

[0023] The problem addressed by this invention is that existing meeting management methods typically focus on document playback, device linkage, and process logging, making it difficult to identify whether participants are in abnormal states, or to determine whether key security steps are skipped, confirmations are perfunctory, or formalities are being circumvented during meeting execution. Compared to common approaches, this invention directly incorporates the results of participant abnormal state identification into the meeting node control chain, and integrates the timing and operational behavior of the meeting execution phases into a unified judgment framework. This transforms the meeting processing process from a simple recording workflow into a standardized management process with process constraints and immediate intervention capabilities.

[0024] Specific implementation methods of the hardware integration module: Before a mine shift meeting enters formal node control, the primary focus of meeting processing is on the physical status of attendees and their corresponding identities. Sign-in results or routine attendance information alone are insufficient to determine whether individuals are within a stable state suitable for entering the pre-shift meeting process. Therefore, the hardware integration module needs to establish a common-source data collection relationship based on the multidimensional physiological characteristics and biometric identifiers of attendees. Under a unified object boundary, it must complete the fitting and deviation identification of the stable-state envelope, shifting the meeting processing starting point from simple attendance confirmation to a structured processing stage oriented towards pre-meeting status assessment.

[0025] The following sub-steps S101 to S103 are executed by the hardware integration module.

[0026] S101 Simultaneous acquisition of multidimensional physiological signs and biometric identifiers.

[0027] In the initial processing stage of the hardware integration module, the hardware integration module in the standardized intelligent management system for the entire process of mine team meetings first enters the pre-meeting data collection state, using the participants of the current team entering the meeting check-in area as the data collection targets. Only after the identity of the same participant is confirmed will the hardware integration module activate the corresponding multi-dimensional physiological characteristic sampling link, ensuring a one-to-one correspondence between identity information and vital sign information from the source of data collection.

[0028] In practice, the hardware integration module first generates a biometric identifier using a facial recognition or fingerprint recognition unit. This biometric identifier uniquely identifies the current attendee. Subsequently, the hardware integration module continuously reads continuously quantifiable physiological parameters of the attendee, such as heart rate, blood oxygen saturation, and body surface temperature, at the same sampling clock. It then combines the values ​​obtained at the same sampling time into a multidimensional physiological characteristic vector in a fixed order. These multidimensional physiological characteristic vectors from multiple consecutive sampling times are then arranged chronologically to form a multidimensional physiological characteristic vector sequence. To ensure a unified input boundary for subsequent spatial mapping, the hardware integration module first removes sampling points corresponding to missing dimensions before forming the multidimensional physiological characteristic vector sequence. Then, the remaining sampling points are grouped according to the biometric identifier, resulting in a current multidimensional physiological characteristic vector sequence that is uniquely bound to each biometric identifier.

[0029] For example, when the early shift coal mining team members enter the pre-shift meeting terminal, the hardware integration module first completes facial recognition in front of the check-in screen to generate a corresponding facial biometric identifier. During the effective collection phase of the facial biometric identifier, the wristband collection terminal continuously transmits the person's heart rate, blood oxygen saturation, and body surface temperature. The hardware integration module combines the three detection values ​​at the same time into a multidimensional physiological characteristic vector, and then writes the multiple vectors into the person's current multidimensional physiological characteristic vector sequence according to the sampling order, for subsequent support vector data description model to call.

[0030] S102 is a steady-state envelope surface fitting model based on support vector data description.

[0031] After the hardware integration module outputs the current multidimensional physiological characteristic vector sequence bound to the biometric identifier, it does not directly make a judgment based on a single threshold. Instead, it first establishes a corresponding steady-state discrimination boundary around each biometric identifier. The starting point for the formation of the steady-state discrimination boundary is not the current sample value itself, but the historical normal participation samples corresponding to that biometric identifier.

[0032] In practice, the hardware integration module retrieves multi-dimensional physiological feature vectors stored during historical normal attendance periods corresponding to each biometric identifier, and assembles them into a steady-state sample sequence based on temporal continuity. Subsequently, each multi-dimensional physiological feature vector in the steady-state sample sequence is used as the training input for the support vector data description model. A Gaussian kernel function is employed to map each multi-dimensional physiological feature vector from the original feature space to a higher-dimensional feature space. Within the higher-dimensional feature space, the minimum envelope boundary capable of encompassing the steady-state sample sequence is calculated. The Gaussian kernel function satisfies:

[0033] in, and Both represent multidimensional physiological characteristic vectors in steady-state sample sequences. This represents the Euclidean distance between two vectors. The kernel width parameter of the Gaussian kernel function is determined by the median of the Euclidean distance between pairwise multidimensional physiological feature vectors in the steady-state sample sequence. The support vector data description model completes the implicit mapping from the original feature space to the high-dimensional feature space through this Gaussian kernel function.

[0034] The corresponding solution relation is:

[0035] in, This represents the envelope center obtained by the support vector data description model in the feature space, used to indicate the center position of steady-state samples; The radius of the envelope represents the steady-state envelope surface and is used to define the spatial boundary of the steady-state envelope surface. Indicates the first The relaxation variables corresponding to each steady-state sample are used to allow for boundary relaxation in individual historical samples; This represents the slack variable penalty coefficient, used to control the tightness of the steady-state envelope and the tolerance for anomalies; This represents the number of samples in the steady-state sample sequence; This represents the feature space mapping corresponding to the Gaussian kernel function.

[0036] Among them, the historical normal attendance samples are selected from the sampling segments of the historical meeting records corresponding to the same biometric identifier that have not generated abnormal status representations and have completed the full process in compliance with regulations; the hardware integration module assembles the historical normal attendance samples into a steady-state sample sequence according to the sampling time sequence, and determines the penalty coefficient based on the proportion of samples in the historical normal attendance samples that are allowed to fall outside the envelope. After training, the hardware integration module retains the support vectors, corresponding Lagrange multipliers, envelope center, and envelope radius as the basis for subsequent anomaly detection. After completing the above calculations, the hardware integration module obtains the steady-state envelope surface parameters corresponding to each biometric identifier. These parameters include at least the envelope center and envelope radius, serving as a unified comparison basis for the current multidimensional physiological characteristic vector sequence entering the anomaly detection stage.

[0037] In the context of mine team meetings, the steady-state envelope formed by the support vector data description model does not set upper and lower limits for a single physiological indicator, but rather comprehensively covers the combination of multi-dimensional physiological characteristics of the same participant under normal pre-shift conditions. For example, although a member of the transportation team may experience slight fluctuations in body surface temperature during historically normal meeting periods, the combined trajectory of heart rate, blood oxygen saturation, and body surface temperature always falls within the same steady-state envelope. The hardware integration module preserves precisely this combined boundary, rather than unrelated individual thresholds.

[0038] Extraction and output of the abnormal state characterization set of S103 deviating from the steady-state envelope.

[0039] After obtaining the steady-state envelope parameters, the hardware integration module feeds the current multidimensional physiological sign vector sequence output by S101 into the support vector data description model one by one for boundary checks. Only after the comparison relationship is formed does the hardware integration module enter the abnormal state collection process, so that the output results can be directly called by the meeting process management module, instead of remaining at the single-point discrimination level.

[0040] In practice, the hardware integration module processes each multidimensional physiological sign vector in the current multidimensional physiological sign vector sequence. Calculate the discriminant value of its relative steady-state envelope surface:

[0041] in, This indicates the current multidimensional physiological characteristic vector sequence at the sampling time. The corresponding multidimensional physiological sign vector is derived from the current multidimensional physiological sign vector sequence output in step S101; and All parameters are taken from the steady-state envelope surface output in step S102, representing the envelope center and envelope radius obtained by the support vector data description model, respectively. Represents the current multidimensional physiological characteristic vector The degree of deviation from the steady-state envelope is used to determine whether the multidimensional physiological characteristic vector is located outside the steady-state envelope.

[0042] In the specific calculation, the hardware integration module calls upon the support vectors and corresponding Lagrange multipliers retained during the training phase, calculates the kernel value between the current multidimensional physiological characteristic vector and each support vector using a Gaussian kernel function, and then obtains the discriminant value by combining it with the envelope radius. The hardware integration module completes the boundary delimitation based on the sign of the discriminant value. At that time, It is determined to be located within the steady-state envelope; when At that time, A deviation from the steady-state envelope is identified, and the corresponding sampling time is recorded as an abnormal sampling point. To ensure that the abnormal results have value for subsequent control calls, the hardware integration module further merges temporally consecutive or adjacent abnormal sampling points according to the same biometric identifier to form an abnormal state representation. The abnormal state representation retains at least the biometric identifier, the time of deviation, the magnitude of deviation, and the source relationship of the corresponding multidimensional physiological sign vectors; multiple abnormal state representations are aggregated according to the scope of participants in the current shift to form an abnormal state representation set.

[0043] When the same biometric identifier meets the requirements at multiple consecutive sampling times When the continuous abnormal sampling points are merged into one abnormal state representation, the hardware integration module records the first abnormal sampling time as the deviation occurrence time and the largest discriminant value among the continuous abnormal sampling points as the deviation amplitude. When there is a single sampling point falling back into the steady-state envelope between two continuous abnormal sampling segments, if the duration of the falling sampling point does not exceed one sampling interval, the hardware integration module still merges the two abnormal sampling segments into the same abnormal state representation. After the abnormal state representation set is generated, the hardware integration module writes the abnormal state representation set into the status control entry of the meeting process management module, so that the meeting process management module can configure the status control logic of each core meeting node and construct the timing chain diagram according to the abnormal state representation set and the job attributes of the current shift. The calling method here is: the meeting process management module reads the deviation results corresponding to each biometric identifier in the abnormal state representation set and uses the participants with abnormal state representations as the key control objects of subsequent core meeting nodes.

[0044] For example, after a tunneling team member completes facial recognition during the pre-meeting data collection phase, the hardware integration module obtains the person's current multidimensional physiological characteristic vector sequence and verifies it point by point using the steady-state envelope formed in step S102. If multiple consecutive sampling points fall outside the steady-state envelope, the hardware integration module generates an abnormal state representation corresponding to the facial biometric identifier and incorporates this abnormal state representation into the abnormal state representation set. After reading this abnormal state representation set, the meeting process management module can set stricter state control logic for core meeting nodes such as safety oaths and risk pre-control, preventing the person from bypassing subsequent interactive confirmation steps.

[0045] Through the hardware integration module, the standardized intelligent management system for the entire process of mine team meetings has completed the unique binding of biometric identifiers of participants, the formation of current multidimensional physiological characteristic vector sequences, the fitting of steady-state envelope surface under the support vector data description model, and the extraction of abnormal state representation sets that deviate from the steady-state envelope surface. Among them, the abnormal state representation set, as the direct output of the hardware integration module, is sent to the meeting process management module, which then configures the state control logic of each core meeting node and constructs a time sequence diagram, so that subsequent process management is based on the objects whose pre-meeting states have been determined.

[0046] Detailed implementation of the meeting process management module: The hardware integration module has output an abnormal state representation set that corresponds one-to-one with each biometric identifier. Based on this, the meeting process management module has the input basis to identify the range of abnormal personnel in the current shift before the meeting. However, the abnormal state representation set alone cannot directly drive the shift meeting to proceed according to the mine safety regulations. The meeting process management module still needs to introduce the job type and operation attributes of the current shift, configure executable state control logic for each core meeting node, and build a timing chain diagram on the node sequence constraints and the shortest effective dwell time constraints. Finally, a navigation framework is generated for the database management module to parse and for the meeting management module to execute.

[0047] The following sub-steps S201-S203 are executed by the meeting process management module.

[0048] S201 receives the abnormal state representation set and determines the core meeting node set.

[0049] After the hardware integration module outputs the abnormal state representation set, the meeting workflow management module first reads this set and matches each abnormal state representation with the current shift's participant list based on its biometric identifier. This identifies the participants within the current shift who deviate from the steady-state envelope. Only after the abnormal state representation set and the current shift's participant list are matched does the meeting workflow management module retrieve the current shift's job attributes to form a unified input basis for node configuration.

[0050] In practice, the meeting workflow management module reads the job type attributes of the current shift from the shift schedule record. The job type attributes of the current shift refer to the job category and work scenario constraints corresponding to the current shift in this underground operation. Subsequently, the meeting workflow management module organizes the abnormal state representation set and the job type attributes of the current shift according to the same shift identifier to form a node configuration input set. After the node configuration input set enters the pre-fixed mine safety procedure node templates inside the meeting workflow management module, it first filters out the core meeting nodes that must be executed in this meeting according to job type, and then removes node templates that are irrelevant to the current shift according to work scenario constraints, resulting in a set of core meeting nodes that correspond one-to-one with the current shift.

[0051] In this process, the meeting workflow management module does not alter the biometric identifiers and abnormal status representations already established in the hardware integration module. Instead, it inherits the abnormal status representation set as the triggering basis for subsequent node control in situ. Taking a tunneling team as an example, after reading the job type and operation attributes of the current team, the meeting workflow management module first determines the set of core meeting nodes applicable to tunneling operations from the mine safety regulations node template. Then, it links the biometric identifiers belonging to the team in the abnormal status representation set to the determined set of core meeting nodes one by one, providing the input basis for configuring the status control logic in the next sub-step.

[0052] S202 configures the state control logic of each core conference node based on the node configuration input set.

[0053] Once the core meeting node set is formed, the meeting workflow management module uses each core meeting node as a processing unit. It sequentially writes the abnormal state representation set from the node configuration input set and the job attributes of the current shift into that core meeting node, generating the corresponding state control logic. This state control logic refers to the set of rules that constrain the entry, dwell, state interaction confirmation, and state transition of the core meeting node. This set of rules serves as the execution boundary when the meeting management module records node transition logs.

[0054] In practice, the meeting workflow management module first writes entry conditions and minimum effective dwell time for each core meeting node based on the job attributes of the current work group. The minimum effective dwell time is used to construct a standard preset timestamp sequence, and the entry conditions are used to ensure that the next core meeting node can only be activated after the previous core meeting node has reached the allowed flow state. Subsequently, the meeting workflow management module replaces the status interaction confirmation method of each core meeting node in situ based on the abnormal status representation set: when a biometric identifier does not correspond to an abnormal status representation, the biometric identifier uses the batch confirmation path in the corresponding core meeting node; when a biometric identifier corresponds to an abnormal status representation, the meeting workflow management module switches the status interaction confirmation method of the biometric identifier in the corresponding core meeting node to a single-person confirmation path, and updates the allowed flow conditions of the core meeting node to allow flow only after the single-person confirmation path is completed.

[0055] After the replacement is completed, the meeting workflow management module continues to solidify the node state transition rules for each core meeting node. These rules include at least four sequential states: pending entry, resident execution, state interaction confirmation, and allowed transition. A state transition relationship is established with the boundary that each subsequent state can only be triggered by the preceding state. Therefore, each core meeting node is no longer a static display element, but an executable business node with entry conditions, resident constraints, and interaction confirmation constraints. For example, in a pre-shift meeting of a coal mining team, if the abnormal state representation set shows that a participant has an abnormal state representation that deviates from the steady-state envelope, the meeting workflow management module will write the corresponding biometric identifier of that person into the individual confirmation path of the relevant core meeting node. When the on-site broadcast reaches that core meeting node, only after the biometric identifier completes the state interaction confirmation will the core meeting node transition from the state interaction confirmation state to the allowed transition state.

[0056] S203 constructs a timing chain diagram based on the state control logic.

[0057] After each core meeting node completes its state control logic configuration, the meeting process management module arranges the core meeting nodes into a directed temporal structure, forming a temporal chain graph, according to the sequential requirements of the core meeting procedures in the mine safety regulations. The graph nodes in the temporal chain graph are the core meeting nodes whose state control logic has been written into them, and the graph edges represent the preceding and following trigger relationships. Only when the preceding graph node reaches the allowed flow state will the entry condition corresponding to the following graph node be satisfied.

[0058] In practice, the meeting workflow management module first arranges the core meeting nodes into a node sequence in a fixed order. Then, it retrieves the shortest effective dwell time of each node corresponding to a core meeting node from the mine safety regulations, and accumulates these timestamps according to the order of the node sequence to generate a standard preset timestamp sequence. If the node in the node sequence... The shortest effective dwell time of each core meeting node is denoted as . Then the first in the standard preset timestamp sequence A standard preset timestamp satisfy:

[0059] in, For the first The minimum effective dwell time of each core meeting node is derived from the minimum effective dwell time requirement set by the mine safety regulations for that core meeting node. This refers to the standard preset timestamps accumulated according to the node sequence. The k-th standard preset timestamp in the sequence represents the standard time point accumulated from the start of the meeting until the k-th core meeting node completes its shortest effective dwell time. The meeting workflow management module will then... After being written into the time sequence graph in the order of the node sequence, the time sequence graph simultaneously has the node sequence relationship and the time base relationship. When the subsequent meeting management module extracts the actual execution timestamp sequence, it directly uses the standard preset timestamp sequence as the comparison benchmark.

[0060] In the context of mine team meetings, after the time sequence diagram is formed, the meeting process management module also binds the state control logic corresponding to each diagram node to that node. This ensures that each diagram node represents both the progress of the meeting and the flow constraints at that point. For example, when generating the time sequence diagram for a transportation team, the meeting process management module first writes the core meeting nodes according to the node sequence, and then binds the single-person confirmation path triggered by the abnormal state representation set to the corresponding diagram node. Ultimately, this ensures that the nodes in the time sequence diagram reflect both the order in which the transportation team meeting should proceed and whether a specific biometric identifier must be confirmed by a single person before proceeding to the next node.

[0061] S204 generates the navigation framework and outputs the call relationships.

[0062] After the timing graph is constructed, the meeting workflow management module encapsulates the graph nodes, edges, state control logic, and standard preset timestamp sequences in the timing graph to generate a navigation framework. The navigation framework is a meeting execution structure that can be directly called by subsequent modules. Internally, it no longer recalculates the exception judgment results of the hardware integration module, but directly inherits the control results already written to the graph nodes.

[0063] In practice, the meeting workflow management module assigns node association attributes to each graph node according to their arrangement in the time-series chain. These attributes include at least the current graph node's predecessor and successor relationships, state control logic, and the corresponding time point in the standard preset timestamp sequence. When a graph node has been triggered by an abnormal state representation set to form a single-person confirmation path, the meeting workflow management module writes this path into the corresponding graph node's node association attributes. After all graph nodes are encapsulated, the meeting workflow management module outputs a navigation framework and simultaneously sends it to both the database management module and the meeting management module. The database management module parses the association attributes of each node in the navigation framework and constructs a graph neural network composed of business nodes. The meeting management module drives the audiovisual equipment to synchronously output a security warning data graph according to the navigation framework, using the standard preset timestamp sequence and state control logic in the navigation framework as the execution benchmark for node transition log generation and process compliance judgment.

[0064] Through the meeting process management module, based on the abnormal state representation set output by the hardware integration module and combined with the job attributes of the current shift, the core meeting node set has been determined. The state control logic configuration of each core meeting node has been completed, a time-series chain diagram with preceding and following trigger relationships and standard preset timestamp sequences has been constructed, and a navigation framework containing node association attributes has been further generated. Among them, the navigation framework, as the direct output of the meeting process management module, is used by the database management module to parse the business node association attributes, and by the meeting management module to drive audiovisual equipment, generate node change logs, and conduct process compliance judgments.

[0065] Specific implementation of the database management module: While the meeting workflow management module has generated a navigation framework and clarified the execution relationships of each graph node, the framework itself primarily defines the organization and flow rules of meeting nodes, but has not yet established a stable correspondence between each graph node and the specific broadcast content. Safety warning materials in mine team meetings are not simply isolated collections of data; they need to be matched based on the position of graph nodes in the meeting workflow, their status control logic, and the relationships between preceding and subsequent nodes. Therefore, it is necessary to continue organizing and processing the data along the association attributes of each node in the navigation framework, transforming the node execution structure into a data retrieval structure.

[0066] The following sub-steps S301-S303 are executed by the database management module.

[0067] S301 parses the associated attributes of each node in the navigation framework and constructs business nodes.

[0068] After the meeting process management module outputs the navigation framework, the database management module first reads each graph node in the navigation framework and extracts the fixed node association attributes of each graph node. Only after the node association attributes have been extracted and reorganized does the database management module enter the node construction stage of the graph neural network, so that subsequent aggregation operations are no longer directly oriented towards the original graph nodes, but towards computable business nodes.

[0069] In practical implementation, the database management module uses a single graph node in the navigation framework as a foundation. It reads the successor-successor relationship, state control logic, and corresponding time points from the standard preset timestamp sequence of that graph node, and organizes the above content corresponding to the same graph node into a node association attribute vector in a fixed order. The node association attribute vector is formed by concatenating the sequential position of the business node in the navigation framework, the corresponding standard preset timestamp, the state control logic code, and the successor-successor relationship code in a fixed order. For single-person confirmation paths and batch confirmation paths in the state control logic, the database management module assigns different discrete codes to each, so that the business node retains the differences in execution constraints triggered by the abnormal state representation set when it is formed. The graph calculation unit formed based on this node association attribute vector is defined as a business node. The business node is used to represent the execution position, execution constraints, and time base of a core meeting node in the current shift meeting. Since the state control logic in the meeting process management module has already absorbed the abnormal state representation set and the job type attributes of the current shift, the business node inherits the single-person confirmation path triggered by the abnormal state and the node execution rules written according to the job type attributes when it is formed.

[0070] After business nodes are formed, the database management module constructs directed connections between them according to the successor-successor relationships in the navigation framework, and generates an adjacency matrix accordingly. The row and column indices in the adjacency matrix correspond to the order of the business nodes in the navigation framework. The first business node can be directly transferred to the second... When there are multiple business nodes, the corresponding positions in the adjacency matrix are marked as connected; when there is no direct flow relationship between them, they are marked as disconnected. To maintain the continuous transmission of the original constraints of a single business node in the graph neural network, the database management module adds self-connections to the adjacency matrix for each business node, thereby obtaining the graph structure input for subsequent aggregation iterations.

[0071] For example, in the pre-shift meeting of the tunneling team, the database management module sequentially reads the node association attributes of the safety oath node, risk pre-control node, and task handover node from the navigation framework. If the risk pre-control node has been written into the single-person confirmation path in the meeting process management module due to the abnormal state representation corresponding to a certain biometric identifier, then the single-person confirmation path enters the business node corresponding to the risk pre-control node along with the state control logic. Subsequently, the database management module establishes directed connections in the order of safety oath node to risk pre-control node and risk pre-control node to task handover node, forming a business node graph structure that can be computed by the graph neural network.

[0072] S302 performs aggregated iteration on the features of adjacent business nodes.

[0073] After the business nodes and adjacency matrix are formed, the database management module uses the node association attribute vectors of each business node as initial node features to construct a graph neural network composed of business nodes. The input starting point of the graph neural network is not static data entries, but business nodes whose execution rules have been written by the meeting process management module, so that the context constraints between adjacent meeting stages can enter the same round of feature propagation.

[0074] In practice, the database management module stacks the initial node features of all business nodes into an initial feature matrix according to the node order in the navigation framework. Let the adjacency matrix with self-connections be denoted as Press again The degree matrix is ​​obtained by calculating the degree of each row. Subsequently, the database management module uses a graph convolutional network as the aggregation model of the graph neural network, in the first... In round aggregation, the business node representation is updated according to the following formula:

[0075] in, The initial node feature matrix representing all business nodes is derived from the stacked result of the node association attribute vectors constructed in step S301. Indicates the first The node representation matrix of all business nodes during round aggregation; Indicates the first The node representation matrix output by the round aggregation; This represents an adjacency matrix with self-connections, used to characterize the connection relationships between business nodes; Indicates basis The degree matrix obtained by calculating the degree by row is used for normalization during adjacency propagation. Indicates the first The trainable mapping matrix corresponding to the round aggregation is used to map the propagated node features to a new feature space; This represents a non-linear activation function used to maintain the non-linear representational capability of graph neural networks.

[0076] The execution order of the above calculations is as follows: first, based on... Determine the range of adjacent business nodes for each business node, and then utilize... Complete the normalization and transmission of features from adjacent business nodes, and then transmit the results to... Multiplication and tracing The new node representation is output. In the offline phase, the graph neural network uses the navigation framework of completed historical meetings and the corresponding manually confirmed data retrieval results as training samples, and the actually retrieved security alert data as positive samples. It updates the trainable mapping matrix corresponding to each round of aggregation by minimizing the matching loss between business nodes and target security alert data. In the online execution phase, the database management module directly calls the offline-trained mapping matrix to perform aggregation iterations on the current business node, without retraining during meeting execution. After continuously executing a preset number of aggregation iterations, the database management module obtains an aggregated node representation sequence containing semantic constraints of upstream and downstream nodes.

[0077] During the aggregation process, if the state control logic of a business node includes a single-person confirmation path triggered by an abnormal state representation set, this constraint is not only retained in the node representation of that business node but also propagated to the preceding and following nodes through adjacent business nodes. This ensures that when the database management module selects security alert materials, it no longer considers a single core meeting node in isolation but also takes into account the execution connection relationship between the preceding and following nodes. For example, after a risk pre-control node is written into a single-person confirmation path, after aggregation through a graph convolutional network, the aggregated node representations corresponding to the security pledge node and the task disclosure node will also incorporate this execution constraint. This ensures that the subsequently pushed security alert materials not only cover the content of this node but also maintain consistency with the broadcast connection of adjacent nodes.

[0078] In one implementation, for example, the database management module can set the aggregation model of the graph neural network as a 3-layer graph convolutional network. The input layer receives an initial node feature matrix formed by stacking node association attribute vectors. The node association attribute vectors can be composed of 24-dimensional features according to the sequential position of the business nodes in the navigation framework, the corresponding standard preset timestamp, the state control logic encoding, and the predecessor and successor relationship encoding. The first layer of the graph convolutional network outputs a 64-dimensional node representation, the second layer outputs a 32-dimensional node representation, and the third layer outputs a 16-dimensional aggregated node representation. The first and second layers use a linear rectified activation function, and the third layer uses an identity mapping to output the final aggregated node representation. The deactivation ratio between adjacent layers is 0.2. The training dataset can be composed of navigation frameworks from completed historical meetings and corresponding manually confirmed data retrieval results. Each training sample includes at least a set of business nodes, an adjacency matrix, a node association attribute vector, the actual security alert data retrieved for each business node, and the security alert data not retrieved in the same batch. The text descriptions, image descriptions, or broadcast scripts of the security alert data are uniformly encoded into 16-dimensional data feature vectors, ensuring consistency with the 16-dimensional aggregated node representation output by the third-layer graph convolutional network. Batch supervised training can be used, with a batch size of 32, 120 training epochs, and a learning rate of 0.001. The optimization method is adaptive moment estimation. The loss function can be matching ranking loss, requiring that the matching score of the actual retrieved security alert data corresponding to the same business node is higher than the matching score of the unretrieved security alert data. If the difference is less than a preset interval of 0.5, the difference is updated in reverse. The pre-training parameters of the trainable mapping matrix can be obtained through offline training. In one embodiment, the dimensions of the trainable mapping matrices of the first, second, and third layers are 24×64, 64×32, and 32×16, respectively. The initial parameters are generated using a random distribution with a mean of 0 and are fixed and saved after offline training. During the online execution phase, the fixed trainable mapping matrix is ​​directly called to perform aggregation iteration on the current business node.

[0079] S303 generates a security warning data map based on the sequence of aggregated nodes.

[0080] After the aggregation node representation sequence is formed, the database management module retrieves a candidate set of data related to the mine team meeting from the database. The candidate set refers to a collection of safety warning data in the database that has been categorized for safety alert purposes and can be directly broadcast by the meeting management module. Before filtering, the database management module extracts the feature vector of each safety warning data in the candidate set, then performs node-by-node matching with the aggregation node representation sequence, ultimately forming a safety warning data map that can be directly attached to the navigation framework.

[0081] The candidate data set is pre-collected by the database management module according to the security topics corresponding to each core meeting node in the navigation framework. Each security alert in the candidate data set establishes a basic topic correspondence with at least one core meeting node. The database management module uniformly encodes the text descriptions, image descriptions, or broadcast scripts of the security alerts to form a data feature vector consistent with the representation dimension of the business node aggregation node, and uses this data feature vector as the candidate vector for business node matching. The uniform encoding rules remain fixed within the same database version, and the dimension of the uniformly encoded data feature vector is consistent with the representation dimension of the aggregation node. The same data feature vector is used for the same security alert during the matching stage.

[0082] In practice, the database management module uses the aggregation node corresponding to each business node as the query vector and the feature vector corresponding to each security alert document in the candidate data set as the matching vector to calculate the matching score between the business node and the security alert document. The matching score can be expressed as:

[0083] in, Indicates the first The aggregation node representation of each business node after the aggregation iteration is derived from the aggregation output of the graph neural network on the business nodes in step S302; Indicates the first The data feature vector of the safety warning document comes from the database management module's analysis of the first... The unified coding result of the safety warning materials; Indicates the first The business node and the first The matching score between security alert documents is used to measure the degree of matching between a business node and the security alert document. The database management module first retrieves the aggregated node representation of each business node in the order of nodes in the navigation framework, then sorts all matching scores corresponding to the same business node in descending order, selects the security alert documents with the highest ranking and consistent with the status control logic of the business node, and establishes a one-to-one push relationship between the business node and the security alert document. The push relationships of each business node are then concatenated according to the node order of the navigation framework to form a security alert document map.

[0084] After generating the safety alert data map, the database management module writes the push relationship between business nodes and safety alert data back to the corresponding graph nodes in the navigation framework, ensuring that each graph node in the navigation framework has a corresponding safety alert data retrieval result. The safety alert data map is then sent to the meeting management module. When the meeting management module drives the audiovisual equipment output according to the navigation framework, it directly retrieves the corresponding safety alert data from the safety alert data map in the order of the graph nodes and completes the on-site broadcast and status interaction confirmation.

[0085] For example, in the pre-shift meeting of the transportation team, after the database management module matches the aggregated node representation corresponding to the risk pre-control node with the data candidate set item by item, if the top-ranked safety warning data is the transportation line obstacle clearing warning data and the mechanical start / stop confirmation warning data, the database management module writes these two safety warning data into the corresponding graph node according to the status control logic of the risk pre-control node; when the meeting management module executes to the graph node, it calls the above safety warning data from the safety warning data graph for broadcasting, without needing to search the database again.

[0086] Through the database management module, based on the navigation framework generated by the meeting workflow management module, node association attribute parsing, business node construction, graph neural network aggregation iteration, and security warning data matching have been completed, forming a security warning data map organized according to the node order of the navigation framework. The push relationship between business nodes and security warning data is then attached back to the corresponding graph nodes of the navigation framework. Among them, the security warning data map, as the direct output of the database management module, is directly called by the meeting management module to drive the audiovisual equipment to complete on-site broadcasting and status interaction confirmation node by node.

[0087] Detailed implementation of the meeting management module: The navigation framework formed by the meeting workflow management module has determined the execution skeleton of the meeting nodes, and the security warning data map formed by the database management module has determined the data content corresponding to the nodes. The meeting processing has moved from the pre-meeting configuration stage to the on-site execution stage. At this point, it is necessary to place the graph node flow, audio-visual equipment output, status interaction confirmation, on-site audio and video acquisition, and node change recording within the same processing chain, so that the meeting is no longer just a matter of data playback and manual confirmation, but a process object that can be continuously recorded, continuously compared, and continuously judged.

[0088] The following sub-steps S401-S404 are executed by the conference management module.

[0089] S401 drives the audiovisual equipment to output synchronously according to the navigation framework and generates node transition logs.

[0090] After receiving the navigation framework output by the conference workflow management module and the safety warning data map output by the database management module, the conference management module first reads the currently accessible core conference nodes according to the order of the graph nodes in the navigation framework. Then, it activates the playback channel of the audiovisual equipment according to the status control logic corresponding to the core conference node. Only when the preceding graph nodes of the core conference node have reached the allowed flow state will the conference management module allow the current graph node to enter the waiting state, and simultaneously retrieve the safety warning data corresponding to the graph node in the safety warning data map, driving the display terminal and broadcast terminal to start outputting, thereby completing the on-site broadcast of the core conference node.

[0091] During the on-site broadcast, the meeting management module, based on the fixed state control logic of the graph node, sequentially advances through three subsequent states: resident execution, state interaction confirmation, and allowed flow. When the state control logic includes a single-person confirmation path triggered by an abnormal state representation set, the meeting management module binds the corresponding biometric identifier to the interactive terminal. Only after the biometric identifier completes state interaction confirmation is the current graph node written to the allowed flow state. Each time a state transition occurs, the meeting management module immediately records the graph node identifier, state name, and state transition time, continuously forming a node transition log. The node transition log is a time-series log recording the state changes of each core meeting node from pending entry to allowed flow, in the order of graph nodes. The subsequent extraction of the actual execution timestamp sequence by the meeting management module relies solely on this node transition log.

[0092] For example, when a risk pre-control node has been written into the single-person confirmation path, the meeting management module first drives the audio-visual device to broadcast the security warning information corresponding to the node, and then keeps the node in the resident execution state until the minimum effective resident time of the node is reached; afterwards, the interactive terminal only accepts touch or button confirmation from the person corresponding to the target biometric identification. After the confirmation is completed, the meeting management module switches the node from the status interaction confirmation to allow flow, and writes the above status switching times into the node transition log in sequence.

[0093] S402 performs spatiotemporal alignment encapsulation of on-site audio and video streams and node transition logs.

[0094] After the node transition log is generated, the meeting management module continuously receives live audio and video streams from the live video and audio capture terminals, and integrates these streams with the node transition log into the same meeting clock. Only after the node transition log and the live audio and video streams are unified to the same time base does the meeting management module enter the spatiotemporal alignment and encapsulation stage. This ensures that the records subsequently stored in the data traceability pool not only reflect the video and audio content, but also the specific graph node and state range corresponding to that content.

[0095] In practice, the meeting management module uses the time interval formed by two adjacent state transition times in the node transition log as the dividing boundary to segment the on-site audio and video streams. It then establishes a correspondence between each segment and the graph node identifier and state name corresponding to that time interval. When multiple consecutive state intervals exist for the same graph node, the meeting management module concatenates the corresponding on-site audio and video stream segments according to the state sequence of the graph node, forming a spatiotemporally aligned encapsulated record package for that graph node. The spatiotemporally aligned encapsulated record package refers to a record object formed by synchronously encapsulating on-site audio and video stream segments with the state trajectory of that graph node in the node transition log, using the graph node as the organizational unit. This record object serves as the entry carrier for subsequent entry into the pool, regardless of whether the path is compliant or blocked.

[0096] For example, during the residency execution phase of a security oath node, the meeting management module extracts the on-site audio and video streams within the corresponding time period based on the time when the node enters the residency execution state and the time of the state entry interaction confirmation in the node transition log, and encapsulates them together with the security oath node identifier and residency execution state; when the node further enters the state interaction confirmation and completes the allowed flow, the meeting management module continues to extract the on-site audio and video streams of subsequent time periods, and finally generates a complete spatiotemporally aligned encapsulated record package corresponding to the security oath node.

[0097] S403 extracts node flow time difference and instruction concurrency divergence.

[0098] After the spatiotemporal alignment encapsulation record package is generated, the meeting management module extracts compliance detection features from the node transition log as the core input. First, the meeting management module extracts the transition time corresponding to each graph node that has reached the allowed transition state from the node transition log, forming an actual execution timestamp sequence according to the graph node order. Then, it reads the standard preset timestamp sequence corresponding to each of the above graph nodes from the navigation framework. Item, of which This represents the number of graph nodes that have completed state transitions. Then, a dynamic time warping algorithm is used to perform non-linear spatiotemporal alignment on the two sets of timestamp sequences. The recursive relationship is as follows:

[0099] in, Indicates the previous The actual execution timestamp and the previous The cumulative cost value after aligning with the standard preset timestamps is used to describe the cumulative time deviation in the local alignment state. The cumulative cost value is obtained by adding the absolute value of the difference between the current actual execution timestamp and the current standard preset timestamp to the minimum value among the cumulative cost values ​​of the three adjacent paths. Indicates the first in the actual execution timestamp sequence The actual execution timestamp is derived from the first one in the node transition log. The time it takes for a graph node to enter the allowed flow state; Represents the first in the standard preset timestamp sequence Each standard preset timestamp is derived from the cumulative result of the shortest effective dwell time of the corresponding graph node in the navigation framework; and These represent the position indices of the actual execution timestamp and the standard preset timestamp, respectively, for the currently participating alignment. After the meeting management module completes global path optimization, it retrieves... As the node transition time difference, it is used to characterize the degree to which the current meeting progress sequence deviates from the standard preset timestamp sequence in the macro time dimension.

[0100] While extracting the time difference of node transitions, the meeting management module opens a discrete time window at each status interaction confirmation stage to poll and collect the underlying concurrent operation events generated by the interactive terminal at the microsecond level. The sampling interval of the discrete time window is determined by the polling frequency of the underlying events of the interactive terminal and remains fixed throughout the execution of the same shift meeting. The meeting management module uses this sampling interval to continuously divide the underlying concurrent operation events of the status interaction confirmation stage into windows, with each discrete time window being contiguous and non-overlapping on the timeline. The operation types are at least divided into two categories: screen touch events and button events, and can be further subdivided based on the underlying event encoding of the interactive terminal. The discrete time window refers to the continuous sampling window divided at a fixed microsecond sampling interval within the status interaction confirmation stage, and the underlying concurrent operation events include screen touch events and button events.

[0101] For the Within a discrete time window, the meeting management module first counts the number of triggers for each operation type. Then calculate the probability distribution of each operation type within the discrete time window. And based on this, the window divergence is calculated:

[0102] in, Indicates the sequential numbering of the discrete-time window within the state interaction confirmation phase; This indicates the operation type number, which must be categorized into at least two types: screen touch events and button events. Indicates the first Within the discrete time window, the first... The number of times the operation type is triggered; Indicates the first Within the discrete time window, the first... The probability of occurrence of class operation types; Indicates the first The window divergence corresponding to each discrete-time window is used to reflect the degree of disorder in the operation distribution within that discrete-time window. The meeting management module will handle all the status interaction confirmations throughout the entire process. The window divergence sequence is constructed in chronological order, and the peak value is taken as the instruction concurrency divergence to characterize the degree of chaos at the micro-operational level of the current meeting. If a graph node is in a reactive state, multiple operation types will occur concurrently in a very short time, which will increase the window divergence of the corresponding discrete time window, eventually causing the instruction concurrency divergence to reach a high value at that node.

[0103] For example, during the status interaction confirmation phase of a task handover node, if participants repeatedly click the screen and press the confirmation button to expedite the process, the meeting management module will simultaneously detect screen touch events and button events within multiple consecutive discrete time windows, resulting in a high window divergence peak. At the same time, if the allowed transfer time of the node in the node transition log is significantly earlier than the corresponding time point in the standard preset timestamp sequence, then the node transfer time difference and instruction concurrency divergence will simultaneously enter the subsequent judgment stage.

[0104] S404 completes process compliance determination based on covariance Markov divergence and solidifies it into the data traceability pool.

[0105] After extracting the node flow time difference and instruction concurrency divergence, the meeting management module constructs a two-dimensional feature vector from the two in a fixed order. ,in The node transition time difference is derived from the cumulative shortest generation value output by the dynamic time warping in step S403. The instruction concurrency divergence is represented by the peak value of the window divergence sequence throughout the entire status interaction confirmation process in step S403. Subsequently, the meeting management module reads the distribution center data from the historical compliance execution logs. and the benchmark covariance matrix and the two-dimensional feature vector Directly map to the reference covariance matrix space and calculate the covariance Mahalanobis divergence according to the following formula:

[0106] in, It represents the covariance Mahalanobis divergence and is also a continuous numerical output as a process deviation index, used to measure the statistical deviation of the current meeting operation status from the historical compliance execution status. This represents a two-dimensional feature vector composed of node flow time difference and instruction concurrency divergence. This represents the distribution center of historical compliance execution logs in a two-dimensional feature space; This represents the baseline covariance matrix generated from historical compliance execution logs; Represents the benchmark covariance matrix The inverse matrix.

[0107] Among them, historical compliance execution logs are selected from team meeting records that have not triggered a freeze and have been verified after the meeting. The meeting management module extracts the corresponding node flow time difference and instruction concurrency from each historical compliance execution log to form a historical two-dimensional feature sample set; distribution center The benchmark covariance matrix is ​​calculated from the mean vector of the historical two-dimensional feature sample set. It is calculated from the sample covariance of this historical two-dimensional feature sample set. When When numerical instability occurs during the inversion process, the meeting management module... A fixed small regularization term is added to the diagonal before inverting the result to keep the covariance Mahalanobis divergence calculation stable.

[0108] The meeting management module will deviate from the process index. The initial blocking limit is compared with a preset blocking threshold, which is determined by the mine's safety tolerance red line. The preset blocking threshold is determined by the meeting management module based on the process deviation index distribution corresponding to historical compliance execution logs. Specifically, the process deviation index of the historical compliance execution logs is sorted from smallest to largest, and the value corresponding to the preset high percentile is taken as the initial blocking threshold. This initial blocking threshold is then confirmed or tightened by the mine's safety tolerance red line. The preset high percentile is pre-defined by the calibration rules corresponding to the current version of the meeting system applicable to the mining area. Under the same mining area and the same version of the meeting system, the preset blocking threshold remains fixed within a calibration period. When the preset blocking threshold is not exceeded, the meeting management module maintains the current graph node's allowed flow state and solidifies the corresponding spatiotemporal alignment encapsulation record package, node flow time difference, instruction concurrency, and process deviation index into the data traceability pool; when When the preset blocking boundary is exceeded, the conference management module immediately drives the audio-visual equipment to output a violation warning, freezes the state transition of the current graph node, and intercepts the spatiotemporal alignment encapsulation writing of subsequent graph nodes. Only the spatiotemporal alignment encapsulation record package formed before the freezing time, along with the violation mark and process deviation index, is solidified into the data traceability pool.

[0109] In a mine team meeting scenario, if the node flow time difference corresponding to the risk pre-control node is large and the instruction concurrency divergence reaches a peak during the status interaction confirmation stage, after the process deviation index calculated by the meeting management module exceeds the preset blocking limit, the display terminal immediately stops at the current risk pre-control node, the broadcast terminal outputs a violation warning simultaneously, and the interactive terminal no longer accepts flow instructions to enter the next diagram node; at the same time, the meeting management module writes the spatiotemporal alignment encapsulation record package previously formed for the risk pre-control node, the freeze mark of the current risk pre-control node, and the corresponding process deviation index into the data traceability pool for subsequent traceability and review.

[0110] Through the meeting management module, based on the navigation framework and safety warning data map, we have completed the graph node-driven, status interaction confirmation, node transition log generation, and spatiotemporal alignment encapsulation of on-site audio and video streams and node transition logs. Furthermore, we have extracted the node flow time difference and instruction concurrency divergence, and used the benchmark covariance matrix to calculate the process deviation index. Finally, the meeting management module outputs the allowed flow or frozen result according to the preset blocking limit, and solidifies the corresponding spatiotemporal alignment encapsulation record package and judgment result into the data traceability pool.

[0111] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A standardized intelligent management system for the entire process of mine team meetings, characterized in that, include: Hardware integration module, meeting workflow management module, database management module, and meeting management module; The hardware integration module is used to acquire the multidimensional physiological characteristics and biometric identifiers of the participants, organize the multidimensional physiological characteristics according to the biometric identifiers, use the support vector data description model to perform spatial mapping and fit the steady-state envelope surface, and extract the abnormal state representation set that deviates from the steady-state envelope surface. The meeting process management module is used to receive the abnormal status representation set, and configure the status control logic of each core meeting node in combination with the job operation attributes of the current shift, construct the time sequence diagram, and generate a navigation framework containing node association attributes and standard preset timestamp sequences. The database management module is used to parse the association attributes of each node in the navigation framework, construct a graph neural network composed of business nodes, aggregate and iterate the features of adjacent business nodes, and generate a security warning data map that is back to the navigation framework. The meeting management module is used to drive the audio-visual equipment to output safety warning data maps according to the navigation framework and complete the status interaction confirmation. It performs spatiotemporal alignment and encapsulation of the on-site audio and video streams and node transition logs, extracts process compliance detection features for node transition logs to determine process compliance, and freezes the status flow of the current meeting node when it crosses the preset blocking threshold.

2. The standardized intelligent management system for the entire process of mine team meetings according to claim 1, characterized in that, The hardware integration module first generates a biometric identifier through a face recognition unit or a fingerprint recognition unit, and then continuously reads the heart rate, blood oxygen saturation and body surface temperature of the participants under the same sampling clock. The detection values ​​of each dimension obtained at the same sampling time are combined into a multidimensional physiological sign vector in a fixed order, and the sampling points corresponding to the missing dimensions are removed before forming the multidimensional physiological sign vector sequence.

3. The standardized intelligent management system for the entire process of mine team meetings according to claim 2, characterized in that, The hardware integration module retrieves historical normal attendee samples corresponding to biometric identifiers to form a steady-state sample sequence. It uses a Gaussian kernel function to complete spatial mapping and fit a steady-state envelope surface, retaining support vectors, corresponding Lagrange multipliers, envelope center and envelope radius, and merging multiple consecutive abnormal sampling points that deviate from the steady-state envelope surface into corresponding abnormal state representations.

4. The standardized intelligent management system for the entire process of mine team meetings according to claim 1, characterized in that, The meeting process management module reads the job attributes from the current shift's schedule record, and filters out the core meeting node set corresponding to the current shift based on the mine safety regulations node template. It then arranges each core meeting node in a fixed order to form a time-series chain diagram and generates a standard preset timestamp sequence based on the shortest effective dwell time of each core meeting node.

5. The standardized intelligent management system for the entire process of mine team meetings according to claim 4, characterized in that, The status control logic includes at least four sequential states: pending entry, resident execution, status interaction confirmation, and allowed flow. When the biometric identifier in the abnormal status representation set corresponds to a core meeting node, the meeting process management module switches the status interaction confirmation method of the core meeting node to a single-person confirmation path and uses the completion of the single-person confirmation path as a condition for allowing flow.

6. The standardized intelligent management system for the entire process of mine team meetings according to claim 1, characterized in that, The database management module organizes the successor-successor relationships, state control logic, and corresponding time points in the standard preset timestamp sequence of each graph node in the navigation framework into node association attribute vectors in a fixed order. Business nodes are constructed using node association attribute vectors, and adjacency matrices with self-connections are generated based on the successor-successor relationships in the navigation framework.

7. The standardized intelligent management system for the entire process of mine team meetings according to claim 6, characterized in that, The database management module uses the navigation framework of historical completed meetings and the corresponding manually confirmed data retrieval results as training samples to update the trainable mapping matrix of the graph neural network offline. During the online execution phase, it calls the trained trainable mapping matrix to aggregate and iterate the business nodes. Subsequently, the security warning data in the data candidate set is uniformly encoded and sorted according to the matching score to form a security warning data map.

8. The standardized intelligent management system for the entire process of mine team meetings according to claim 1, characterized in that, The meeting management module drives the audiovisual equipment to output corresponding safety warning information in the order of the graph nodes in the navigation framework, and promotes the resident execution, status interaction confirmation and permission flow according to the status control logic of the corresponding graph nodes. When switching between states, the graph node identifier, status name and status switching time are recorded to form a node transition log, and the on-site audio and video streams are segmented and encapsulated according to the adjacent status switching time.

9. The standardized intelligent management system for the entire process of mine team meetings according to claim 8, characterized in that, The meeting management module extracts the transition time of each graph node entering the allowed transition state from the node transition log, forms an actual execution timestamp sequence according to the graph node order, and reads the standard preset timestamp sequence corresponding to each graph node in the navigation framework for comparison. Simultaneously, during the state interaction confirmation phase, a discrete time window is opened to statistically analyze the underlying concurrent operation events generated by the interactive terminal according to the operation type.

10. The standardized intelligent management system for the entire process of mine team meetings according to claim 9, characterized in that, The meeting management module determines the node flow time difference based on the comparison between the actual execution timestamp sequence and the standard preset timestamp sequence, determines the instruction concurrency divergence based on the statistical results of underlying concurrent operation events within the discrete time window, and maps the node flow time difference and instruction concurrency divergence to the benchmark covariance matrix space generated by the historical compliant execution log for judgment. When the preset blocking limit is exceeded, a violation warning is output and the state flow of the current meeting node is frozen.

Images