Comprehensive traffic hub personnel abnormal behavior emergency disposal method and system based on group risk state transition and self-learning decision

Abstract

This invention relates to an emergency response method and system for abnormal personnel behavior in integrated transportation hubs based on group risk state transition and self-learning decision-making. The system includes: a group perception and interaction modeling module, a risk state transition prediction module, a self-learning decision-making module, and a spatial execution and feedback module. The group perception and interaction modeling module receives real-time data from within the integrated transportation hub and external events and environmental information to construct a group risk state representation. The risk state transition prediction module establishes a time-series prediction model to predict and evaluate the evolution trend and risk level changes of the group risk state. The self-learning decision-making module generates emergency response objectives and constraints based on the risk prediction results and generates and evaluates candidate response strategies. The spatial execution and feedback module executes the emergency response strategies. The beneficial effects of this invention are: it can provide more efficient, robust, and adaptive emergency response support, demonstrating significant technological advancement and practical application value.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an emergency response method and system for abnormal behavior of personnel in integrated transportation hubs based on group risk state transition and self-learning decision-making. Background Technology

[0002] 1. A method, system, device, and storage medium for detecting abnormal behavior of people in subways (CN117036140A): This invention proposes a method and system for detecting abnormal behavior of people in high-density pedestrian areas such as subways. It detects each person entering the detection area, acquiring facial recognition information, physiological characteristics, and behavioral features of the target. Based on a preset threshold, it determines whether the target is a suspicious person. If a suspicious person is identified, its abnormal behavior information is further collected, and its risk level is determined using an abnormal behavior model. When a risk is detected, an emergency response plan corresponding to that risk is triggered. The key to this technical solution is the use of multi-source sensing information to screen, mark, and assess the risk of individual targets, enabling early detection and warning of potentially dangerous individuals. It also reduces the likelihood of safety accidents through preset emergency response plans, forming a complete abnormal behavior detection and handling process centered on "individual identification—risk assessment—emergency response."

[0003] 2. A Method and Device for Detecting Abnormal Behavior in Public Places Based on Cloud-Edge-Device Collaborative Scheduling (CN202311534125.8): This invention proposes a method for detecting abnormal behavior in public places based on a cloud-edge-device collaborative architecture. It involves deploying camera devices at the entrance of public places to collect real-time video data, uploading the video data to a cloud server via a local area network, and preprocessing and optimizing the video stream in the cloud. The HOTR neural network model is used to detect and identify human-object and human-human interactions in the video frames, thereby determining whether abnormal behavior exists. Upon detection of an anomaly, an anomaly alarm is sent to the corresponding entrance. The core of this technical solution lies in improving the real-time performance and recognition efficiency of abnormal behavior detection in multi-entry, multi-video-stream scenarios through a collaborative scheduling mechanism of centralized computing power in the cloud and data collection at the edge. This allows abnormal behavior to be quickly located and alerted to on-site management personnel, thus assisting in the safety supervision and risk prevention work of public places.

[0004] 3. A method, device, medium, and equipment for detecting abnormal passenger behavior in escalators (CN202510801738.6): This patent belongs to a general evidence fusion method, mainly addressing the credibility and uncertainty issues of multi-source conflicting evidence. It calculates the trust factor, divergence, belief entropy, and information weight of the evidence vector, performs a weighted average of the evidence, and outputs the final decision result using the DS combination rule. Its innovation lies in the evidence weight correction and conflict suppression mechanism. While the method is relatively complete at the algorithm level, its application scenario is represented by automotive system fault diagnosis, belonging to the paradigm of "constructed evidence → optimized fusion rules." It does not involve the evidence generation process, nor does it consider issues such as risk classification, prediction lead time, or decision timeliness.

[0005] Deficiencies of existing technology:

[0006] 1. The existing system (a method, system, equipment and storage medium for detecting abnormal behavior of subway pedestrians CN117036140A) mainly focuses on the identification and risk assessment of abnormal behavior of a single detection target. Its emergency response method relies on a preset emergency plan, focusing on the identification and handling of abnormal individuals. It does not model and analyze the behavioral changes and risk diffusion process of the surrounding crowd after the occurrence of abnormal behavior, making it difficult to reflect the dynamic impact of abnormal events on the overall state of the crowd, nor can it adaptively adjust the function and passage mode of the public space according to the evolution of the group's risk state.

[0007] 2. Existing System (Method and Device for Detecting Abnormal Behavior in Public Places Based on Cloud-Edge-Device Collaborative Scheduling CN202311534125.8) This technology improves the real-time performance and recognition efficiency of abnormal behavior detection through cloud-edge-device collaborative scheduling. Its technical focus is on video data processing and abnormal alarm output. After detecting abnormal behavior, it mainly uses alarm feedback as the handling result. It lacks a mechanism for continuous analysis of changes in the crowd situation caused by abnormal events, does not establish a correlation model between abnormal behavior and group risk state, and does not form an emergency response closed loop that dynamically adjusts spatial access rules and guidance strategies based on changes in risk state.

[0008] 3. Existing System (A Method, Device, Medium and Equipment for Detecting Abnormal Passenger Behavior on Escalators CN202510801738.6) This technology achieves refined identification of typical abnormal behaviors in escalator scenarios. Its core is to improve the accuracy of abnormal detection under video analysis conditions. However, the technical goal is still limited to the identification of abnormal behaviors. It does not consider the chain reaction of abnormal behavior on the behavior of surrounding passenger groups after the occurrence of abnormal behavior, nor does it involve an emergency response mechanism that dynamically adjusts the escalator passage mode or space usage based on changes in the risk status of the crowd. It is difficult to cope with the group-level risk evolution problem in complex public spaces. Summary of the Invention

[0009] This invention provides an emergency response method and system for abnormal behavior of personnel in integrated transportation hubs based on group risk state transition prediction and self-learning decision-making. It aims to solve the problems that existing technologies related to abnormal behavior in integrated transportation hubs generally have, such as focusing on individual abnormality identification in emergency response, lacking a systematic characterization of the group risk state evolution process after an abnormal event occurs, difficulty in predicting risk development trends in advance, reliance on static plans for emergency response objectives that are difficult to dynamically optimize with changes in the state of the population, and difficulty in adaptively reconfiguring the spatial access and functional configuration of transportation hubs according to changes in risk state.

[0010] This invention provides an emergency response system for abnormal personnel behavior in integrated transportation hubs based on group risk state transition and self-learning decision-making, comprising: a group perception and interaction modeling module, a risk state transition prediction module, a self-learning decision-making module, and a spatial execution and feedback module, wherein...

[0011] The group perception and interaction modeling module is used to receive real-time data from inside the integrated transportation hub and external events and environmental information, perform multi-source fusion perception of abnormal behavior events and the state of the crowd within their range of influence, and construct a group risk state representation that reflects the interaction relationship of the crowd.

[0012] The risk state transition prediction module is used to establish a time-series prediction model based on the group risk state characterization, predict and evaluate the evolution trend and risk level changes of the group risk state, and generate risk development results for future moments.

[0013] The self-learning decision-making module is used to generate emergency response objectives and constraints based on risk prediction results, generate and evaluate candidate response strategies through a self-learning strategy optimization mechanism, and determine the confidence level of the strategies. When the confidence level is insufficient, a fallback simulation and manual collaboration mechanism is triggered to correct the response plan.

[0014] The spatial execution and feedback module is used to execute the emergency response strategy, including the dynamic reconstruction of the spatial functions of the transportation hub and crowd guidance and control, and to continuously collect feedback information on crowd behavior during the execution process. The feedback is then transmitted back to the risk state transition prediction module to update the risk prediction model and decision-making strategy, thereby forming a closed-loop collaborative operation system of continuous adaptive evolution of perception, prediction, decision-making, execution, and feedback.

[0015] As a further improvement of the present invention, the group perception and interaction modeling module performs the following steps:

[0016] Step a1: Receive multi-source data stream input from the hub, including video surveillance data, turnstile passenger flow data, escalator and equipment operating status and abnormal event information, and perform time synchronization and spatial alignment processing on the above data to form a unified data input basis;

[0017] Step a2: Extract and estimate the crowd density, movement speed, direction of movement, and local aggregation characteristics in the affected area through the crowd state perception unit, and construct a basic feature set reflecting the current movement state of the crowd;

[0018] Step a3: Determine the completeness and reliability of the perceived data. When the perceived data meets the quality requirements, construct a dynamic crowd interaction relationship model through the crowd interaction modeling unit to model the crowd interaction behavior. When the perceived data is missing or occluded, execute the statistical estimation method through the robust state estimation unit to maintain the continuity and stability of the crowd state estimation.

[0019] Step a4: Generate a group risk state representation, comprehensively encode it through multi-dimensional risk indicators to form a structured group risk state vector, and simultaneously output the corresponding interaction graph structure information;

[0020] Step a5: Output the group risk state vector and interaction structure as standardized interface data to the risk state transition prediction module to drive subsequent time-series risk prediction and intelligent decision-making calculation, thereby providing a highly reliable population perception basis for dynamic emergency response to abnormal behavior events.

[0021] As a further improvement of the present invention, the risk state transition prediction module performs the following steps:

[0022] Step b1: Receive the group risk state vector and interaction graph structure at the current and historical times, construct a risk state time series cache through a sliding time window, and perform sequence alignment and consistency processing on the data at different time steps;

[0023] Step b2: Jointly encode the risk state evolution features and spatial correlation in the temporal feature encoding unit, and form a temporal embedding representation suitable for the prediction model by extracting the risk change rate, regional coupling strength and structural topology features;

[0024] Step b3: In the group risk transition prediction unit, multi-step forward inference is performed on the encoded time-series features based on the time-series graph neural network to predict the risk state distribution within several future time steps, and the prediction uncertainty and confidence index are calculated simultaneously.

[0025] Step b4: Determine the prediction confidence level. When the prediction result meets the confidence level requirements, generate a structured risk level output in the risk level assessment unit through risk classification mapping and trend stability analysis. When the prediction confidence level is insufficient or abnormal drift is detected, trigger the prediction correction mechanism to correct the prediction result through historical risk pattern matching and robust compensation calculation.

[0026] Step b5: The risk transition trend generation unit comprehensively outputs the future risk state estimation results, risk transition direction and diffusion trend information, and sends the results as standardized prediction interface data to the self-learning decision module to drive the generation and optimization of subsequent emergency response strategies, thereby achieving forward-looking risk prediction support for abnormal behavior events.

[0027] As a further improvement of the present invention, the self-learning decision module performs the following steps:

[0028] Step c1: Receive the predicted future risk state distribution and its evolution trend, and construct a decision state space by combining the spatial structure constraints and operation rules of the transportation hub. Encode the information into a decision state representation for strategy calculation. The information includes risk state, spatial accessibility, and facility capacity.

[0029] Step c2: In the emergency response target generation unit, the control targets for the current stage are determined based on the risk-driven principle. The targets are jointly characterized through multi-objective constraint modeling, and a set of targets with priority relationships is formed. The targets include evacuation efficiency, congestion suppression, and operational stability.

[0030] Step c3: In the self-learning strategy optimization unit, reinforcement learning is used to search and optimize candidate disposal strategies. The comprehensive performance of different strategies in terms of risk suppression effect, execution cost and system stability is evaluated through the reward function to generate a set of candidate strategies.

[0031] Step c4: Perform security constraint checks on candidate strategies. When a strategy meets the security conditions, determine the optimal strategy in the strategy evaluation and selection unit based on cost-benefit analysis and stability tests. When a strategy is detected to violate security constraints or has unexecutable risks, trigger the security correction mechanism to correct the strategy through rule constraint projection or conservative strategy rollback.

[0032] Step c5: The self-learning strategy optimization unit outputs structured emergency response instructions and their execution parameters, and sends the decision results to the spatial execution module. At the same time, it provides basic data support for subsequent online learning and strategy updates based on execution feedback, thereby forming a prediction-driven, self-learning optimization intelligent decision-making mechanism.

[0033] As a further improvement of the present invention, the space execution and feedback module performs the following steps:

[0034] Step d1: Receive structured emergency response strategies in the decision instruction parsing and execution plan generation unit, convert the strategies into executable control sequences, and generate an execution scheduling plan in combination with transportation hub operation resources and time constraints;

[0035] Step d2: In the spatial function dynamic reconstruction unit, the functional status of channels, areas and entrances and exits is adjusted in real time, and the spatial passage structure is changed by implementing measures such as flow restriction control, one-way passage or zone isolation.

[0036] Step d3: In the crowd guidance and system linkage control unit, crowd behavior is guided through broadcast prompts, display guidance, and signal control, and linkage control is performed with the equipment management system and security system;

[0037] Step d4: Continuously monitor the execution effect. When the execution result reaches the expected control target, collect changes in crowd behavior and system operation status in the execution monitoring and feedback unit, and generate structured feedback data. When the execution effect is detected to deviate from the expectation or an anomaly occurs, trigger the execution strategy adjustment mechanism to correct the control strategy through local replanning or manual intervention.

[0038] Step d5: In the online learning and model update unit, the execution feedback data is sent back to the perception and prediction module to update the risk prediction model and decision-making strategy, thereby forming a closed-loop control system of continuous adaptive evolution of prediction, decision-making, execution, and feedback.

[0039] As a further improvement of the present invention, in step a3, the group interaction behavior includes inter-regional flow coupling, directional conflict and congestion propagation; the statistical estimation method includes density correction, optical flow compensation or count correction; in step a4, the multidimensional risk indicators include congestion risk, diffusion risk, conflict risk and stagnation risk.

[0040] This invention also discloses an emergency response method for abnormal personnel behavior in integrated transportation hubs based on group risk state transition and self-learning decision-making, including:

[0041] Step 1: Receive relevant information on abnormal behavior events and spatial environment information of the integrated transportation hub, and construct an initial crowd perception state reflecting the current state of the crowd at the transportation hub;

[0042] Step 2: Model and analyze the interaction behavior of the crowd within the scope of the abnormal event, taking into account the crowd density distribution, changes in movement direction, local clustering characteristics and inter-individual interaction relationships, to generate a group risk state representation to characterize the degree of impact of the abnormal event;

[0043] Step 3: Introduce a risk prediction mechanism based on time-series group interaction modeling to predict and track the evolution trend of group risk state in the time dimension, and determine the risk state transition trend and the corresponding risk level change.

[0044] Step 4: Based on the self-learning decision-making mechanism, the emergency response objectives and their execution priorities under the current abnormal event are adaptively generated according to the risk prediction results in Step 3, so that the response strategy can be dynamically optimized and adjusted as the risk status of the population changes.

[0045] Step 5: Generate a spatial function dynamic reconstruction strategy and a crowd guidance strategy that match the disposal objectives of Step 4. By adjusting the functional attributes and traffic rules of passages, areas or entrances and exits within the transportation hub in real time, orderly guidance of crowd behavior and risk suppression can be achieved.

[0046] Step 6: Execute the emergency response strategy generated in Step 5, and continuously obtain feedback information on changes in population behavior and the effectiveness of the response. Use the feedback to update the risk prediction model through an online learning mechanism, thereby forming a closed-loop operation mechanism in which prediction, decision-making and execution are coupled and continuously adaptively evolved during the emergency response to abnormal events.

[0047] As a further improvement of the present invention, in step 1, the relevant information of the abnormal behavior event includes the type of abnormal behavior event, the location of occurrence, and the time characteristics.

[0048] The beneficial effects of this invention are as follows: Compared with existing technologies for handling abnormal behavior in public spaces, which mainly rely on single-point event detection, static emergency plans, or manual experience-based decision-making, this invention effectively overcomes the shortcomings of existing technologies, such as insufficient characterization of group risk evolution, delayed response, and limited strategy adaptability, by introducing a closed-loop emergency response mechanism based on group risk state transition prediction and self-learning decision-making. This invention, through a group perception and interactive modeling mechanism, transforms the impact of abnormal behavior on surrounding populations into a structured representation of group risk state, realizing a shift from individual abnormality identification to group risk modeling, significantly improving the ability to characterize complex crowd behavior patterns. Furthermore, by introducing a temporal risk transition prediction model, this invention can proactively infer future risk development trends, transforming emergency response from a reactive response to a prediction-driven proactive control, thereby reducing the possibility of risk spread and congestion escalation. Further, this invention, through a reinforcement learning-driven self-learning decision-making mechanism, automatically generates and optimizes response strategies under multi-objective constraints, exhibiting stronger environmental adaptability and strategy optimization capabilities compared to fixed rules or preset schemes. Meanwhile, through a spatial execution and feedback closed-loop mechanism, this invention achieves continuous linkage between execution results and model updates, enabling the system to continuously improve prediction accuracy and decision quality during long-term operation, thereby enhancing the stability and sustainable operation capability of engineering deployment. In summary, this invention achieves predictive identification, intelligent decision-making, and closed-loop control of abnormal behavior events in complex transportation hub environments, providing more efficient, robust, and adaptive emergency response support, demonstrating significant technological advancements and practical application value. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the overall operation process of the system of the present invention;

[0050] Figure 2 This is the overall system architecture diagram of the present invention;

[0051] Figure 3 This is a schematic diagram of the operation of the group perception and interaction modeling module of the present invention;

[0052] Figure 4 This is a schematic diagram of the operation of the risk state transition prediction module of the present invention;

[0053] Figure 5 This is a flowchart of the self-learning decision-making module operation of the present invention;

[0054] Figure 6 This is a flowchart of the spatial execution and feedback module of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0056] This invention uses abnormal behavioral events and their spatiotemporal characteristics as trigger inputs. By constructing a group risk state representation model oriented towards the interaction characteristics of people in integrated transportation hubs, it uniformly transforms the impact of abnormal events on the behavior of surrounding passenger groups into a calculable and evolvable risk state. Based on this, it introduces a risk state transition prediction mechanism based on temporal group interaction modeling to achieve dynamic identification and forward-looking judgment of group risk development trends. Furthermore, combined with a self-learning decision-making mechanism, it adaptively generates and continuously optimizes emergency response targets based on the risk state prediction results, linking them to form matching spatial function temporary reconstruction and group guidance strategies, thereby establishing a closed-loop emergency response system of "risk state prediction—self-learning decision-making—spatial response execution." Through the above technical solution, this invention realizes the transformation of emergency response to abnormal events in integrated transportation hubs from the traditional model mainly based on individual control and static response to an intelligent emergency response model with group risk state prediction and self-learning optimization as its core, possessing dynamic adjustment capabilities and overall coordination. This provides more stable, efficient, and forward-looking technical support for the management of abnormal events in high-density transportation hub environments.

[0057] like Figure 1 As shown, this invention discloses an emergency response method for abnormal behavior of personnel in integrated transportation hubs based on group risk state transition and self-learning decision-making, including:

[0058] Step 1: Taking the passenger group in the area where the abnormal behavior event occurred and its influence range as the analysis object, first receive the type, location, and time characteristics of the abnormal behavior event, as well as environmental information such as the spatial structure layout and traffic conditions of the integrated transportation hub. Based on the above inputs, construct an initial crowd perception state that reflects the current state of the crowd in the transportation hub.

[0059] Step 2: After completing the initial crowd perception state construction, the system models and analyzes the crowd interaction behavior within the scope of the abnormal event's impact. Taking into account crowd density distribution, changes in movement direction, local aggregation characteristics, and inter-individual interaction relationships, the system generates a group risk state representation to characterize the degree of impact of the abnormal event.

[0060] Step 3: The system introduces a risk prediction mechanism based on time-series group interaction modeling to predict and track the evolution trend of group risk state in the time dimension, determine the risk state transition trend and the corresponding risk level change, so as to identify different evolution states such as risk diffusion, escalation or mitigation.

[0061] Step 4: Based on the self-learning decision-making mechanism, the system adaptively generates emergency response targets and their execution priorities under the current abnormal event according to the risk prediction results in Step 3, so that the response strategy can be dynamically optimized and adjusted as the risk status of the population changes.

[0062] Step 5: After clarifying the emergency response objectives, the system generates a spatial function dynamic reconstruction strategy and a crowd guidance strategy that match the response objectives in Step 4. By adjusting the functional attributes and traffic rules of passages, areas or entrances and exits within the transportation hub in real time, the system achieves orderly guidance of crowd behavior and risk suppression.

[0063] Step 6: The system executes the emergency response strategy generated in Step 5 and continuously acquires feedback information on changes in population behavior and the effectiveness of the response. Through an online learning mechanism, the feedback is used to update the risk prediction model, thereby forming a closed-loop operation mechanism in which prediction, decision-making, and execution are coupled and continuously adaptively evolve during the emergency response to abnormal events.

[0064] like Figure 2 As shown, this invention also discloses an emergency response system for abnormal behavior of personnel in integrated transportation hubs based on group risk state transition and self-learning decision-making. By constructing a technical architecture that integrates multi-source perception-driven, risk prediction, self-learning decision-making, and execution feedback closed-loop collaborative operation, it achieves intelligent emergency response and dynamic optimization for abnormal behavior events in high-density transportation hub environments. The system includes: a group perception and interaction modeling module, a risk state transition prediction module, a self-learning decision-making module, and a spatial execution and feedback module. The specific functions of each module are as follows:

[0065] Group perception and interaction modeling module: It is used to receive real-time data inside the integrated transportation hub and external events and environmental information, perform multi-source fusion perception of abnormal behavior events and the state of the crowd within their scope of influence, and construct a group risk state representation that reflects the interaction relationship of the crowd.

[0066] like Figure 3 As shown, the group perception and interaction modeling module performs the following steps:

[0067] Step a1: This module first receives multi-source data stream input from the hub, including video surveillance data, turnstile passenger flow data, escalator and equipment operating status and abnormal event information, and performs time synchronization and spatial alignment processing on the above data to form a unified data input basis;

[0068] Step a2: Extract and estimate the crowd density, movement speed, direction of movement, and local aggregation characteristics in the affected area through the crowd state perception unit, and construct a basic feature set reflecting the current movement state of the crowd;

[0069] Step a3: Determine the completeness and reliability of the sensing data. When the sensing data meets the quality requirements, construct a dynamic crowd interaction relationship model through the crowd interaction modeling unit to model crowd interaction behaviors such as inter-regional flow coupling, directional conflict and congestion propagation. When the sensing data is missing or occluded, execute statistical estimation methods such as density correction, optical flow compensation or count correction through the robust state estimation unit to maintain the continuity and stability of crowd state estimation.

[0070] Step a4: After completing the modeling of the interaction relationship among the crowd, a group risk state representation is generated. By comprehensively encoding multi-dimensional risk indicators such as crowding risk, diffusion risk, conflict risk and stagnation risk, a structured group risk state vector is formed, and the corresponding interaction graph structure information is output simultaneously.

[0071] Step a5: Output the group risk state vector and interaction structure as standardized interface data to the risk state transition prediction module to drive subsequent time-series risk prediction and intelligent decision-making calculation, thereby providing a highly reliable population perception basis for dynamic emergency response to abnormal behavior events.

[0072] Risk state transition prediction module: used to establish a time series prediction model based on the group risk state representation, predict and evaluate the evolution trend of the group risk state and the changes in risk level, and generate the risk development results for future time.

[0073] like Figure 4 As shown, the risk state transition prediction module performs the following steps:

[0074] Step b1: Receive the group risk state vector and interaction graph structure at the current and historical times, construct a risk state time series cache through a sliding time window, and perform sequence alignment and consistency processing on the data at different time steps;

[0075] Step b2: Jointly encode the risk state evolution features and spatial correlation in the temporal feature encoding unit, and form a temporal embedding representation suitable for the prediction model by extracting the risk change rate, regional coupling strength and structural topology features;

[0076] Step b3: In the group risk transition prediction unit, multi-step forward inference is performed on the encoded time-series features based on the time-series graph neural network to predict the risk state distribution within several future time steps, and the prediction uncertainty and confidence index are calculated simultaneously.

[0077] Step b4: Determine the prediction confidence level. When the prediction result meets the confidence level requirements, generate a structured risk level output in the risk level assessment unit through risk classification mapping and trend stability analysis. When the prediction confidence level is insufficient or abnormal drift is detected, trigger the prediction correction mechanism to correct the prediction result through historical risk pattern matching and robust compensation calculation.

[0078] Step b5: The risk transition trend generation unit comprehensively outputs the future risk state estimation results, risk transition direction and diffusion trend information, and sends the results as standardized prediction interface data to the self-learning decision module to drive the generation and optimization of subsequent emergency response strategies, thereby achieving forward-looking risk prediction support for abnormal behavior events.

[0079] The self-learning decision-making module is used to generate emergency response objectives and constraints based on risk prediction results, and to generate and evaluate candidate response strategies through a self-learning strategy optimization mechanism. At the same time, it determines the confidence level of the strategy and triggers a fallback simulation and human collaboration mechanism to correct the response plan when the confidence level is insufficient.

[0080] like Figure 5 As shown, the self-learning decision module performs the following steps:

[0081] Step c1: First, receive the predicted future risk state distribution and its evolution trend, and construct a decision state space by combining the spatial structure constraints and operation rules of the transportation hub. Information such as risk state, spatial accessibility and facility capacity are uniformly encoded into a decision state representation for strategy calculation.

[0082] Step c2: In the emergency response target generation unit, the control targets for the current stage are determined based on the risk-driven principle. Through multi-objective constraint modeling, targets such as evacuation efficiency, congestion suppression and operational stability are jointly characterized, and a set of targets with priority relationships is formed.

[0083] Step c3: In the self-learning strategy optimization unit, reinforcement learning is used to search and optimize candidate disposal strategies. The comprehensive performance of different strategies in terms of risk suppression effect, execution cost and system stability is evaluated through the reward function to generate a set of candidate strategies.

[0084] Step c4: To ensure the safe and executable nature of the generated strategy in the actual transportation hub environment, the module further performs safety constraint checks on the candidate strategies. When the strategy meets the safety conditions, the optimal strategy is determined in the strategy evaluation and selection unit based on cost-benefit analysis and stability test. When a strategy is detected to violate safety constraints or has an unexecutable risk, a safety correction mechanism is triggered to correct the strategy through rule constraint projection or conservative strategy rollback.

[0085] Step c5: The self-learning strategy optimization unit outputs structured emergency response instructions and their execution parameters, and sends the decision results to the spatial execution module. At the same time, it provides basic data support for subsequent online learning and strategy updates based on execution feedback, thereby forming a prediction-driven, self-learning optimization intelligent decision-making mechanism.

[0086] Spatial Execution and Feedback Module: Used to execute emergency response strategies, including dynamic reconstruction of the spatial functions of transportation hubs and crowd guidance and control, and continuously collect feedback information on crowd behavior during the execution process. The feedback is then transmitted back to the risk state transition prediction module to update the risk prediction model and decision-making strategy, thereby forming a closed-loop collaborative operation system of continuous adaptive evolution of perception, prediction, decision-making, execution and feedback.

[0087] like Figure 6 As shown, the spatial execution and feedback module performs the following steps:

[0088] Step d1: Receive structured emergency response strategies in the decision instruction parsing and execution plan generation unit, convert the strategies into executable control sequences, and generate an execution scheduling plan in combination with transportation hub operation resources and time constraints;

[0089] Step d2: In the spatial function dynamic reconstruction unit, the functional status of channels, areas and entrances and exits is adjusted in real time, and the spatial passage structure is changed by implementing measures such as flow restriction control, one-way passage or zone isolation.

[0090] Step d3: In the crowd guidance and system linkage control unit, crowd behavior is guided through broadcast prompts, display guidance, and signal control, and linkage control is performed with the equipment management system and security system;

[0091] Step d4: Continuously monitor the execution effect. When the execution result reaches the expected control target, collect changes in crowd behavior and system operation status in the execution monitoring and feedback unit, and generate structured feedback data. When the execution effect is detected to deviate from the expectation or an anomaly occurs, trigger the execution strategy adjustment mechanism to correct the control strategy through local replanning or manual intervention.

[0092] Step d5: In the online learning and model update unit, the execution feedback data is sent back to the perception and prediction module to update the risk prediction model and decision-making strategy, thereby forming a closed-loop control system of continuous adaptive evolution of prediction, decision-making, execution, and feedback.

[0093] In one specific embodiment of the present invention: First, an emergency response system for abnormal personnel behavior based on group risk state transition prediction and self-learning decision-making is deployed in the emergency management platform or intelligent operation system of an integrated transportation hub. This system connects to multi-source data interfaces such as video surveillance systems, passenger flow detection systems, equipment operation monitoring systems, and event alarm systems, constructing a real-time perception environment covering spatial units such as station halls, platforms, passageways, and entrances / exits. Based on this, the group perception and interaction modeling module performs time synchronization and spatial alignment processing on the input multi-source data. Through crowd detection and state extraction, it generates basic crowd state information such as density, speed, and aggregation characteristics. Based on the reliability judgment of the perception data, it constructs a dynamic group interaction model and a group risk state representation, outputting the risk state as standardized interface data. Subsequently, the risk state transition prediction module caches and encodes historical risk state sequences based on a sliding time window. It infers the risk evolution trend within several future time steps using a time-series prediction model and performs stability assessment and necessary correction processing on the prediction results using a prediction confidence judgment mechanism, generating a structured risk level and transition trend output. After obtaining the prediction results, the self-learning decision-making module maps the future risk state and the spatial constraints of the transportation hub into a decision state space. It generates candidate emergency response strategies through a risk-driven target generation mechanism and a reinforcement learning strategy optimization process, and determines the optimal executable strategy based on safety constraint verification. Finally, the spatial execution and feedback module converts the response strategy into specific control commands, dynamically reconstructing spatial functions and providing crowd guidance control for passageways, areas, and entrances / exits. Simultaneously, it continuously monitors the execution effect and collects crowd behavior feedback data, which is then fed back to the perception and prediction module for online model updates and strategy optimization. This forms a closed-loop adaptive emergency response system integrating perception, prediction, decision-making, and execution, enabling real-time response and dynamic risk suppression of abnormal behavioral events within the integrated transportation hub.

[0094] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A comprehensive traffic hub personnel abnormal behavior emergency disposal system based on group risk state transition and self-learning decision, characterized in that, include: The modules include group perception and interaction modeling, risk state transition prediction, self-learning decision-making, and spatial execution and feedback. The group perception and interaction modeling module is used to receive real-time data from inside the integrated transportation hub and external events and environmental information, perform multi-source fusion perception of abnormal behavior events and the state of the crowd within their range of influence, and construct a group risk state representation that reflects the interaction relationship of the crowd. The risk state transition prediction module is used to establish a time-series prediction model based on the group risk state characterization, predict and evaluate the evolution trend and risk level changes of the group risk state, and generate risk development results for future moments. The self-learning decision-making module is used to generate emergency response objectives and constraints based on risk prediction results, generate and evaluate candidate response strategies through a self-learning strategy optimization mechanism, and determine the confidence level of the strategies. When the confidence level is insufficient, a fallback simulation and manual collaboration mechanism is triggered to correct the response plan. The spatial execution and feedback module is used to execute emergency response strategies, including dynamic reconstruction of the spatial functions of transportation hubs and crowd guidance and control, and to continuously collect feedback information on crowd behavior during the execution process. The feedback information is then sent back to the risk state transition prediction module to update the risk prediction model and decision-making strategy, thereby forming a closed-loop collaborative operation system of continuous adaptive evolution of perception, prediction, decision-making, execution, and feedback. The risk state transition prediction module performs the following steps: Step b1: Receive the group risk state vector and interaction graph structure information at the current and historical times, construct a risk state time series cache through a sliding time window, and perform sequence alignment and consistency processing on the data at different time steps; the group risk state vector and interaction graph structure information are obtained by encoding the group risk state representation; Step b2: Jointly encode the risk state evolution features and spatial correlation in the temporal feature encoding unit, and form a temporal embedding representation suitable for the prediction model by extracting the risk change rate, regional coupling strength and structural topology features; Step b3: In the group risk transition prediction unit, multi-step forward inference is performed on the encoded time-series features based on the time-series graph neural network to predict the risk state distribution within several future time steps, and the prediction uncertainty and confidence index are calculated simultaneously. Step b4: Determine the prediction confidence level. When the prediction result meets the confidence level requirements, generate a structured risk level output in the risk level assessment unit through risk classification mapping and trend stability analysis. When the prediction confidence level is insufficient or abnormal drift is detected, trigger the prediction correction mechanism to correct the prediction result through historical risk pattern matching and robust compensation calculation. Step b5: The risk transition trend generation unit comprehensively outputs the future risk state estimation results, risk transition direction and diffusion trend information, and sends the results as standardized prediction interface data to the self-learning decision module to drive the generation and optimization of subsequent emergency response strategies, thereby achieving forward-looking risk prediction support for abnormal behavior events.

2. The integrated transportation hub personnel abnormal behavior emergency handling system according to claim 1, characterized in that, The group perception and interaction modeling module performs the following steps: Step a1: Receive multi-source data stream input from the hub, including video surveillance data, turnstile passenger flow data, escalator and equipment operating status and abnormal event information, and perform time synchronization and spatial alignment processing on the above data to form a unified data input basis; Step a2: Extract and estimate the crowd density, movement speed, direction of movement, and local aggregation characteristics in the affected area through the crowd state perception unit, and construct a basic feature set reflecting the current movement state of the crowd; Step a3: Determine the completeness and reliability of the perceived data. When the perceived data meets the quality requirements, construct a dynamic crowd interaction relationship model through the crowd interaction modeling unit to model the crowd interaction behavior. When the perceived data is missing or occluded, execute the statistical estimation method through the robust state estimation unit to maintain the continuity and stability of the crowd state estimation. Step a4: Generate a group risk state representation, comprehensively encode it through multi-dimensional risk indicators to form a structured group risk state vector, and simultaneously output the corresponding interaction graph structure information; Step a5: Output the group risk state vector and interaction graph structure information as standardized interface data to the risk state transition prediction module to drive subsequent time-series risk prediction and intelligent decision-making calculation, thereby providing a highly reliable population perception basis for dynamic emergency response to abnormal behavior events.

3. The emergency response system for abnormal personnel behavior in integrated transportation hubs according to claim 1, characterized in that, The self-learning decision-making module performs the following steps: Step c1: Receive the predicted future risk state distribution and its evolution trend, and construct a decision state space by combining the spatial structure constraints and operation rules of the transportation hub. Encode the information into a decision state representation for strategy calculation. The information includes risk state, spatial accessibility, and facility capacity. Step c2: In the emergency response target generation unit, the control targets for the current stage are determined based on the risk-driven principle. The targets are jointly characterized through multi-objective constraint modeling, and a set of targets with priority relationships is formed. The targets include evacuation efficiency, congestion suppression, and operational stability. Step c3: In the self-learning strategy optimization unit, reinforcement learning is used to search and optimize candidate disposal strategies. The comprehensive performance of different strategies in terms of risk suppression effect, execution cost and system stability is evaluated through the reward function to generate a set of candidate strategies. Step c4: Perform security constraint checks on candidate strategies. When a strategy meets the security conditions, determine the optimal strategy in the strategy evaluation and selection unit based on cost-benefit analysis and stability tests. When a strategy is detected to violate security constraints or has unexecutable risks, trigger the security correction mechanism to correct the strategy through rule constraint projection or conservative strategy rollback. Step c5: The self-learning strategy optimization unit outputs structured emergency response instructions and their execution parameters, and sends the decision results to the spatial execution module. At the same time, it provides basic data support for subsequent online learning and strategy updates based on execution feedback, thereby forming a prediction-driven, self-learning optimization intelligent decision-making mechanism.

4. The emergency response system for abnormal personnel behavior in integrated transportation hubs according to claim 3, characterized in that, The spatial execution and feedback module performs the following steps: Step d1: Receive structured emergency response strategies in the decision instruction parsing and execution plan generation unit, convert the strategies into executable control sequences, and generate an execution scheduling plan in combination with transportation hub operation resources and time constraints; Step d2: In the spatial function dynamic reconstruction unit, the functional status of channels, areas and entrances and exits is adjusted in real time, and the spatial passage structure is changed by implementing corresponding measures; the implemented measures include flow control, one-way passage or zone isolation; Step d3: In the crowd guidance and system linkage control unit, crowd behavior is guided through broadcast prompts, display guidance, and signal control, and linkage control is performed with the equipment management system and security system; Step d4: Continuously monitor the execution effect. When the execution result reaches the expected control target, collect changes in crowd behavior and system operation status in the execution monitoring and feedback unit, and generate structured feedback data. When the execution effect is detected to deviate from the expectation or an anomaly occurs, trigger the execution strategy adjustment mechanism to correct the control strategy through local replanning or manual intervention. Step d5: In the online learning and model update unit, the execution feedback data is sent back to the perception and prediction module to update the risk prediction model and decision-making strategy, thereby forming a closed-loop control system of continuous adaptive evolution of prediction, decision-making, execution, and feedback.

5. The emergency response system for abnormal personnel behavior in integrated transportation hubs according to claim 2, characterized in that, In step a3, the group interaction behavior includes inter-regional flow coupling, directional conflict, and congestion propagation; the statistical estimation method includes density correction, optical flow compensation, or count correction. In step a4, the multidimensional risk indicators include congestion risk, diffusion risk, conflict risk, and stagnation risk.

6. An emergency response method for abnormal personnel behavior in integrated transportation hubs based on group risk state transition and self-learning decision-making, wherein the method is applied to the system described in any one of claims 1-5, characterized in that, include: Step 1: Receive relevant information on abnormal behavior events and spatial environment information of the integrated transportation hub, and construct an initial crowd perception state reflecting the current state of the crowd at the transportation hub; Step 2: Model and analyze the interaction behavior of the crowd within the scope of the abnormal event, taking into account the crowd density distribution, changes in movement direction, local clustering characteristics and inter-individual interaction relationships, to generate a group risk state representation to characterize the degree of impact of the abnormal event; Step 3: Introduce a risk prediction mechanism based on time-series group interaction modeling to predict and track the evolution trend of group risk state in the time dimension, and determine the risk state transition trend and the corresponding risk level change. Step 4: Based on the self-learning decision-making mechanism, the emergency response objectives and their execution priorities under the current abnormal event are adaptively generated according to the risk prediction results in Step 3, so that the response strategy can be dynamically optimized and adjusted as the risk status of the population changes. Step 5: Generate a spatial function dynamic reconstruction strategy and a crowd guidance strategy that match the disposal objectives of Step 4. By adjusting the functional attributes and traffic rules of passages, areas or entrances and exits within the transportation hub in real time, orderly guidance of crowd behavior and risk suppression can be achieved. Step 6: Execute the emergency response strategy generated in Step 5, and continuously obtain feedback information on changes in population behavior and the effectiveness of the response. Use the feedback to update the risk prediction model through an online learning mechanism, thereby forming a closed-loop operation mechanism in which prediction, decision-making and execution are coupled and continuously adaptively evolved during the emergency response to abnormal events.

7. The emergency response method for abnormal personnel behavior in integrated transportation hubs according to claim 6, characterized in that, In step 1, the relevant information of the abnormal behavior event includes the type of abnormal behavior event, the location of occurrence, and the time characteristics.

Images