CN-121980316-A - Subway carriage passenger abnormal behavior perception method and system based on self-adaptive decision

CN121980316ACN 121980316 ACN121980316 ACN 121980316ACN-121980316-A

Abstract

The invention discloses a subway carriage passenger abnormal behavior perception method and system based on self-adaptive decision, and relates to the technical field of urban rail transit operation safety, wherein the method comprises the steps of dispersing a subway carriage into two-dimensional grid units and constructing a passenger behavior state set, and iteratively updating to obtain abnormal behavior space distribution under each time step; setting a manager and an action set, constructing an observation vector by combining the distribution and the manager position, inputting a PPO decision model to determine a target action, acquiring instant feedback information after execution and updating a strategy, realizing abnormal behavior self-adaptive sensing and decision, and outputting an intervention decision result. The method solves the technical problems that the traditional method is difficult to dynamically capture the propagation rule of the abnormal state of the crowd in the closed public space and cannot realize accurate and timely intervention decision, achieves the technical effects of dynamically grasping the evolution situation of the abnormal state of the crowd, autonomously optimizing the intervention mode and the opportunity, and improving the accuracy and the timeliness of the management and control of the abnormal state of the closed public space.

Inventors

DENG QING
DAI WENTAO
YU FENG
JIANG HUILING

Assignees

北京科技大学
上海交通大学

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (8)

1. The subway carriage passenger abnormal behavior perception method based on the self-adaptive decision is characterized by comprising the following steps of: Dispersing the internal space of a subway carriage into two-dimensional grid units, and constructing a passenger behavior state set on the two-dimensional grid units; Based on a neighborhood interaction rule and behavior state transition parameters, carrying out iterative updating on the two-dimensional grid units to obtain abnormal behavior space distribution of passengers in the subway carriage under each time step; setting a manager object and defining an action set of the manager object; at each time step, constructing an observation vector for self-adaptive decision according to the abnormal behavior space distribution of the passengers and the position information of the manager objects; inputting the observation vector into an adaptive decision strategy model, outputting corresponding action probability distribution and determining a target action; executing the target action in a simulation environment, updating the object state of a manager to obtain instant feedback information, carrying out strategy updating on the self-adaptive decision strategy model based on the instant feedback information, and carrying out self-adaptive sensing and decision on the abnormal behavior of the passengers of the subway carriage to obtain an abnormal behavior intervention decision result.
2. The adaptive decision-based subway car passenger abnormal behavior perception method according to claim 1, wherein the set of passenger behavior states comprises: The passenger behavior state set at least comprises normal operation behaviors, controllable abnormal behaviors and disruptive upgrading behaviors; The normal operation behavior corresponds to a passenger in an abnormal risk-free stable state conforming to subway operation specifications, the controllable abnormal behavior corresponds to a transition state in which the passenger is in an observable abnormal state without destroying order and the occurrence probability of the abnormal behavior is increased, and the disruptive upgrading behavior corresponds to a state in which the abnormal behavior of the passenger has disturbance to order or safety and the risk exceeds a preset threshold; The passenger behavior state is comprehensively judged based on the passenger position density characteristics, the neighborhood behavior influence characteristics and the state transition rules.
3. The adaptive decision-based subway car passenger abnormal behavior perception method according to claim 1, wherein obtaining the passenger abnormal behavior spatial distribution of the subway car under each time step comprises: the behavior state transition parameters at least comprise one or more of neighborhood impact strength parameters, behavior recovery parameters, behavior decay parameters and behavior fatigue parameters; At each time step, based on a preset neighborhood interaction rule and in combination with the behavior state transition parameters, carrying out state update on each target grid unit in the two-dimensional grid units, wherein the behavior state of each target grid unit at the next time step is determined according to the grid unit proportion in the neighborhood of the target grid unit, which is in a disruptive upgrade type behavior, and in combination with the behavior state transition parameters; and forming the abnormal behavior space distribution of passengers in the subway carriage under the corresponding time step based on the behavior state of each target grid unit in the next time step.
4. The adaptive decision-based subway car passenger abnormal behavior perception method according to claim 1, wherein the action set of the manager object comprises: The action set of the manager object is a discrete action set, and the discrete action is obtained by combining and encoding a mobile action index and an intervention action index; Wherein the movement action index corresponds to at least one or more of up, down, left, right, and stop.
5. The adaptive decision-based subway car passenger abnormal behavior perception method according to claim 4, wherein the intervention action index comprises: the intervention action index comprises at least a local behavior intervention and a global broadcast intervention; the local behavior intervention acts on a preset local window around the manager object, and abnormal behavior grades of grid units in the preset local window are reduced according to a first preset probability; The global broadcasting intervention acts on a plurality of two-dimensional grid units in the carriage range, and the proportion of the disruptive upgrading behavior is reduced or the behavior recovery probability is improved according to the second preset probability.
6. The adaptive decision-based subway car passenger abnormal behavior perception method according to claim 1, wherein the observation vector comprises: The proportion characteristics of normal operation behaviors, controllable abnormal behaviors and disruptive upgrading behaviors in the global range of the carriage; the density characteristics of normal operation behavior controllable abnormal behavior and disruptive upgrading behavior in a preset local neighborhood window with the manager object as a center; and a position coordinate feature of the manager object in an interior region of the subway car.
7. The adaptive decision-based subway car passenger abnormal behavior perception method according to claim 1, wherein the instant feedback information comprises: the instant feedback information at least comprises one or more of abnormal behavior state variable quantity, abnormal behavior proportion maintaining item, intervention cost item and risk penalty item; wherein the risk penalty term is determined based on a comparison of a jammer escalation behavioral proportion to a preset risk threshold, and when the jammer escalation behavioral proportion exceeds the preset risk threshold, And introducing a preset punishment value or a punishment value calculated according to a preset punishment function into the instant feedback information based on the risk punishment item.
8. An adaptive decision-based subway car passenger abnormal behavior sensing system for implementing the adaptive decision-based subway car passenger abnormal behavior sensing method of any one of claims 1 to 7, the system comprising: the behavior state set construction module is used for dispersing the internal space of the subway carriage into two-dimensional grid units and constructing a passenger behavior state set on the two-dimensional grid units; the abnormal behavior space distribution acquisition module is used for carrying out iterative updating on the two-dimensional grid units based on neighborhood interaction rules and behavior state transition parameters to obtain abnormal behavior space distribution of passengers in the subway carriage under each time step; the action set definition module is used for setting a manager object and defining an action set of the manager object; the observation vector construction module is used for constructing an observation vector for self-adaptive decision according to the space distribution of the abnormal behaviors of the passengers and the position information of the manager object at each time step; The target action acquisition module is used for inputting the observation vector into the self-adaptive decision strategy model, outputting corresponding action probability distribution and determining a target action; The intervention decision result acquisition module is used for executing the target action in the simulation environment, updating the object state of the manager to obtain instant feedback information, carrying out policy updating on the self-adaptive decision policy model based on the instant feedback information, and carrying out self-adaptive perception and decision on the abnormal behavior of the passengers in the subway carriage to obtain an abnormal behavior intervention decision result.

Description

Subway carriage passenger abnormal behavior perception method and system based on self-adaptive decision Technical Field The invention relates to the technical field of urban rail transit operation safety, in particular to a subway carriage passenger abnormal behavior perception method and system based on self-adaptive decision. Background The operation safety of the sealed public space such as subway carriages is critical, and the abnormal behaviors of passengers are easy to spread rapidly, so that secondary disasters such as crowding, panic and the like are caused. The prior art relies on fixed rule intervention or manual monitoring to play a certain role in a stable scene, but with the improvement of safety control requirements, obvious limitations are exposed when the safety control system is applied to subway carriages. Due to the characteristics of high-density crowd and strong local interaction of subway carriages, the traditional method cannot dynamically sense the spatial distribution and propagation situation of abnormal behaviors, so that data are incomplete, decision is delayed, accurate intervention is difficult to realize, and the requirement for efficiently managing and controlling the abnormal behaviors in a closed public space cannot be met. Disclosure of Invention The application provides a subway carriage passenger abnormal behavior perception method and system based on self-adaptive decision, which solve the technical problems that the traditional method is difficult to dynamically capture the propagation rule of crowd abnormal states in a closed public space and cannot realize accurate and timely intervention decision. The method comprises the steps of dispersing the inner space of a subway carriage into two-dimensional grid units, constructing a passenger behavior state set on the two-dimensional grid units, carrying out iterative updating on the two-dimensional grid units based on neighborhood interaction rules and behavior state transition parameters to obtain passenger abnormal behavior spatial distribution of the subway carriage under each time step, setting manager objects, defining an action set of the manager objects, constructing an observation vector for self-adaptive decision according to the passenger abnormal behavior spatial distribution and the position information of the manager objects in each time step, inputting the observation vector into a self-adaptive decision strategy model, outputting corresponding action probability distribution and determining target actions, executing the target actions in a simulation environment, updating the manager object states to obtain instant feedback information, carrying out strategy updating on the self-adaptive decision strategy model based on the instant feedback information, and carrying out self-adaptive sensing and decision on abnormal behaviors of the subway carriage to obtain abnormal behavior intervention decision results. The application provides a subway carriage passenger abnormal behavior perception system based on self-adaptive decisions, which comprises a behavior state set construction module, an abnormal behavior space distribution acquisition module, an action set definition module, an observation vector construction module, a target action acquisition module and an intervention decision result acquisition module, wherein the behavior state set construction module is used for dispersing the internal space of a subway carriage into two-dimensional grid units, constructing a passenger behavior state set on the two-dimensional grid units, the abnormal behavior space distribution acquisition module is used for carrying out iterative updating on the two-dimensional grid units based on neighborhood interaction rules and behavior state transition parameters to obtain the passenger abnormal behavior space distribution of the subway carriage under each time step, the action set definition module is used for setting manager objects and defining action sets of the manager objects, the observation vector construction module is used for constructing an observation vector for self-adaptive decisions according to the passenger abnormal behavior space distribution and the position information of the manager objects, the target action acquisition module is used for inputting the observation vector into a self-adaptive decision strategy model, outputting corresponding action probability distribution and determining target actions, the intervention decision result acquisition module is used for executing the target actions in a simulation environment, updating the manager object states to obtain instant feedback information, updating the self-adaptive decision strategy and carrying out self-adaptive decision strategy and self-adaptive decision strategy based on the instant decision strategy. One or more technical schemes provided by the application have at least the following technical effects or advantages: According to the method,