Dangerous goods operation behavior safety management and control system based on localized AI model

Abstract

The present application relates to the technical field of dangerous goods operation safety management and control, in particular to a dangerous goods operation behavior safety management and control system based on a localized AI model, comprising a data modeling unit, a risk monitoring unit, a risk inference unit, a risk assessment and hierarchical control unit, a result output and feedback unit, wherein the risk inference unit is used to start a space-time coupled multi-behavior chain risk inference engine under the local reinforcement inference state, and to perform behavior-time-space correlation analysis on the continuous behavior sequence of the relevant target; the risk inference unit is used to start a space-time coupled multi-behavior chain risk inference engine under the local reinforcement inference state, and to perform behavior-time-space correlation analysis on the continuous behavior sequence of the relevant target; the risk assessment and hierarchical control unit is used to perform risk assessment and hierarchical control on the relevant target; and the result output and feedback unit is used to output and feed back the result. The present application performs preliminary screening on the weak risk behavior in the presequence on the basis of normal monitoring, implements local reinforcement inference on the relevant area when triggering the weak risk event, continuously tracks the target and develops behavior, time and space correlation analysis and accident precursor chain identification, realizes focused analysis on the risk evolution area and the relevant target, and thus realizes dynamic determination and hierarchical control of the dangerous goods operation site risk.

Description

Technical Field

[0001] This invention relates to the field of hazardous materials operation safety management technology, and more specifically, to a hazardous materials operation behavior safety management system based on a localized AI model. Background Technology

[0002] With the continuous growth of dangerous goods container throughput at ports, the requirements for safety management during the storage, transshipment, and on-site operations of dangerous goods are increasing. Dangerous goods container operation areas typically contain multiple dynamic targets simultaneously, including personnel, equipment, and dangerous goods containers. Furthermore, the area boundaries, access routes, and operational activities are complex and intertwined. Abnormal behaviors such as unauthorized personnel entry, abnormal equipment approach, or unauthorized lingering can easily trigger safety accidents. Therefore, how to combine information on the port yard's spatial environment and operational behavior to conduct timely and effective risk identification and control during dangerous goods operations has become a pressing technical problem to be solved in this field.

[0003] In existing technologies, some solutions have been developed to assess and predict the risks of dangerous goods transportation or storage in port areas. For example, Chinese patent CN202410138563.0 discloses a system and method for assessing the safety risks of dangerous goods containers during loading, unloading, and storage in port areas. The method includes: acquiring external environmental data associated with the cargo container and internal lithium battery status data; generating a directed graph of the cargo transportation path based on the external environmental data; correcting and reconstructing a lithium battery early warning model hidden behind interference data based on deep learning and computer vision; executing the lithium battery early warning model to obtain the real-time transportation risk assessment result of the cargo container. Another example is Chinese patent CN202310498306.3, which discloses a method for predicting the safety risks of dangerous goods containers based on port berth congestion. The method includes: extracting relevant information about dangerous goods containers based on cargo container transportation data; constructing a dangerous goods container transportation chain network based on the relevant information and complex network theory; extracting target features from the dangerous goods container transportation chain network; and inputting the target features into a pre-constructed port berth congestion prediction model to obtain a first output result for predicting the safety risks of dangerous goods containers.

[0004] However, the aforementioned existing technologies focus primarily on environmental risk assessment, internal status early warning, or system-level congestion anomaly prediction during the transportation of dangerous goods containers. Their emphasis is mainly on the transportation link, external environment, or the condition of the goods themselves, lacking a detailed characterization of the on-site interactions between personnel, equipment, and dangerous goods containers within the container storage area. In particular, for weakly risky preceding behaviors such as abnormal approach, abnormal stoppage, unplanned equipment approach, and path deviation, there is a lack of a safety control mechanism capable of dynamically reasoning based on regional spatial distribution, continuous target tracking, and the progressive evolution of behaviors. This makes it difficult to promptly identify accident precursor chains that gradually evolve from multiple consecutive abnormal behaviors, leading to delayed risk detection and insufficient tiered response. Therefore, we propose a dangerous goods operation behavior safety control system based on a localized AI model. Summary of the Invention

[0005] The purpose of this invention is to provide a safety management and control system for hazardous materials operations based on a localized AI model, in order to solve the problems mentioned in the background art, which are that the existing technology is difficult to accurately identify and dynamically infer the on-site interaction behavior between operators, operating equipment and hazardous materials containers in the hazardous materials storage area, especially the inability to continuously track the preceding low-risk behaviors and identify the accident precursor chain by combining time, space and behavioral evolution relationship, thus resulting in untimely risk detection and insufficient targeted hierarchical management.

[0006] To address the aforementioned technical problems, the present invention aims to provide a safety management system for hazardous materials operations based on a localized AI model, comprising: The data modeling unit is used to perform spatial modeling of the dangerous goods container storage area in the port yard. It accesses the basic information and location information of the operators, operating equipment and dangerous goods containers, and divides the storage area into regions according to the stacking layout, passage path and operation boundary to construct a spatial distribution model that includes dangerous goods container stacking area, passage area and operation boundary area. The risk monitoring unit analyzes the activities of personnel, equipment, and dangerous goods containers in the vicinity of the port in real time based on an artificial intelligence model deployed locally at the port. Under normal monitoring conditions, it performs preliminary screening of weak risk events in each area to identify early abnormal behaviors such as abnormal approach or stay of personnel, unplanned approach or deviation of equipment from its path. When a weak risk event occurs in any area, it triggers that area to enter a local enhanced reasoning state and continuously tracks the relevant targets. The risk reasoning unit is used to activate a spatiotemporally coupled multi-behavior chain risk reasoning engine under the local enhanced reasoning state, perform behavior-time-space correlation analysis on the continuous behavior sequence of the relevant targets, and identify accident precursor chains formed by multiple abnormal behaviors that occur continuously and are spatially correlated in the vicinity of the dangerous goods container based on the progressive relationship of risk behaviors; when the subsequent behavior and the preceding event satisfy the temporal continuity, spatial proximity and evolutionary correlation relationship, the risk level is progressively increased; when no subsequent correlated behavior occurs or the target leaves the risk range, the local enhanced reasoning state is exited and normal monitoring is restored. The risk assessment and hierarchical control unit determines the risk level of each area in the storage area based on the risk status output by the risk reasoning unit, and triggers corresponding hierarchical control operations based on the determination results. The hierarchical control operations include outputting early warning information, triggering on-site prompts, or sending control commands. The result output and feedback unit is used to output the processing results of the risk assessment and hierarchical control unit, and send the risk area, risk level and corresponding behavioral chain information to the external system or management platform.

[0007] As a further improvement to this technical solution, the data modeling unit includes an information acquisition and access module, a region parsing module, a coordinate mapping module, and a model generation module, wherein: The information collection and access module is used to collect the identity information, category information, operation attribute information, and location information of operators, operating equipment, and dangerous goods containers; The area analysis module performs area analysis and area division on the dangerous goods container storage area of ​​the port yard based on the stacking layout, passage path and operation boundary information, forming dangerous goods container stacking area, passage area and operation boundary area. The coordinate mapping module is used to perform coordinate registration of the location information of operators, operating equipment and dangerous goods containers, and to determine the regional affiliation of each target object; The model generation module generates a spatial distribution model that includes boundary information of each region and spatial distribution relationships of target objects, based on the region division results and region affiliation relationships.

[0008] As a further improvement to this technical solution, the risk monitoring unit includes a real-time analysis module, a weak risk initial screening module, a status triggering module, and a continuous tracking module, wherein: The real-time analysis module analyzes the activities of personnel, equipment and dangerous goods containers in the vicinity of the port based on an artificial intelligence model deployed locally, and obtains the behavioral status information of each target object in the storage area. The low-risk screening module is used to perform preliminary screening of low-risk events in each area under normal monitoring conditions, based on the behavioral status information of each target object, in order to identify early abnormal behaviors such as abnormal approach of personnel, abnormal stay, unplanned approach of equipment, or deviation from the path. The state triggering module is used to trigger the region to enter a local enhanced reasoning state when a preceding weak risk event occurs in any region. The continuous tracking module is used to continuously track target objects related to preceding weak-risk events in the local enhanced reasoning state.

[0009] As a further improvement to this technical solution, the preliminary screening process for weak risk events in the weak risk screening module includes the following steps: S22.1 Based on the behavioral status information of each target object output by the real-time analysis module, extract the location, trajectory, direction of movement, area affiliation, and dwell status information of operators, operating equipment, and dangerous goods containers within the current monitoring area; S22.2. Based on the spatial proximity between the operators and the dangerous goods container, identify whether the operators are exhibiting abnormal approach behavior. Specifically, when an operator enters the dangerous goods container, a preset safe distance threshold is established. When the corresponding adjacent range is considered, it is determined that there is abnormal proximity behavior; S22.3. Based on the continuous stay of personnel in the vicinity of the dangerous goods container, extract the duration of stay. and the duration of stay Compared with a preset dwell time threshold, when the dwell time... If the preset dwell time threshold is exceeded, abnormal dwelling behavior is determined. S22.4. Based on the planned operating area, current position and direction of movement of the operating equipment, identify whether the operating equipment deviates from the planned operating range and makes an unplanned approach to the vicinity of the dangerous goods container, and determine whether there is any unplanned approach behavior of the equipment. S22.5. Based on the deviation between the actual movement trajectory of the operator or equipment and the preset travel path, a path deviation degree is calculated. and the path deviation The path deviation is compared with a preset path deviation threshold. When the path deviation exceeds the preset path deviation threshold, it is determined that there is path deviation behavior; S22.6 Summarize the identification results of abnormal approach behavior, abnormal stay behavior, unplanned equipment approach behavior and path deviation behavior to form the preceding weak risk events for the corresponding area, and output the preceding weak risk events to the status triggering module.

[0010] As a further improvement to this technical solution, the state triggering module includes a region identification submodule, a trigger determination submodule, and a state switching submodule, wherein: When the region identification submodule detects a preceding weak risk event, it identifies the target region where the preceding weak risk event occurred and determines the target region as the region to be strengthened for analysis. The trigger determination submodule determines whether the region to be strengthened for analysis meets the local strengthening reasoning trigger conditions based on the region location, event type, and target object corresponding to the preceding weak risk event. When the region to be reinforced for analysis meets the local reinforcement reasoning triggering condition, the state switching submodule switches the region to be reinforced for analysis from the normal monitoring state to the local reinforcement reasoning state, so as to trigger the risk reasoning unit to perform continuous behavior sequence analysis on the relevant target objects.

[0011] As a further improvement to this technical solution, the risk reasoning unit includes a behavior sequence construction module, a spatiotemporal correlation analysis module, and an accident precursor chain reasoning module, wherein: The behavior sequence construction module is used to sort the relevant target object behavior information output by the continuous tracking module by time and aggregate events under the local reinforcement reasoning state to form a continuous behavior sequence of the corresponding target object. The spatiotemporal correlation analysis module is used to perform behavior-time-space correlation analysis on each behavior event in the continuous behavior sequence and calculate the spatiotemporal correlation degree between behaviors. And identify spatiotemporal relationships with temporal continuity and spatial proximity; The accident precursor chain reasoning module identifies a continuous chain of abnormal behaviors in the vicinity of a dangerous goods container based on the spatiotemporal correlation, and performs accident precursor chain reasoning based on the evolutionary relationship between each behavioral event in the chain, while simultaneously calculating the risk progression value. This is to achieve state control that allows for progressive risk levels and locally enhanced reasoning.

[0012] As a further improvement to this technical solution, the behavior sequence construction module includes a behavior event extraction submodule, an event sorting submodule, and a sequence generation submodule, wherein: The behavior event extraction submodule extracts abnormal behavior events from the relevant target object trajectory information and behavior status information output by the continuous tracking module, and records the corresponding occurrence time, occurrence location and behavior category; The event sorting submodule sorts the abnormal behavior events according to the order of their occurrence time, forming a set of behavior events with temporal continuity. The sequence generation submodule constructs a continuous behavior sequence of the corresponding target object based on the set of behavior events, and outputs the continuous behavior sequence to the spatiotemporal correlation analysis module for subsequent behavior-time-space correlation analysis.

[0013] As a further improvement to this technical solution, the behavior-time-space correlation analysis process of the spatiotemporal correlation analysis module includes the following steps: S32.1 Receive the continuous behavior sequence output by the behavior sequence construction module, and extract the occurrence time, spatial location and behavior type information corresponding to each behavior event; S32.2 Extract behavioral temporal correlation features based on the time interval between occurrences of adjacent behavioral events in a continuous behavioral sequence, and form corresponding temporal correlation indicators; S32.3 Extract spatial correlation features of behaviors based on the spatial relationship between the location of each behavioral event and the adjacent area of ​​the dangerous goods container, and form corresponding spatial correlation indicators; S32.4. Conduct a comprehensive analysis of the aforementioned time-related indicators and spatial-related indicators to form the spatiotemporal correlation degree between behaviors. ; S32.5, Based on the aforementioned spatiotemporal correlation degree Determine whether the behavioral events in a continuous behavioral sequence meet the behavioral-temporal-spatial correlation conditions, and output the behavioral events that meet the correlation conditions to the accident precursor chain inference module.

[0014] As a further improvement to this technical solution, the accident precursor chain reasoning and risk progression process of the accident precursor chain reasoning module includes the following steps: S33.1 Receive the combination of behavioral events that meet the behavioral-time-space correlation conditions output by the spatiotemporal correlation analysis module, and extract the occurrence time, spatial location and behavioral type information corresponding to each behavioral event; S33.2. Construct a chain-like association of the behavioral events based on the chronological order of their occurrence to form an abnormal behavior chain; S33.3. Based on the spatial relationship between the location of each behavioral event in the abnormal behavior chain and the adjacent area of ​​the dangerous goods container, the abnormal behavior chain is screened to identify and form an accident precursor chain. S33.4 Calculate the risk progression value based on the behavioral type, order of occurrence, and correlation strength of each behavioral event in the accident precursor chain. ; S33.5, Based on the aforementioned risk progression value The risk level is progressively upgraded, and the upgraded risk level is output to the risk assessment and hierarchical management unit. S33.6. If no subsequent related behaviors or related target objects are detected to have moved out of the vicinity of the dangerous goods container, interrupt the progression of the accident precursor chain and control the system to exit the local enhanced reasoning state to restore normal monitoring.

[0015] As a further improvement to this technical solution, the risk assessment and hierarchical control unit includes a risk level determination module, a hierarchical control module, and a processing result output module, wherein: The risk level determination module determines the risk level of each area of ​​the storage area based on the risk status output by the risk reasoning unit. The hierarchical control module triggers corresponding hierarchical control operations based on the judgment result. The hierarchical control operations include outputting early warning information, triggering on-site prompts, or sending control commands. The processing result output module is used to output the processing result of the hierarchical control operation and output the processing result to the result output and feedback unit.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention performs preliminary screening of preceding low-risk behaviors based on routine monitoring, and triggers the corresponding area to enter a local enhanced reasoning state when a preceding low-risk event occurs. It continuously tracks relevant targets and conducts behavior-time-space correlation analysis and accident precursor chain identification. It can focus reasoning analysis on the risk evolution area and related targets, realize dynamic judgment and hierarchical control of risks at hazardous materials operation sites, and solve the problems of existing technologies that are difficult to accurately identify and dynamically reason about on-site interactive behaviors, cannot continuously track preceding low-risk behaviors and identify accident precursor chains, and lack targeted enhanced analysis mechanisms for local risk areas, resulting in untimely risk detection and insufficient targeted hierarchical control. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the system framework of the present invention; The meanings of the labels in the diagram are as follows: 1. Data Modeling Unit; 11. Information Acquisition and Access Module; 12. Region Analysis Module; 13. Coordinate Mapping Module; 14. Model Generation Module; 2. Risk monitoring unit; 21. Real-time analysis module; 22. Initial screening module for weak risks; 23. Status triggering module; 24. Continuous tracking module; 3. Risk Reasoning Unit; 31. Behavioral Sequence Construction Module; 32. Spatiotemporal Correlation Analysis Module; 33. Accident Precursor Chain Reasoning Module; 4. Risk assessment and hierarchical control unit; 41. Risk level determination module; 42. Hierarchical control module; 43. Processing result output module; 5. Result output and feedback unit. Detailed Implementation

[0018] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, this embodiment provides a safety management system for hazardous materials operations based on a localized AI model, including: Data Modeling Unit 1 is used to perform spatial modeling of the dangerous goods container storage area in the port yard. It accesses the basic information and location information of operators, operating equipment and dangerous goods containers, and divides the storage area into regions according to the stacking layout, passage path and operation boundary, and constructs a spatial distribution model that includes dangerous goods container stacking area, passage area and operation boundary area. In this embodiment, the data modeling unit 1 includes an information acquisition and access module 11, a region parsing module 12, a coordinate mapping module 13, and a model generation module 14, wherein: The information collection and access module 11 is used to collect the identity information, category information, operation attribute information, and location information of operators, operating equipment, and dangerous goods containers; Specifically, identity information can be used to distinguish different operators, different operating equipment, and different dangerous goods containers; category information can be used to characterize the type of target object; operation attribute information can be used to characterize the job attributes of operators, the purpose attributes of operating equipment, and the operation-related attributes corresponding to dangerous goods containers; location information can be used to characterize the current location of each target object in the dangerous goods container storage area of ​​the port yard. After the information collection and access module 11 receives the above information, it can summarize and organize data from different sources to form basic target object data that can be used for subsequent regional analysis and coordinate mapping.

[0020] The area analysis module 12 analyzes and divides the storage area based on the layout of the dangerous goods container storage area in the port yard, the access routes and the operation boundary information, forming dangerous goods container storage area, access area and operation boundary area. Specifically, the stacking layout characterizes the distribution of dangerous goods containers in the yard, the access path characterizes the pre-defined access routes for personnel or equipment within the storage area, and the operation boundary information defines the boundary range of dangerous goods operations. Based on the above information, the area analysis module 12 can divide the dangerous goods container storage area in the port yard into different functional areas, enabling subsequent systems to differentiate and process target object activities based on area attributes.

[0021] The coordinate mapping module 13 is used to perform coordinate registration of the location information of operators, operating equipment and dangerous goods containers, and to determine the regional affiliation of each target object; Specifically, the coordinate mapping module 13 can uniformly map the location information accessed by the information acquisition access module 11 to the spatial coordinate system corresponding to the dangerous goods container storage area of ​​the port yard, so that the location data of different target objects can be expressed under the same spatial reference.

[0022] Subsequently, based on the region division results obtained by the region analysis module 12, the coordinate mapping module 13 determines which region each target object is currently located in among the dangerous goods container stacking area, passage area, or operation boundary area, thereby establishing the affiliation relationship between the target object and the region.

[0023] The model generation module 14 generates a spatial distribution model that includes boundary information of each region and spatial distribution relationship of the target object based on the region division results and region affiliation relationship.

[0024] Specifically, the spatial distribution model includes, on the one hand, the boundary information of the dangerous goods container stacking area, the passage area, and the operation boundary area; on the other hand, it also includes the distribution status and correspondence of operators, operating equipment, and dangerous goods containers in each area. The spatial distribution model generated by the model generation module 14 can intuitively reflect the spatial relationship between various target objects and each operation area within the dangerous goods container stacking area of ​​the port yard, providing a foundation for subsequent real-time analysis of activities in the vicinity of dangerous goods containers, identification of preceding weak risk events, and local reinforcement reasoning.

[0025] Furthermore, in this embodiment, the information acquisition and access module 11, the area analysis module 12, the coordinate mapping module 13, and the model generation module 14 can operate sequentially according to a preset process: First, the information acquisition and access module 11 acquires relevant information about operators, operating equipment, and dangerous goods containers; then, the area analysis module 12 performs area analysis and division of the dangerous goods container storage area in the port yard; next, the coordinate mapping module 13 performs coordinate registration on the location information of each target object and determines its area affiliation; finally, the model generation module 14 combines the area division results and area affiliation relationships to generate the corresponding spatial distribution model. Thus, the data modeling unit 1 can complete area modeling and target mapping in the dangerous goods storage operation scenario, providing a unified spatial modeling result for subsequent functional units of the system.

[0026] Risk Monitoring Unit 2, based on an AI model deployed locally at the port, performs real-time analysis of the activities of personnel, equipment, and dangerous goods containers in the vicinity. Under normal monitoring conditions, it performs preliminary screening of pre-emptive low-risk events in each area to identify early abnormal behaviors such as abnormal personnel approach, abnormal lingering, unplanned equipment approach, or path deviation. When a pre-emptive low-risk event occurs in any area, it triggers that area to enter a local enhanced inference state and continuously tracks related targets. Risk Monitoring Unit 2 includes a real-time analysis module 21, a low-risk preliminary screening module 22, a state triggering module 23, and a continuous tracking module 24. Data transmission between these modules can be achieved via bus, shared memory, or data interface, or they can be integrated and deployed collaboratively on the same local computing platform. In this embodiment, the real-time analysis module 21 performs real-time analysis of the activities of operators, operating equipment, and dangerous goods containers in the vicinity of the port based on an artificial intelligence model deployed locally, and obtains the behavioral status information of each target object in the storage area. The local deployment here refers to the direct deployment of the artificial intelligence model on the edge server, local computing node, or local area network computing platform at the port site, so as to process the on-site video stream, positioning stream, or sensing data stream locally, thereby meeting the requirements of the dangerous goods operation area for real-time performance and independent on-site operation capability.

[0027] In practical implementation, the real-time analysis module 21 can interface with input data from camera equipment, positioning terminals, vehicle-mounted terminals, or on-site sensing devices to perform target identification, target location, and status analysis on operators, operating equipment, and dangerous goods containers. It then outputs behavioral status information for each target object at the current monitoring time, including its position, trajectory, direction of movement, area affiliation, and dwell status. Specifically, position represents the target object's current spatial coordinates, all of which are located within the same port yard spatial coordinate system; trajectory represents the target object's positional change sequence over a continuous period; direction of movement represents the target object's current orientation; area affiliation represents the monitoring area to which the target object belongs; and dwell status represents whether the target object is continuously stationed.

[0028] Furthermore, the behavioral state information output by the real-time analysis module 21 can be cached and updated according to timestamps, so that the weak risk screening module 22 can call the state data of the current time and adjacent time periods to determine the weak risk. To ensure the consistency of continuous analysis, the real-time analysis module 21 can maintain a unified identifier for the same target object, thereby forming a corresponding target state sequence at multiple consecutive time points.

[0029] It should be added that the AI ​​model deployed locally at the port can receive video frame data from camera equipment, location data from positioning terminals and vehicle-mounted terminals, and status perception data from on-site sensing devices. It then performs target identification, target location calculation, multi-source target association, trajectory update, and behavior status analysis on this data. After processing, it can output a structured target status record, which includes at least: target object identifier, timestamp, current location coordinates, historical trajectory point sequence, current direction of movement, monitoring area, and dwell state. The low-risk initial screening module 22 can perform subsequent abnormal behavior identification based on the structured target status record.

[0030] In this embodiment, the weak risk screening module 22 is used to perform preliminary screening of weak risk events in each area under normal monitoring conditions, based on the behavioral status information of each target object, to identify early abnormal behaviors such as abnormal approach of personnel, abnormal stay, unplanned approach of equipment, or deviation from the path. The weak risk screening module 22 prioritizes the rapid identification of single behavioral anomalies, and its output is used to characterize the initial risk signs that have appeared on site. However, at this time, it does not directly perform continuous behavioral chain reasoning, but instead uses it as the input basis for subsequent state triggering. The preliminary screening process of weak risk events by the weak risk screening module 22 includes the following steps: S22.1 Based on the behavioral status information of each target object output by the real-time analysis module 21, extract the location, trajectory, direction of movement, area affiliation, and dwell status information of the operators, operating equipment, and dangerous goods containers within the current monitoring area. In specific implementation, the status information of each target object at continuous sampling times can be organized into a time-series record to form a target status set within the current monitoring area. Based on this target status set, further analysis can be conducted on spatial proximity between personnel and dangerous goods containers, dwell time analysis, equipment planned area deviation analysis, and deviation analysis between actual trajectories and preset travel paths.

[0031] The position of the operator can be recorded as a two-dimensional coordinate. The location of a dangerous goods container can be represented by two-dimensional coordinates. The current position of the working equipment can be recorded as two-dimensional coordinates. All the above coordinates are located in the same port yard spatial coordinate system. The position sequence of the target object at consecutive time points can be used as trajectory information. The current direction of movement of the target object can be obtained by the position difference between adjacent time points. The dwell state can be determined based on whether the position change amplitude within a preset time window is lower than a set threshold.

[0032] In this embodiment, to facilitate unified processing of target location data from camera equipment, positioning terminals, vehicle-mounted terminals, or on-site sensing equipment, a port yard planar spatial coordinate system can be pre-established. The target image location acquired by the camera equipment can be converted into two-dimensional coordinates in the planar spatial coordinate system through a pre-calibrated mapping relationship. The location data output by the positioning terminal and the vehicle-mounted terminal are also uniformly converted to the planar spatial coordinate system. Therefore, the locations of personnel, equipment, and hazardous materials containers can be calculated for distance, analyzed for trajectory, determined for regional affiliation, and analyzed for spatiotemporal correlation under the same spatial reference.

[0033] S22.2. Based on the spatial proximity between the operators and the dangerous goods container, identify whether the operators are exhibiting abnormal approach behavior. Specifically, when an operator enters the dangerous goods container, a preset safe distance threshold is established. When the corresponding adjacent range is considered, it is determined that there is abnormal proximity behavior; Specifically, the real-time spatial distance between the workers and the dangerous goods container can be expressed as: ; in, This indicates the real-time spatial distance between the workers and the dangerous goods container; The x-coordinate represents the current position of the worker; The vertical coordinate represents the current position of the worker; The x-coordinate represents the current location of the dangerous goods container; The vertical coordinate represents the current position of the dangerous goods container.

[0034] An abnormal approach behavior is determined to exist when the following formula is met: ; in, This indicates the preset safe distance threshold for dangerous goods containers. It can be set according to the type of dangerous goods container, operating procedures, or on-site safety management requirements. It means the safe boundary distance at which personnel are allowed to approach the dangerous goods container.

[0035] Furthermore, to avoid misjudgments due to momentary detection jitter, in practice, the above-mentioned judgment conditions can be required to be met within a certain number of consecutive sampling times before an abnormal proximity behavior recognition result is output. For example, the conditions can be met for three consecutive sampling times. When this occurs, the behavior is recorded as a valid abnormal approach. The improvement of this approach lies in that it does not rely on a single instantaneous distance for direct judgment, but rather combines stability over continuous time periods for confirmation, thereby enhancing the feasibility and on-site adaptability of the initial screening for weak risks.

[0036] The calculation example is as follows: Assume that at a certain moment, the worker's position is (12,8) and the dangerous goods container's position is (15,12). ; in, This indicates that the real-time spatial distance between the workers and the hazardous materials container at that moment is 5. If the preset safety distance threshold is... Then it satisfies Therefore, it can be determined that the worker entered the dangerous goods container at the preset safe distance threshold. The corresponding neighborhood exhibits abnormal proximity behavior.

[0037] S22.3. Based on the continuous stay of personnel in the vicinity of the dangerous goods container, extract the duration of stay. and the duration of stay Compared with a preset stay duration threshold, when the stay duration If the preset dwell time threshold is exceeded, abnormal dwelling behavior is determined. Specifically, in this embodiment, the duration of stay This can be defined as the duration of time a worker remains in the vicinity of a dangerous goods container. Let the start time of entering the vicinity and meeting the stay criteria be denoted as... The time of the current state or the time of exiting the dwell state is recorded as Then the duration of stay It can be represented as: ; An abnormal dwell behavior is determined to exist when the following formula is met: ; in, This indicates the preset dwell time threshold. It can be set according to the requirements of dangerous goods operation management, and it means the maximum safe duration for personnel to stay continuously in the vicinity of dangerous goods containers.

[0038] Furthermore, the state of inactivity can be determined by the magnitude of positional changes of the target object within a continuous time window. For example, if the positional changes of personnel within a continuous time window do not exceed a set micro-displacement threshold, and they remain consistently within the vicinity of the hazardous materials container, they are considered to be in an inactivity state. The improvement of this approach lies in requiring not only that personnel be located within the vicinity, but also that they exhibit continuous inactivity characteristics, thereby distinguishing between "passing proximity" and "staying proximity," making the identification of weak risks more consistent with the logic of on-site hazardous materials supervision.

[0039] It should be added that, in this embodiment, the length can be [missing information]. The range of personnel position changes is calculated within a continuous time window, provided that the maximum distance between any two positions within that time window does not exceed a small displacement threshold. When the target remains within the vicinity of the dangerous goods container, it is determined that the target is in a stationary state. This can be represented as: ; in, This represents a continuous time window used to determine whether a target object is in a stationary state. and This represents any two sampling times within the continuous time window; Indicates the target object at the sampling time Spatial location coordinates; Indicates the target object at the sampling time Spatial location coordinates; Indicates the target object at the sampling time With sampling time The positional distance between them; This represents the maximum positional distance between any two sampling moments of the target object within the continuous time window; This indicates a preset threshold for minute displacement.

[0040] S22.4. Based on the planned operating area, current position and direction of movement of the operating equipment, identify whether the operating equipment deviates from the planned operating range and makes an unplanned approach to the vicinity of the dangerous goods container, and determine whether there is any unplanned approach behavior of the equipment. Specifically, the system first obtains the planned operating area corresponding to the equipment, then determines whether the equipment's current position exceeds the planned operating area. If the equipment exceeds the planned operating area, it further determines whether the equipment's current direction of movement is pointing towards the vicinity of the hazardous materials container. If both conditions are met, it can be determined that there is unplanned approach behavior of the equipment. Here, "unplanned approach" is not judged solely based on distance changes, but rather by combining "planned operating area constraints" and "movement trend towards the vicinity of the hazardous materials container." This design incorporates both the equipment's task boundary and its dynamic movement status into the low-risk initial screening process, making the identification of abnormal equipment behavior more aligned with the operation scheduling scenario.

[0041] Furthermore, if the current position of the equipment has deviated from the planned operation area, and its movement direction for several consecutive moments is towards the area adjacent to the dangerous goods container, the stability of the identification result can be improved, avoiding misjudgment caused by short-term turning or local obstacle avoidance.

[0042] It should be added that, in practical implementation, the current direction of movement of the working equipment can be represented as the position difference vector of the equipment at two adjacent sampling times. The direction from the current location of the equipment to the center point or nearest boundary point of the adjacent area of ​​the dangerous goods container is represented as a vector. If the included angle between the two meets a preset direction threshold condition, then the current direction of movement of the equipment is determined to be towards the area adjacent to the hazardous materials container. Correspondingly, this can be expressed as: ; in, This indicates the angle between the current direction of movement of the operating equipment and the direction of the area adjacent to the dangerous goods container; This indicates the threshold for direction determination.

[0043] S22.5. Based on the deviation between the actual movement trajectory of the operator or equipment and the preset travel path, a path deviation degree is calculated. and path deviation Compared with a preset path deviation threshold, when the path deviation is... When the path deviation exceeds the preset path deviation threshold, it is determined that there is path deviation behavior; Specifically, path deviation It can be used to characterize the degree of deviation of the actual movement trajectory of a target object from a preset travel path. Assume that the operator or equipment is continuously... The shortest distances from the actual trajectory points on each sampling point to the preset travel path are respectively Then the path deviation It can be represented as: ; in, Indicates the number of trajectory sampling points involved in the calculation; Indicates the first The shortest distance from each actual trajectory point to the preset travel path; This indicates the sequence number of the trajectory sampling point.

[0044] When the following equation is satisfied, path deviation behavior is determined to exist: ; in, This indicates that the preset path deviates from the threshold. It can be set according to the actual passage regulations of the port yard, which means the maximum degree of deviation that the actual trajectory of the target object is allowed to deviate from the preset passage path.

[0045] It is understandable that the above method of calculating path deviation uses the average deviation of multiple sampling points as the criterion, rather than taking only the maximum deviation value at a single moment. Its advantages lie in that by quantifying the overall offset of continuous trajectory segments, the impact of instantaneous detours, short-term avoidances, or local measurement noise on the results can be reduced, thus making path deviation identification more stable and more consistent with the continuous characteristics of on-site traffic behavior.

[0046] The calculation example is as follows: Suppose a target object is in The shortest distances from each consecutive sampling point to the preset travel path are respectively ,but: ; in, This indicates that the average path deviation of the target object on this continuous trajectory segment is 0.85. If a preset path deviation threshold is set... Then it satisfies Therefore, it was determined that there was path deviation behavior.

[0047] S22.6. The identification results of abnormal approach behavior, abnormal stay behavior, unplanned equipment approach behavior, and path deviation behavior are summarized to form the preceding weak risk events for the corresponding area, and the preceding weak risk events are output to the status trigger module 23. The weak risk screening module 22 can associate and record each identified early abnormal behavior with its corresponding target object, area, time of occurrence, and behavior category; if any of the above-mentioned abnormal behaviors occur in the same area during the current monitoring period, the area is marked as the area where the preceding weak risk event occurred, and a preceding weak risk event record containing area identifier, event type, associated target object, and event occurrence time is generated. Thus, the weak risk screening module 22 can transform the scattered individual weak risk behavior identification results into regional-level preceding weak risk event outputs.

[0048] In this embodiment, the state triggering module 23 is used to trigger a region to enter a local enhanced inference state when a preceding weak risk event occurs in any region. The function of the state triggering module 23 is that it does not always perform continuous inference of equal intensity in all regions, but only enhances the state of the corresponding target region after detecting a preceding weak risk event, thereby forming an enhanced analysis mechanism for regions with local risk signs. The state triggering module 23 includes a region identification submodule, a trigger determination submodule, and a state switching submodule, wherein: When a preceding weak risk event is detected, the area identification submodule identifies the target area where the preceding weak risk event occurred and designates it as an area to be strengthened for analysis. Specifically, the area identification submodule can read the area identification information from the preceding weak risk event records and identify the hazardous materials container stacking area, passage area, or operational boundary area where the weak risk event occurred as the area to be strengthened for analysis. If multiple areas experience preceding weak risk events simultaneously, the corresponding areas can be identified as multiple areas to be strengthened for analysis in parallel.

[0049] The trigger determination submodule determines whether the area to be strengthened for analysis meets the local strengthening inference trigger conditions based on the location, event type, and target object of the preceding weak risk event. These trigger conditions may include at least the following elements: first, a preceding weak risk event confirmed by the weak risk screening module 22 already exists in the area; second, the preceding weak risk event has a clearly defined target object and area affiliation; and third, the event is still within the effective monitoring period. When the above conditions are met, the area to be strengthened for analysis is determined to meet the local strengthening inference trigger conditions. The "effective monitoring period" here can be understood as a pre-set continuous monitoring window after the occurrence of the preceding weak risk event, used to ensure that the status trigger and on-site risk signs remain connected in time.

[0050] When the region to be analyzed meets the triggering conditions for local enhanced inference, the state switching submodule switches the region from normal monitoring to local enhanced inference state, thereby triggering the risk inference unit 3 to perform continuous behavioral sequence analysis on the relevant target objects. For the risk monitoring unit 2 itself, the purpose of the state switching submodule is to complete the switching of the region's state flag and the enhanced tracking marking of relevant target objects, rather than directly undertaking subsequent inference analysis. In specific implementation, the state switching submodule can write a "local enhanced inference state" identifier to the region to be analyzed and simultaneously send the identifiers of target objects that need to be tracked closely to the continuous tracking module 24, thus enabling the subsequent monitoring process to continuously unfold around the associated targets corresponding to the preceding weak risk events.

[0051] Meanwhile, the key design feature of this state triggering mechanism is that it does not uniformly increase the analysis intensity of the entire storage area, but rather switches the state of the target area locally based on the preceding weak risk events, so as to achieve a conditional transition from normal regional monitoring to enhanced regional analysis, thereby making the monitoring process more in line with the actual characteristics of "risks first appearing locally and then expanding locally" in hazardous materials operation sites.

[0052] In this embodiment, the continuous tracking module 24 is used to continuously track target objects related to preceding weak risk events in a local enhanced inference state. These related target objects include the personnel and equipment that triggered the preceding weak risk event, as well as other target objects associated with the vicinity of the hazardous materials container corresponding to the preceding weak risk event.

[0053] In practice, after receiving the target object identifier sent by the state triggering module 23, the continuous tracking module 24 continuously records the continuous position changes, trajectory updates, movement direction changes, area switching, and dwell state changes of the relevant target object, and maintains the consistency of the target object identifier across multiple consecutive moments. In this way, the continuous tracking module 24 can generate continuous and uninterrupted behavioral state update results for the relevant target object during the local reinforcement inference state.

[0054] Furthermore, during implementation, the continuous tracking module 24 can focus on tracking and updating the target object around the area where the preceding weak risk event occurred and its adjacent areas. Tracking continues as long as the target object remains within the area to be strengthened or continues to have a spatial relationship with that area; when the target object leaves the current scope of strengthened analysis, the strengthened tracking and marking of that target object is stopped. This approach ensures that the continuous tracking module 24 maintains a correspondence with the preceding weak risk event, thereby guaranteeing that the subsequent input target state always revolves around the triggered local risk scenario.

[0055] It should be added that the overall operation process of risk monitoring unit 2 is as follows: First, the real-time analysis module 21 analyzes the activities of operators, operating equipment, and the vicinity of hazardous materials containers in real time, and outputs the behavioral status information of each target object; then, under normal monitoring conditions, the weak risk screening module 22 performs abnormal approach identification, abnormal stay identification, unplanned equipment approach identification, and path deviation identification according to the above behavioral status information, and forms the corresponding pre-emptive weak risk events for the area; next, the status triggering module 23 identifies the target area where the pre-emptive weak risk event occurred, determines whether it meets the local reinforcement inference triggering conditions, and switches the area to the local reinforcement inference state when the conditions are met; finally, the continuous tracking module 24 continuously tracks the target objects related to the pre-emptive weak risk event. Through the above methods, risk monitoring unit 2 can form a closed-loop monitoring mechanism for early abnormal behavior in hazardous materials storage operation scenarios.

[0056] Risk Reasoning Unit 3 is used to activate a spatiotemporally coupled multi-behavioral chain risk reasoning engine under local enhanced reasoning state. It performs behavior-time-space correlation analysis on continuous behavioral sequences of relevant targets and identifies accident precursor chains formed by multiple abnormal behaviors occurring consecutively and spatially in the vicinity of dangerous goods containers based on the progressive relationship of risk behaviors. When subsequent behaviors satisfy the temporal continuity, spatial proximity, and evolutionary correlation with preceding events, the risk level is progressively increased. When no subsequent correlated behaviors occur or the target leaves the risk range, the local enhanced reasoning state is exited and normal monitoring is restored. Risk Reasoning Unit 3 includes a behavior sequence construction module 31, a spatiotemporal correlation analysis module 32, and an accident precursor chain reasoning module 33, wherein: In this embodiment, the behavior sequence construction module 31 is used to perform time sorting and event aggregation on the relevant target object behavior information output by the continuous tracking module 24 under the local reinforcement reasoning state, forming a continuous behavior sequence of the corresponding target object; the behavior sequence construction module 31 transforms the continuous trajectory information and behavior state information output by the continuous tracking module 24 into a structured behavior event sequence suitable for subsequent spatiotemporal correlation analysis. That is, the behavior sequence construction module 31 does not directly perform risk determination, but first extracts, sorts, and organizes the target object behaviors related to the preceding weak risk events according to a unified event format. The behavior sequence construction module 31 includes a behavior event extraction submodule, an event sorting submodule, and a sequence generation submodule, wherein: The behavior event extraction submodule extracts abnormal behavior events from the relevant target object trajectory information and behavior status information output by the continuous tracking module 24, and records the corresponding occurrence time, occurrence location and behavior category; In practice, each identified abnormal behavior event can be uniformly represented as an event item. Each event item contains at least three basic fields: event time, event location, and behavior category. Correspondingly, a single behavior event can be represented as: ; in, Indicates the first One behavioral event; Indicates the first The time of occurrence of each behavioral event; Indicates the first The location where the behavioral event occurred; Indicates the first The behavioral category of each behavioral event.

[0057] in, This can be further represented as the spatial coordinates of the location where the event occurred, i.e.: ; in, Indicates the first The location where the behavioral event occurred; Indicates the first The x-coordinate of the location where each behavioral event occurred; Indicates the first The vertical coordinate of the location where each behavioral event occurs.

[0058] In this embodiment, behavior category This can correspond to any of the abnormal approach behavior, abnormal stay behavior, unplanned equipment approach behavior, or path deviation behavior output by the aforementioned risk monitoring unit 2. The behavior event extraction submodule can extract the moment when a state transition occurs or an anomaly judgment condition is met based on the continuous changes in the target trajectory and the changes in behavior state output by the continuous tracking module 24, and generate the corresponding behavior event item.

[0059] Furthermore, to avoid repeatedly recording the same abnormal process as multiple discrete events within a very short time, in this embodiment, the behavior event extraction submodule can aggregate original abnormal records that have the same target object, the same behavior category, and occur within a continuous short time window, retaining only the start time and representative location of the abnormal behavior as the behavior event output. The advantage of this processing method is that by replacing the original frame-level records with event-level extraction, subsequent event sorting and chain reasoning are based on "abnormal behavior events" rather than "perceptual sampling points".

[0060] The event sorting submodule sorts abnormal behavior events according to the order of their occurrence, forming a set of behavior events with temporal continuity. Specifically, the event sorting submodule receives multiple behavioral events output by the behavioral event extraction submodule. And according to the time of occurrence of each behavioral event. Sort the events from earliest to latest to form an ordered set of events: ; in, This represents a set of behavioral events ordered by time. Indicates the sorted order of the first... One behavioral event; Indicates the sorted order of the first... The time of occurrence of each behavioral event; This represents the total number of behavioral events extracted from the target object during the current local reinforcement reasoning phase.

[0061] In this embodiment, if multiple behavioral events occur at the same time, they can be further arranged according to the order of event collection or the order of event writing to ensure the uniqueness of the sequence generation result. The function of the event sorting submodule is to provide strictly ordered behavioral input for subsequent spatiotemporal correlation analysis, so that the time interval and sequence of adjacent behavioral events have a clear correspondence.

[0062] The sequence generation submodule constructs a continuous sequence of behaviors for the corresponding target object based on the set of behavior events, and outputs the continuous sequence of behaviors to the spatiotemporal correlation analysis module 32 for subsequent behavior-time-space correlation analysis.

[0063] Specifically, the sequence generation submodule can organize the sorted set of behavioral events into a continuous sequence of behaviors for each relevant target object. : ; in, This represents a continuous sequence of actions corresponding to the target object; This represents a sequence structure formed by connecting various behavioral events in chronological order; This represents the number of behavioral events in a continuous sequence of behaviors.

[0064] Furthermore, if multiple related target objects exist under the same local reinforcement inference state, the sequence generation submodule can generate corresponding continuous behavior sequences for each target object and output them one by one to the spatiotemporal correlation analysis module 32. Thus, the behavior sequence construction module 31 completes the structural transformation from continuous target tracking information to continuous behavior sequences.

[0065] In this embodiment, the spatiotemporal correlation analysis module 32 is used to perform behavior-time-space correlation analysis on each behavior event in a continuous behavior sequence and calculate the spatiotemporal correlation degree between behaviors. It identifies spatiotemporal relationships with temporal continuity and spatial proximity; the behavior-time-space correlation analysis process of the spatiotemporal correlation analysis module 32 includes the following steps: S32.1 Receive the continuous behavior sequence output by the behavior sequence construction module 31, and extract the occurrence time, spatial location and behavior type information corresponding to each behavior event; Specifically, the spatiotemporal correlation analysis module 32 analyzes continuous behavioral sequences. Read any two adjacent behavior events and and extract its event time. , Event location , and behavioral categories , The above information will serve as the input basis for temporal correlation analysis and spatial correlation analysis.

[0066] S32.2 Extract behavioral temporal correlation features based on the time interval between occurrences of adjacent behavioral events in a continuous behavioral sequence, and form corresponding temporal correlation indicators; Specifically, the time interval between adjacent behavioral events can be expressed as: ; in, Indicates the sorted order of the first... The first behavioral event and the first The time interval between the occurrence of each behavioral event; Indicates the sorted order of the first... The time of occurrence of each behavioral event; Indicates the sorted order of the first... The time when a behavioral event occurs.

[0067] To characterize the temporal continuity of two behavioral events, in this embodiment, the temporal correlation index can be defined as: ; in, Indicates the first Temporal correlation indicators for adjacent behavioral events; This represents the time interval between the occurrences of the pair of behavioral events. From this definition, it can be seen that the shorter the interval between behavioral events, the more... The larger the value, the stronger the temporal continuity; the longer the interval between behavioral events, the stronger the continuity. The smaller the value, the weaker the temporal continuity.

[0068] The improvement of this processing method lies in the fact that instead of simply using a fixed time threshold for binarization screening, it first transforms the event time interval into a continuously quantified time-related indicator, providing a unified dimensional basis for subsequent comprehensive analysis with spatially related indicators.

[0069] The calculation example is as follows: If the times of occurrence of two adjacent behavioral events are respectively and ,but: ; ; in, This indicates that the time interval between the occurrence of the two events is 3 time units; This indicates that the time correlation index for this pair of behavioral events is 0.25.

[0070] S32.3 Extract spatial correlation features of behaviors based on the spatial relationship between the location of each behavioral event and the adjacent area of ​​the dangerous goods container, and form corresponding spatial correlation indicators; Specifically, the shortest distance from the location of the behavioral event to the boundary of the adjacent area of ​​the dangerous goods container can be defined as follows: When a behavioral event occurs within the vicinity of a dangerous goods container, it is permissible to take [the appropriate action]. When a behavioral event occurs outside the vicinity of a dangerous goods container, it is the shortest spatial distance from the location of the event to the boundary of the vicinity.

[0071] Accordingly, the spatial correlation index of the i-th behavioral event can be defined as: ; in, Indicates the first Spatial correlation indicators for individual behavioral events; Indicates the first The shortest distance between the location of a behavioral event and the area adjacent to the dangerous goods container. From this definition, it can be seen that the closer the behavioral event is to the area adjacent to the dangerous goods container, the shorter the distance between the event and the container's location. The larger the value; the farther away the behavioral event is, the greater the value. The smaller.

[0072] For two adjacent behavioral events and To uniformly characterize the spatial proximity of the event pair to the area adjacent to the dangerous goods container, in this embodiment, the spatial correlation index of the pair of behavioral events can be defined as the average of the spatial correlation indices of the two single events: ; in, Indicates the first Spatial correlation indicators for adjacent behavioral events; Indicates the first Spatial correlation indicators for behavioral events.

[0073] The improvement of this approach lies in the fact that it does not only examine whether the location of a single behavioral event enters the vicinity of the dangerous goods container, but also examines the spatial relationship between the preceding and following behavioral events and the vicinity of the dangerous goods container. This makes it more suitable for identifying the risk evolution characteristics of "multiple abnormal behaviors occurring consecutively in the vicinity and forming spatial correlations".

[0074] The calculation example is as follows: If the shortest distances from two adjacent events to the boundary of the adjacent area of ​​the dangerous goods container are respectively and ,but: ; ; ; in, Indicates the first The incident occurred in the area adjacent to the dangerous goods container; Indicates the first The incident occurred in a location close to the vicinity of the hazardous materials container; This indicates that the pair of behavioral events as a whole have a strong spatial correlation.

[0075] S32.4. Conduct a comprehensive analysis of time-related and spatial-related indicators to determine the spatiotemporal correlation between behaviors. ; Specifically, the first Spatiotemporal correlation between adjacent behavioral events It can be represented as: ; in, Indicates the first The spatiotemporal correlation between adjacent behavioral events; Indicates the first Temporal correlation indicators for adjacent behavioral events; Indicates the first Spatial correlation indicators for adjacent behavioral events; Indicates the weighting coefficient of time-related indicators; Represents the weight coefficients of spatial correlation indicators, and satisfies .

[0076] In practical implementation, the importance of temporal continuity and spatial proximity in port hazardous materials operations can be determined based on these factors. and Configure settings. When a greater emphasis is placed on the rapid continuity of abnormal behavior over time, the settings can be appropriately increased. When greater emphasis is placed on the spatial proximity of abnormal behavior to the vicinity of dangerous goods containers, the appropriate increase can be made. .

[0077] The calculation example is as follows: If a time correlation index of a pair of adjacent behavioral events Spatial correlation indicators ,Pick , ,but: ; in, This indicates that the overall spatiotemporal correlation between the two adjacent behavioral events is 0.55.

[0078] S32.5, Based on spatiotemporal correlation Determine whether the behavioral events in a continuous behavioral sequence meet the behavioral-temporal-spatial correlation conditions, and output the behavioral events that meet the correlation conditions to the accident precursor chain reasoning module 33.

[0079] Specifically, a threshold for determining spatiotemporal correlation can be set. When the following formula is satisfied, the corresponding behavior event is determined to satisfy the behavior-time-space association condition: ; in, This represents the threshold for determining spatiotemporal correlation.

[0080] If multiple sets of adjacent behavioral events in a continuous behavioral sequence all satisfy the above conditions, the spatiotemporal correlation analysis module 32 can group these satisfying behavioral events into behavioral event combinations according to their chronological order and output them to the accident precursor chain inference module 33. Thus, the spatiotemporal correlation analysis module 32 completes the screening process from "continuous behavioral sequence" to "behavioral event combinations that satisfy correlation conditions".

[0081] In this embodiment, the accident precursor chain reasoning module 33 identifies a chain of abnormal behaviors that occur continuously in the vicinity of the dangerous goods container based on spatiotemporal correlation, and performs accident precursor chain reasoning based on the evolutionary relationship between each behavioral event in the behavior chain, while calculating the risk progression value. This is to achieve state control for progressive risk level progression and localized reinforcement reasoning. The accident precursor chain reasoning module 33 is used to further construct chains, spatially filter, and progressively evaluate combinations of behavioral events that already meet the behavioral-temporal-spatial correlation conditions, in order to identify accident precursor chains. The accident precursor chain reasoning and risk progression process of the accident precursor chain reasoning module 33 includes the following steps: S33.1 Receive the combination of behavioral events that satisfy the behavior-time-space correlation conditions output by the spatiotemporal correlation analysis module 32, and extract the occurrence time, spatial location and behavior type information corresponding to each behavioral event; Specifically, the accident precursor chain reasoning module 33 can record the input behavioral events as follows: ; in, This represents a combination of behavioral events that satisfy the behavioral-temporal-spatial correlation condition. Indicates the first in the combination of behavioral events One behavioral event; This indicates the number of events in the event combination for this behavior.

[0082] S33.2. Construct a chain of abnormal behavior chains by linking the behavioral events according to their chronological order. Specifically, the abnormal behavior chains can be organized according to the chronological order of the behavioral events in the combination of behavioral events. : ; in, Indicates a chain of abnormal behavior; This represents a chain structure formed by connecting multiple behavioral events that satisfy the spatiotemporal correlation conditions in chronological order of their occurrence. This indicates the number of behavioral events in the abnormal behavior chain.

[0083] The basis for the formation of the abnormal behavior chain is that adjacent behavioral events have been determined to have temporal continuity and spatial proximity through the spatiotemporal correlation analysis module 32, and therefore can be connected in a chain-like manner to represent a continuously developing abnormal behavior process.

[0084] S33.3. Based on the spatial relationship between the location of each behavioral event in the abnormal behavior chain and the adjacent area of ​​the dangerous goods container, the abnormal behavior chain is screened to identify and form an accident precursor chain. Specifically, each behavioral event in the abnormal behavior chain can be individually assessed to determine whether it is located within the area adjacent to the dangerous goods container, or whether it maintains a predetermined spatial proximity to that area. When behavioral events in the abnormal behavior chain continuously occur within the area adjacent to the dangerous goods container, or when individual events are located outside the adjacent area but the overall chain maintains significant spatial proximity, the abnormal behavior chain can be identified as an accident precursor chain. .

[0085] To facilitate implementation, spatial filtering conditions for the accident precursor chain can be set: if the proportion of behavioral events in the abnormal behavior chain that are located in the vicinity of the dangerous goods container or have a preset spatial proximity relationship with it is not less than a preset threshold. If so, then the chain of abnormal behavior is determined to form a precursor chain of an accident. This can be represented as: ; in, This indicates the percentage of behavioral events in the abnormal behavior chain that satisfy the spatial constraints of the adjacent area of ​​the dangerous goods container; This represents the number of behavioral events in the abnormal behavior chain that satisfy the spatial constraints of the adjacent area of ​​the dangerous goods container; This represents the total number of behavioral events in the abnormal behavior chain; This indicates the threshold for the spatial screening ratio of the accident precursor chain.

[0086] The improvement of this approach lies in the fact that it does not rely solely on whether a single abnormal behavior occurs in the vicinity of a dangerous goods container to identify an accident precursor. Instead, it starts from the entire abnormal behavior chain and combines the continuous spatial relationship between multiple behavioral events and the vicinity of the dangerous goods container for screening. This is more in line with the characteristic that an "accident precursor chain" is formed by the evolution of continuous behaviors.

[0087] S33.4 Calculate the risk progression value based on the behavioral type, order of occurrence, and correlation strength of each behavioral event in the accident precursor chain. ; Specifically, we can start with the first link in the chain of warning signs of an accident. Set behavior type weights for each behavior event. This weight is used to characterize the relative role of different abnormal behavior types in risk progression; then an order coefficient is set. This is used to characterize the impact of the order of occurrence of behavioral events in the accident precursor chain on risk evolution; and to introduce the spatiotemporal correlation between adjacent behavioral events. As a measure of correlation strength, the risk progression value... The calculation formula is: ; in, Indicates the progressive risk value; Indicates the first The event to the 1 Evolution weights between events; This indicates the first in the chain of precursory events. The order coefficients corresponding to each behavioral event; This indicates the first in the chain of precursory events. The first behavioral event and the first The spatiotemporal correlation between individual behavioral events; Indicates the number of behavioral events in the warning chain of an incident; This indicates the event sequence number in the precursor chain of an accident.

[0088] Among them, behavior type weight Used to illustrate the differences in the impact of different types of abnormal behavior on risk progression; order coefficient This is used to indicate that the later a behavioral event appears in the chain, the closer it is to a further stage of risk evolution. For example, a larger sequence coefficient can be assigned to later behavioral events to reflect the gradual progression of consecutive abnormal behaviors.

[0089] The advantage of this formula lies in the fact that it does not simply count the behavioral events in the accident precursor chain, but simultaneously introduces three factors—behavioral type, order of occurrence, and spatiotemporal correlation strength—for joint calculation, thus enabling the risk progression value to be calculated. This better aligns with the technical logic that "multiple abnormal behaviors occur consecutively in the vicinity of a dangerous goods container and are spatially correlated to form a chain of precursory events for accidents."

[0090] The calculation example is as follows: Suppose a certain accident precursor chain contains three behavioral events, thus there are two sets of adjacent event associations. ; ; Then the risk progression value for: ; ; in, The risk progression value corresponding to the precursor chain of the incident is 1.642. This result indicates that the risk state progressively increases as subsequent behavioral events occur in the chain and as there is a strong spatiotemporal correlation between adjacent behaviors.

[0091] S33.5, Based on risk progression value The risk level is progressively upgraded, and the upgraded risk level is output to the risk assessment and hierarchical control unit 4; Specifically, multiple risk progression threshold ranges can be preset, and the risk progression values ​​can be adjusted accordingly. The range in which the risk level is determined. When As the chain of warning signs lengthens or the behavioral correlation strengthens, the corresponding risk level increases progressively. This can be expressed as: if... If the risk level exceeds the upper limit threshold corresponding to the current risk level, the risk level will be upgraded to the next level.

[0092] It is worth noting that the key point of this step is to progressively update the risk level based on the inference results of the accident precursor chain, and the update is based on the risk progression value. The specific risk level and the corresponding handling actions belong to the subsequent risk assessment and hierarchical control unit 4.

[0093] S33.6. If no subsequent related behaviors or related target objects are detected to have moved out of the vicinity of the dangerous goods container, interrupt the progression of the accident precursor chain and control the system to exit the local enhanced reasoning state to restore normal monitoring.

[0094] Specifically, the accident precursor chain inference module 33 can continuously determine whether the accident precursor chain will continue to extend. For ease of implementation, a preset subsequent observation time window can be set. The time window following the occurrence of the last behavioral event in the current precursory chain of events. If no new subsequent behavioral events satisfying the behavior-time-space correlation conditions are detected within the current event precursor chain, it is determined that the current event precursor chain has not continued to develop.

[0095] In addition, if the relevant target object has moved out of the risk range near the dangerous goods container, and subsequent actions no longer have a spatial correlation basis in the dangerous goods proximity scenario, then the current accident precursor chain progression process is interrupted.

[0096] For ease of implementation, "moving out of the risk zone adjacent to the dangerous goods container" can be defined as any of the following situations: First, the shortest distance between the current location of the relevant target and the area adjacent to the dangerous goods container is greater than a preset exit distance threshold. Secondly, the target object no longer belongs to the area of ​​concern adjacent to the dangerous goods container.

[0097] Furthermore, to avoid accidental exits due to momentary jitter detection or short-term area switching, an exit hold window can be set. When the above-mentioned exit condition is met within a consecutive number of sampling times, or within a consecutive time window... When the condition is continuously established, it is determined that the relevant target has been stably removed from the risk range near the dangerous goods container.

[0098] When either the current accident precursor chain does not continue to develop or the relevant target object has been stably removed from the risk range near the dangerous goods container, the accident precursor chain inference module 33 can end the current local enhanced inference process and output the control result of exiting the local enhanced inference state, so that the system returns to the normal monitoring state.

[0099] The purpose of this exit mechanism is to keep the local reinforcement reasoning state consistent with the actual risk evolution process. That is, reinforcement reasoning is maintained only when the accident precursor chain continues to develop, and exits in a timely manner when the behavioral progression is interrupted or the risk scenario is resolved, so that the risk reasoning process has dynamic convergence characteristics.

[0100] It should be added that the overall operation process of risk reasoning unit 3 may include the following: First, under the local reinforcement reasoning state, the behavior sequence construction module 31 extracts events, sorts them by time, and generates sequences from the relevant target object behavior information output by the continuous tracking module 24 to form a continuous behavior sequence of the corresponding target object; then, the spatiotemporal correlation analysis module 32 performs behavior-time-space correlation analysis on each behavior event in the continuous behavior sequence to calculate the spatiotemporal correlation degree between adjacent behavior events. The system then filters out combinations of behavioral events that meet the behavior-time-space correlation conditions. Next, the accident precursor chain reasoning module 33 constructs chain associations and performs spatial filtering on the combinations of behavioral events that meet the correlation conditions, identifies and forms accident precursor chains, and calculates risk progression values ​​based on behavior type, order of occurrence, and correlation strength. Subsequently, based on the risk progression value The risk level is progressively increased; finally, if no subsequent related behaviors are detected or the relevant target object moves out of the risk range adjacent to the dangerous goods container, the accident precursor chain progression process is interrupted, the local enhanced reasoning state is exited, and normal monitoring is resumed. Through the above method, the risk reasoning unit 3 can complete the overall reasoning process from the construction of continuous behavior sequences to the identification of accident precursor chains, and then to the progressive control of risks.

[0101] Risk assessment and hierarchical control unit 4 determines the risk level of each area in the storage area based on the risk status output by risk reasoning unit 3, and triggers corresponding hierarchical control operations based on the determination results. Hierarchical control operations include outputting early warning information, triggering on-site prompts, or sending control commands. In this way, the risk status formed by risk reasoning unit 3 can be further transformed into hierarchical disposal results for on-site application, so that the risk level of different areas in the hazardous materials storage area can correspond to different control response methods, thereby realizing the connection between risk identification results and on-site control actions.

[0102] In this embodiment, the risk assessment and hierarchical control unit 4 includes a risk level determination module 41, a hierarchical control module 42, and a processing result output module 43. These modules can work collaboratively through data interfaces, bus connections, or internal calls within the same processing platform. Wherein: The risk level determination module 41 determines the risk level of each area in the storage area based on the risk status output by the risk reasoning unit 3. Here, the risk status can be understood as the risk representation result corresponding to the target area output by the risk reasoning unit 3, which may include at least the progressively advancing risk status information, the corresponding area information, and the behavioral chain information associated with that risk status. After receiving the aforementioned risk status, the risk level determination module 41 binds the risk status to the corresponding area, thereby forming a region-level risk level determination result.

[0103] In practical implementation, the risk level determination module 41 can determine the risk level of the dangerous goods container stacking area, passage area, and operation boundary area according to the storage area division results. If a certain area has a risk status output by the risk reasoning unit 3, then that risk status is used as the basis for determining the current risk level of that area; if the same area corresponds to multiple risk statuses within the same time period, then the higher risk status can be used as the current risk level determination result for that area. Thus, the risk level determination module 41 can uniformly convert the reasoning results of the risk reasoning unit 3, which are oriented towards target objects and behavioral chains, into risk level results oriented towards area management.

[0104] Furthermore, the risk level determination module 41 can output a determination result that includes a risk area identifier, risk level information, and risk status source information. The risk area identifier represents the area where the risk has occurred; the risk level information represents the current risk level of that area; and the risk status source information represents the behavioral chain or reasoning result source corresponding to that risk level. Through the above output, a clear basis for triggering actions can be provided for the subsequent hierarchical control module 42.

[0105] The hierarchical control module 42 triggers corresponding hierarchical control operations based on the judgment results. The hierarchical control operations include outputting early warning information, triggering on-site prompts, or sending control commands. Specifically, the function of the graded control module 42 is to provide different response methods according to different risk levels, so that the risk handling process and the risk level determination results are in a corresponding relationship.

[0106] In practice, the hierarchical control module 42 receives the regional risk level determination result output by the risk level determination module 41 and triggers corresponding control operations based on the preset level response relationship. When the risk level reaches the early warning trigger condition, the corresponding early warning information can be output; when the risk level reaches the on-site prompt trigger condition, the on-site prompt can be triggered; when the risk level reaches the control command trigger condition, the corresponding control command can be sent. The above three types of hierarchical control operations can be triggered individually or in combination according to the hierarchical relationship of risk levels.

[0107] Among them, outputting early warning information can be used to notify managers or the duty system that a risk status has appeared in the current area. The early warning information can include the risk area, risk level, and a summary of the related behavior chain. Triggering on-site prompts can be used to remind workers at the hazardous materials storage area to pay attention to the risk area and abnormal operation status through sound, light, text or interface prompts. Sending control commands can be used to send corresponding control information to relevant on-site equipment, control terminals or linkage execution objects to cooperate with risk disposal.

[0108] Furthermore, in this embodiment, the hierarchical control module 42 does not limit the specific correspondence between risk level and action, but triggers corresponding hierarchical control operations based on the judgment result. That is, in practical applications, different early warning information content, on-site prompts, and control command formats can be pre-configured for different risk levels according to port hazardous materials operation management requirements. When the risk level progressively increases, the hierarchical control module 42 triggers a higher-level control operation; when the risk level decreases or the risk status is resolved, the corresponding hierarchical control operation is stopped or released. This implementation method ensures that the hierarchical control process remains consistent with changes in risk level.

[0109] The processing result output module 43 is used to output the processing results of the hierarchical control operation and output the processing results to the result output and feedback unit 5. The processing results here may include early warning information, on-site prompt information and control instruction information that have been executed or generated, and may also include risk area and risk level information corresponding to the above processing actions.

[0110] In practical implementation, the processing result output module 43 can organize and format the various processing results triggered by the hierarchical control module 42, giving the processing results a unified data structure so that subsequent result output and feedback unit 5 can receive and continue outputting. The content output by the processing result output module 43 may include at least: risk area information, risk level information, corresponding hierarchical control operation type, and related behavioral chain information identifiers. Through the above method, the processing result output module 43 can complete the conversion from "hierarchical disposal actions" to "standardized output results".

[0111] Furthermore, in this embodiment, if multiple processing results correspond to the same area, the processing result output module 43 can organize the output according to the risk level or the processing order, so that the result output and the feedback unit 5 can maintain clarity and consistency when sending them externally. The processing result output module 43 does not directly undertake the external sending action, but serves as the result aggregation and output interface within the risk assessment and hierarchical control unit 4.

[0112] The results output and feedback unit 5 is used to output the processing results of the risk assessment and hierarchical control unit 4, and send the risk area, risk level, and corresponding behavioral chain information to external systems or management platforms. In this way, the risk judgment and control results generated internally by the system can be further transmitted to the external management side, realizing the external display, linkage, and management application of hazardous materials operation risk information.

[0113] In practice, the result output and feedback unit 5 receives the processing results output by the processing result output module 43 and sends the risk area, risk level, and corresponding behavioral chain information as key output content to an external system or management platform. Specifically, the risk area indicates the specific area where the risk has occurred; the risk level indicates the current risk level of that area; and the corresponding behavioral chain information characterizes the abnormal behavioral chain or accident precursor chain information that forms the current risk state. By sending the above content, the external system or management platform can obtain information about the current risk distribution in the hazardous materials storage area and the basis for its formation.

[0114] Furthermore, in this embodiment, the result output and feedback unit 5 can output the processing result as a data message, platform message, or management interface display content suitable for external system recognition. However, this embodiment does not limit the specific output protocol, display method, or platform type. Its core is to transmit the processing result formed by the risk assessment and hierarchical control unit 4 to the outside, enabling managers to view, record, or coordinate actions based on risk areas, risk levels, and corresponding behavioral chain information.

[0115] It should be added that the overall operation process of the risk assessment and hierarchical control unit 4 and the result output and feedback unit 5 may include the following: First, the risk level determination module 41 receives the risk status output by the risk reasoning unit 3 and determines the risk level of each area in the storage area; then, the hierarchical control module 42 triggers corresponding hierarchical control operations based on the determination results, including outputting early warning information, triggering on-site prompts, or sending control commands; next, the processing result output module 43 outputs the processing results of the hierarchical control operations; finally, the result output and feedback unit 5 outputs the above processing results externally and sends the risk area, risk level, and corresponding behavioral chain information to an external system or management platform. Through the above method, the entire processing process from risk status determination, hierarchical handling, to result output can be completed.

[0116] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0117] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A safety management system for hazardous materials operations based on a localized AI model, characterized in that: include: Data modeling unit (1) is used to perform spatial modeling of dangerous goods container storage area in port yard, access basic information and location information of operators, operating equipment and dangerous goods containers, and divide the storage area into regions according to the stacking layout, passage path and operation boundary, and construct a spatial distribution model including dangerous goods container stacking area, passage area and operation boundary area. Risk monitoring unit (2) analyzes the activities of workers, equipment and dangerous goods containers in the vicinity of the port in real time based on an artificial intelligence model deployed locally in the port. Under normal monitoring conditions, it performs preliminary screening of weak risk events in each area to identify early abnormal behaviors such as abnormal approach of personnel, abnormal stay, unplanned approach of equipment or deviation from the path. When a preceding weak risk event occurs in any region, it triggers that region to enter a local enhanced reasoning state and continuously tracks the relevant targets; Risk reasoning unit (3) is used to start a spatiotemporally coupled multi-behavior chain risk reasoning engine in the local enhanced reasoning state, perform behavior-time-space correlation analysis on the continuous behavior sequence of the relevant target, and identify an accident precursor chain formed by multiple abnormal behaviors that occur continuously and are spatially correlated in the vicinity of the dangerous goods container based on the progressive relationship of risk behavior; when the subsequent behavior and the preceding event satisfy the temporal continuity, spatial proximity and evolutionary correlation relationship, the risk level is progressively increased; when no subsequent correlated behavior occurs or the target leaves the risk range, the local enhanced reasoning state is exited and normal monitoring is restored. Risk assessment and hierarchical control unit (4), the risk assessment and hierarchical control unit (4) determines the risk level of each area of ​​the storage area according to the risk status output by the risk reasoning unit (3), and triggers corresponding hierarchical control operations according to the determination results. The hierarchical control operations include outputting early warning information, triggering on-site prompts or sending control commands. The result output and feedback unit (5) is used to output the processing results of the risk assessment and hierarchical control unit (4) and send the risk area, risk level and corresponding behavior chain information to the external system or management platform.

2. The hazardous materials operation safety management system based on a localized AI model according to claim 1, characterized in that, The data modeling unit (1) includes an information acquisition and access module (11), a region parsing module (12), a coordinate mapping module (13), and a model generation module (14), wherein: The information collection and access module (11) is used to access the identity information, category information, operation attribute information and location information of operators, operating equipment and dangerous goods containers; The area analysis module (12) performs area analysis and area division on the storage area of ​​dangerous goods container storage area in the port yard according to the stacking layout, passage path and operation boundary information, forming dangerous goods container stacking area, passage area and operation boundary area; The coordinate mapping module (13) is used to perform coordinate registration of the location information of operators, operating equipment and dangerous goods containers, and to determine the regional affiliation of each target object; The model generation module (14) generates a spatial distribution model containing boundary information of each region and spatial distribution relationship of the target object based on the regional division results and regional affiliation relationship.

3. The hazardous materials operation behavior safety management system based on a localized AI model according to claim 1, characterized in that, The risk monitoring unit (2) includes a real-time analysis module (21), a weak risk screening module (22), a status triggering module (23), and a continuous tracking module (24), wherein: The real-time analysis module (21) analyzes the activities of operators, operating equipment and dangerous goods containers in the vicinity of the port based on the artificial intelligence model deployed locally, and obtains the behavioral status information of each target object in the storage area. The weak risk screening module (22) is used to perform preliminary screening of weak risk events in each area under normal monitoring conditions, based on the behavioral status information of each target object, in order to identify early abnormal behaviors such as abnormal approach of personnel, abnormal stay, unplanned approach of equipment or deviation from the path. The state triggering module (23) is used to trigger the region to enter the local enhanced reasoning state when a preceding weak risk event occurs in any region. The continuous tracking module (24) is used to continuously track the target object related to the preceding weak risk event in the local enhanced reasoning state.

4. The hazardous materials operation behavior safety management system based on a localized AI model according to claim 3, characterized in that, The preliminary screening process for weak risk events in the weak risk screening module (22) includes the following steps: S22.1 Based on the behavioral status information of each target object output by the real-time analysis module (21), extract the location, trajectory, direction of movement, area affiliation and dwell status information of the operators, operating equipment and dangerous goods containers in the current monitoring area; S22.

2. Based on the spatial proximity between the operators and the dangerous goods container, identify whether the operators are exhibiting abnormal approach behavior. Specifically, when an operator enters the dangerous goods container, a preset safe distance threshold is established. When the corresponding adjacent range is considered, it is determined that there is abnormal proximity behavior; S22.

3. Based on the continuous stay of personnel in the vicinity of the dangerous goods container, extract the duration of stay. and the duration of stay Compared with a preset dwell time threshold, when the dwell time... If the preset dwell time threshold is exceeded, abnormal dwelling behavior is determined. S22.

4. Based on the planned operating area, current position and direction of movement of the operating equipment, identify whether the operating equipment deviates from the planned operating range and makes an unplanned approach to the vicinity of the dangerous goods container, and determine whether there is any unplanned approach behavior of the equipment. S22.

5. Based on the deviation between the actual movement trajectory of the operator or equipment and the preset travel path, a path deviation degree is calculated. and the path deviation The path deviation is compared with a preset path deviation threshold. When the path deviation exceeds the preset path deviation threshold, it is determined that there is path deviation behavior; S22.6 Summarize the identification results of abnormal approach behavior, abnormal stay behavior, unplanned equipment approach behavior and path deviation behavior to form the preceding weak risk events of the corresponding area, and output the preceding weak risk events to the status triggering module (23).

5. The hazardous materials operation safety management system based on a localized AI model according to claim 4, characterized in that, The state triggering module (23) includes a region identification submodule, a trigger determination submodule, and a state switching submodule, wherein: When the region identification submodule detects a preceding weak risk event, it identifies the target region where the preceding weak risk event occurred and determines the target region as the region to be strengthened for analysis. The trigger determination submodule determines whether the region to be strengthened for analysis meets the local strengthening reasoning trigger conditions based on the region location, event type, and target object corresponding to the preceding weak risk event. When the region to be reinforced for analysis meets the local reinforcement reasoning triggering condition, the state switching submodule switches the region to be reinforced for analysis from the normal monitoring state to the local reinforcement reasoning state, so as to trigger the risk reasoning unit (3) to perform continuous behavior sequence analysis on the relevant target objects.

6. The hazardous materials operation behavior safety management system based on a localized AI model according to claim 5, characterized in that, The risk reasoning unit (3) includes a behavior sequence construction module (31), a spatiotemporal correlation analysis module (32), and an accident precursor chain reasoning module (33), wherein: The behavior sequence construction module (31) is used to sort the relevant target object behavior information output by the continuous tracking module (24) by time and aggregate the events under the local reinforcement reasoning state to form a continuous behavior sequence of the corresponding target object; The spatiotemporal correlation analysis module (32) is used to perform behavior-time-space correlation analysis on each behavior event in the continuous behavior sequence and calculate the spatiotemporal correlation degree between behaviors. And identify spatiotemporal relationships with temporal continuity and spatial proximity; The accident precursor chain reasoning module (33) identifies a chain of abnormal behaviors that occur continuously in the vicinity of the dangerous goods container based on the spatiotemporal correlation, and performs accident precursor chain reasoning based on the evolutionary relationship between each behavioral event in the behavior chain, while calculating the risk progression value. This is to achieve state control that allows for progressive risk levels and locally enhanced reasoning.

7. The hazardous materials operation behavior safety management system based on a localized AI model according to claim 6, characterized in that, The behavior sequence construction module (31) includes a behavior event extraction submodule, an event sorting submodule, and a sequence generation submodule, wherein: The behavior event extraction submodule extracts abnormal behavior events from the relevant target object trajectory information and behavior status information output by the continuous tracking module (24), and records the corresponding occurrence time, occurrence location and behavior category; The event sorting submodule sorts the abnormal behavior events according to the order of their occurrence time, forming a set of behavior events with temporal continuity. The sequence generation submodule constructs a continuous behavior sequence of the corresponding target object based on the set of behavior events, and outputs the continuous behavior sequence to the spatiotemporal correlation analysis module (32) for subsequent behavior-time-space correlation analysis.

8. The hazardous materials operation behavior safety management system based on a localized AI model according to claim 7, characterized in that, The behavior-time-space correlation analysis process of the spatiotemporal correlation analysis module (32) includes the following steps: S32.1 Receive the continuous behavior sequence output by the behavior sequence construction module (31) and extract the occurrence time, spatial location and behavior type information of each behavior event; S32.2 Extract behavioral temporal correlation features based on the time interval between occurrences of adjacent behavioral events in a continuous behavioral sequence, and form corresponding temporal correlation indicators; S32.3 Extract spatial correlation features of behaviors based on the spatial relationship between the location of each behavioral event and the adjacent area of ​​the dangerous goods container, and form corresponding spatial correlation indicators; S32.

4. Conduct a comprehensive analysis of the aforementioned time-related indicators and spatial-related indicators to form the spatiotemporal correlation degree between behaviors. ; S32.5, Based on the aforementioned spatiotemporal correlation degree Determine whether the behavioral events in the continuous behavioral sequence meet the behavioral-temporal-spatial correlation conditions, and output the behavioral events that meet the correlation conditions to the accident precursor chain reasoning module (33).

9. The hazardous materials operation behavior safety management system based on a localized AI model according to claim 8, characterized in that, The accident precursor chain reasoning and risk progression process of the accident precursor chain reasoning module (33) includes the following steps: S33.1 Receive the combination of behavioral events that meet the behavioral-time-space correlation conditions output by the spatiotemporal correlation analysis module (32), and extract the occurrence time, spatial location and behavioral type information of each behavioral event; S33.

2. Construct a chain-like association of the behavioral events based on the chronological order of their occurrence to form an abnormal behavior chain; S33.

3. Based on the spatial relationship between the location of each behavioral event in the abnormal behavior chain and the adjacent area of ​​the dangerous goods container, the abnormal behavior chain is screened to identify and form an accident precursor chain. S33.4 Calculate the risk progression value based on the behavioral type, order of occurrence, and correlation strength of each behavioral event in the accident precursor chain. ; S33.5, Based on the aforementioned risk progression value The risk level is progressively upgraded, and the upgraded risk level is output to the risk assessment and hierarchical control unit (4). S33.

6. If no subsequent related behaviors or related target objects are detected to have moved out of the vicinity of the dangerous goods container, interrupt the progression of the accident precursor chain and control the system to exit the local enhanced reasoning state to restore normal monitoring.

10. The hazardous materials operation behavior safety management system based on a localized AI model according to claim 1, characterized in that, The risk assessment and hierarchical control unit (4) includes a risk level determination module (41), a hierarchical control module (42), and a processing result output module (43), wherein: The risk level determination module (41) determines the risk level of each area of ​​the storage area based on the risk status output by the risk reasoning unit (3). The hierarchical control module (42) triggers corresponding hierarchical control operations based on the judgment result. The hierarchical control operations include outputting early warning information, triggering on-site prompts, or sending control commands. The processing result output module (43) is used to output the processing result of the hierarchical control operation and output the processing result to the result output and feedback unit (5).

Images