Emergency system for supervising dangerous chemical laboratories

Abstract

The application relates to the technical field of laboratory intelligent management, and discloses a hazardous chemical laboratory supervision emergency system, which takes a central processing unit as a core, integrates seven modules of multidimensional perception, account registration, a laboratory digital twin model library, cross verification of taking and using, behavior recognition intervention, emergency response strategies and a mobile terminal interaction assistant, constructs a complete closed-loop management system from hazardous chemical taking compliance verification, personnel operation safety monitoring to intelligent emergency response, and comprehensively improves the safety management level of a hazardous chemical laboratory. The system can realize real-time linkage of data of each module, break information islands, avoid taking violations from the source, block operation risks in real time, efficiently dispose of emergency conditions, rely on the mobile terminal interaction assistant to improve emergency efficiency, comprehensively strengthen the whole life cycle supervision capability of hazardous chemicals, significantly reduce the incidence of laboratory safety accidents, improve the efficiency of event emergency disposal, and improve the intelligentization and refinement level of overall safety management.

Description

Technical Field

[0001] This application relates to the field of intelligent laboratory management technology, and more specifically, to an emergency monitoring system for hazardous chemical laboratories. Background Technology

[0002] Hazardous chemical laboratories, as core locations for scientific research and teaching activities, routinely handle various hazardous chemicals, including flammable, explosive, highly toxic, and corrosive substances. In recent years, numerous safety accidents have occurred within the industry due to improper handling and management of hazardous chemicals, causing not only property damage and personal injury but also severely disrupting the order of scientific research and teaching. This underscores the urgent necessity for systematic and intelligent safety supervision of hazardous chemical laboratories.

[0003] From the perspective of current technology, the safety management of hazardous chemical laboratories has initially introduced some single-point intelligent equipment. However, these technical solutions have obvious systemic defects and are difficult to meet the needs of full-process safety supervision: First, the sensing and verification methods are limited, with most solutions relying on a single sensor to collect data, which cannot accurately identify abnormal consumption of hazardous chemicals and is prone to missed risk assessments; Second, behavioral supervision is lagging, with traditional management models still mainly relying on manual inspections, making it difficult to achieve 24-hour real-time monitoring, and some laboratories have caused safety accidents due to failure to detect violations in a timely manner; Third, the emergency response system is static, with existing emergency plans mostly existing in the form of paper documents or fixed procedures, unable to be dynamically adjusted in conjunction with real-time environmental data, and the storage location of emergency supplies often relies on manual memory, which can easily delay the response time in emergencies due to untimely location; Fourth, the management system is fragmented, and data cannot be shared and linked in real time, resulting in chaotic hazardous chemical inventory and an inability to accurately trace the flow of goods.

[0004] Therefore, there is an urgent need for a hazardous chemical laboratory supervision and emergency response system to address the shortcomings of existing technologies and achieve integrated and intelligent laboratory supervision and emergency response. Summary of the Invention

[0005] This application provides a hazardous chemical laboratory supervision and emergency response system. Based on the laboratory digital twin model library, it realizes integrated and intelligent closed-loop management from compliance verification of hazardous chemical handling and safety monitoring of personnel operations to intelligent emergency response, thereby comprehensively improving the safety management level of hazardous chemical laboratories.

[0006] An emergency monitoring system for hazardous chemical laboratories includes a central processing unit, and a multi-dimensional sensing module, a ledger registration module, a laboratory digital twin model library, a cross-validation module for retrieval, a behavior recognition and intervention module, an emergency response strategy module, and a mobile interactive assistant, all of which are communicatively connected to the central processing unit.

[0007] The multidimensional sensing module is used to collect multidimensional data from the laboratory site;

[0008] The ledger registration module is used to automate the recording of ledger information;

[0009] The laboratory digital twin model library contains pre-stored laboratory model data including building topology, hazardous chemical storage data, experimental equipment data, and emergency material locations.

[0010] The cross-validation module uses at least two dosage detection sensors and a data fusion analysis unit to perform real-time comparison and cross-validation of hazardous chemical consumption, and outputs the final dosage value and abnormal status diagnosis signal.

[0011] The behavior recognition and intervention module analyzes the video stream from the multi-dimensional perception module to identify the experimenter's violation of regulations and trigger active intervention.

[0012] The emergency response strategy module is used to activate the emergency mode in response to alarm signals. Based on the laboratory digital twin model library and the laboratory on-site multidimensional data, it generates dynamic emergency response strategies through the strategy engine and decomposes them into task instructions.

[0013] The mobile interactive assistant is used to receive and push the task instructions to the user, and provide interactive guidance including augmented reality navigation.

[0014] Optionally, the dosage detection sensor combination in the cross-validation module includes a weighing sensor, a pressure sensor, and a temperature sensor;

[0015] The data fusion and analysis unit is used to perform:

[0016] Based on the readings of the pressure sensor and temperature sensor, the first dosage value is calculated according to the ideal gas law.

[0017] The second dosage value is calculated based on the change in the reading of the weighing sensor;

[0018] The first dosage value and the second dosage value are compared in real time to determine the dosage difference;

[0019] When the difference in dosage is within a preset reasonable error range, the weighted average value is output as the final dosage value;

[0020] When the difference in dosage exceeds the reasonable error range, and the second dosage value is greater than the first dosage value by more than a first preset value, it is determined to be an abnormal state and a diagnostic signal for container leakage is generated.

[0021] When the difference in dosage exceeds the reasonable error range, and the first dosage value is greater than the second dosage value by more than the second preset value, it is determined to be an abnormal state and a diagnostic signal for pressure sensor failure is generated.

[0022] Optionally, the dosage detection sensor combination in the cross-validation module may also include an acoustic fingerprint sensor installed at the container valve or pipeline.

[0023] The data fusion and analysis unit is also used to perform:

[0024] The audio stream collected by the acoustic sensor is received and input into a pre-trained acoustic flow analysis model. The acoustic flow analysis model is used to identify specific acoustic patterns that characterize minute gas leaks, and to calculate a third dosage value by identifying the acoustic features of gas flow and establishing a mapping relationship with gas flow.

[0025] The first dosage value, the second dosage value, and the third dosage value are compared in real time.

[0026] When the difference between any two of the first dosage value, the second dosage value and the third dosage value is within the corresponding preset reasonable error range, the weighted average value is output as the final dosage value.

[0027] When the difference between the first dosage value and the second dosage value exceeds the preset reasonable error range, but there is a reasonable item in the first dosage value and the second dosage value whose difference with the third dosage value is within the reasonable error range, then the weighted average of the third dosage value and the reasonable item is used as the final dosage value, and a diagnostic signal for another sensor fault that is not included in the calculation of the final dosage value is generated.

[0028] When the acoustic flow analysis model identifies the specific acoustic pattern, it determines it to be an abnormal state and generates a leakage warning signal.

[0029] Optionally, the cross-validation module may further include a usage compliance verification unit;

[0030] The usage compliance verification unit is used to perform:

[0031] Before the hazardous chemical handling operation begins, receive the pre-use amount input by the experimenter;

[0032] After the hazardous chemical handling operation is completed, obtain the final usage value output by the handling cross-validation module;

[0033] If the deviation between the pre-used amount and the final usage value exceeds a preset deviation threshold, it is determined to be an abnormal state, the current operation record is locked, and a verification report is generated.

[0034] Optionally, when the behavior recognition intervention module detects that the experimenter has started to handle hazardous chemicals, it sends a start command to the handling cross-validation module to trigger the handling cross-validation module to perform dosage cross-validation.

[0035] When the cross-validation module determines that the operation is in an abnormal state, it sends a behavior analysis signal to the behavior recognition intervention module, triggering the behavior recognition intervention module to perform video stream analysis on the current operation area and improve the sensitivity of the identification of the violation.

[0036] Optionally, the emergency response strategy module includes a dynamic risk assessment unit;

[0037] The dynamic risk assessment unit is used to perform:

[0038] Construct a multidimensional feature vector for risk assessment. The dimensions of the multidimensional feature vector for risk assessment include at least the hazard index of the leaked substance, the estimated leakage intensity based on the output of the cross-validation module, the ventilation level of the laboratory space, and the number of trapped personnel in real time obtained through the behavior recognition and intervention module.

[0039] The risk assessment multidimensional feature vector is input into a pre-trained risk level assessment model, which is generated based on historical accident data and simulated data, and outputs a quantified comprehensive risk level.

[0040] The strategy engine triggers a warning information multicast protocol that matches the overall risk level. The warning information multicast protocol defines the target notification department, personnel priority set, and information content template.

[0041] Optionally, the execution process of the mobile interactive assistant in the emergency mode includes:

[0042] Load and render a 3D mesh model based on the aforementioned laboratory digital twin model library;

[0043] By integrating data from UWB positioning tags worn by personnel with data from inertial measurement units, real-time positioning and attitude mapping of personnel are achieved in the three-dimensional mesh model.

[0044] Based on the building topology in the digital twin model library, and with the dangerous areas detected by the multi-dimensional perception module being introduced in real time as dynamic obstacles, a path finding algorithm is used to calculate the optimal evacuation route in real time. The cost function of the path finding algorithm integrates the path length, the estimated travel time, and the real-time perceived environmental risk value.

[0045] The optimal evacuation route is overlaid onto the real-time image of the terminal camera in an augmented reality manner to form a visual navigation guide.

[0046] Optionally, for alerts regarding widespread leakage, the execution process of the policy engine includes:

[0047] The hazardous chemicals storage data and experimental equipment data in the laboratory digital twin model library are called up, and based on the predefined experimental equipment asset value weight and hazardous chemical hazard weight, an equipment priority sequence and a hazardous chemical priority sequence are constructed.

[0048] Based on the multidimensional data simulation and prediction from the laboratory site, leakage diffusion simulation data is obtained. Combined with the equipment priority sequence and the hazardous chemical priority sequence, a dynamic emergency response strategy including equipment transfer sequence and hazardous material isolation scheme is generated.

[0049] The dynamic emergency response strategy is converted and decomposed into task instructions, which are then pushed to the mobile interactive assistants of the corresponding personnel. The mobile interactive assistants provide the optimal action path to the target equipment location or the target hazardous chemical storage location.

[0050] Optionally, the emergency response strategy module may also include a protective equipment planning unit;

[0051] The protective equipment planning unit is used to execute:

[0052] When generating the task instruction, the material task association knowledge graph is accessed, which defines the standard protective equipment and handling tools required to handle different police situations.

[0053] Based on the real-time location of personnel, the nearest target material point with the required standard protective equipment and disposal tools is dynamically calculated from the emergency material points in the digital twin model library using the nearest neighbor search algorithm.

[0054] The target material point is inserted as a preceding task node into the personnel's original task sequence to obtain a composite task chain, and a globally optimal action route is generated for the composite task chain.

[0055] Optionally, the emergency response strategy module dispatches the task instructions to the execution processes of each of the mobile terminal interactive assistants, including:

[0056] Determine the task type, required skills, and task location corresponding to the task instruction, and match the optimal executor from among the online personnel. The matching algorithm takes into account the personnel's skill level, the distance between the real-time location and the task point, and the current task load.

[0057] The task instructions are pushed to the mobile interactive assistant of the optimal executor;

[0058] The system receives the task execution status of the task instruction returned by the mobile interactive assistant of the optimal executor, and re-triggers the optimal executor matching process when the task execution times out or fails.

[0059] Optionally, the ledger registration module includes a registration data acquisition unit and an electronic ledger generation unit;

[0060] The registration data collection unit is used to obtain the final usage value from the cross-validation module, automatically obtain the operator's identity and operation time period from the behavior recognition and intervention module, and obtain the corresponding experimental project information from the enterprise database.

[0061] The electronic ledger generation unit is used to structurally integrate the obtained final usage value, the operator's identity, the operation time period, and the experimental project information to generate electronic ledger records and register them in the hazardous chemicals electronic ledger.

[0062] Optionally, the ledger registration module also includes an access control unit;

[0063] The permission linkage control unit is used to execute:

[0064] The system receives in real time the abnormal status diagnosis signal sent by the cross-validation module and the abnormal usage event sent by the usage compliance verification unit.

[0065] When any of the aforementioned abnormal status diagnostic signals or the aforementioned abnormal dosage event is received, the corresponding operator's identity is determined by querying the currently associated electronic ledger record;

[0066] Based on the operator's identity, an access freeze command is generated and executed, and an access freeze event log containing the abnormal status diagnostic signal or the usage abnormal event is generated.

[0067] Optionally, the mobile interactive assistant has a built-in multimodal interaction engine, which includes at least a voice interaction unit and a video stream scheduling unit.

[0068] The voice interaction unit is used to parse user voice commands and recognize user feedback data on site, and at the same time broadcast the task commands and navigation prompts to the user.

[0069] The video stream scheduling unit is used to invoke and present the real-time video stream of a specific camera in the multi-dimensional perception module according to the user's voice command or the task command.

[0070] Optionally, the emergency response strategy module is also used to execute:

[0071] Receive the user's on-site feedback data uploaded by the mobile interactive assistant;

[0072] When the user's on-site feedback data is inconsistent with the assumptions of the current dynamic emergency response strategy, the strategy adjustment mechanism is triggered;

[0073] The strategy adjustment mechanism, based on the user's on-site feedback data, uses the strategy engine to make real-time corrections to the dynamic emergency response strategy and reassign updated task instructions.

[0074] Optionally, the behavior recognition intervention module includes:

[0075] The video analysis unit is used to decode and perform behavior sequence analysis on the real-time video stream acquired by the multi-dimensional perception module in order to identify the operational behavior of the experimenters.

[0076] The violation pattern library stores the characteristics of violation operation patterns defined based on historical violation cases;

[0077] An intervention triggering unit is used to generate and execute an intervention instruction when the degree of matching between the identified behavioral sequence and any of the features of the violation operation mode exceeds a preset threshold.

[0078] As can be seen from the above technical solutions, the hazardous chemical laboratory supervision and emergency system provided in this application embodiment takes the central processing unit as the core and integrates seven major modules: multi-dimensional perception, ledger registration, laboratory digital twin model library, cross-validation of use, behavior recognition and intervention, emergency response strategy and mobile terminal interactive assistant, to build a complete closed-loop management system from hazardous chemical use compliance verification, personnel operation safety monitoring to intelligent emergency response.

[0079] In the compliance verification stage of hazardous chemical handling, the handling cross-validation module uses at least two types of dosage detection sensors paired with a data fusion analysis unit to compare and verify the consumption of hazardous chemicals in real time. This not only controls the dosage monitoring error to an extremely low range and accurately outputs the final dosage value that matches the actual consumption, but also quickly identifies violations such as over-disposal and abnormal loss and issues diagnostic signals. This completely solves the problems of inaccurate detection and missed risk assessment by single sensors in existing technologies, ensuring compliance in the handling stage from the source. In the personnel operation safety monitoring stage, the behavior recognition and intervention module relies on video stream data collected by the multi-dimensional perception module to automatically identify violations by laboratory personnel through intelligent algorithms, and can trigger active intervention as soon as a violation is detected. Preventive measures replace the traditional, slow-moving manual inspection model, enabling real-time safety monitoring of the entire operation process and preventing accidents caused by human error. In the intelligent emergency response phase, after receiving alarm signals, the emergency response strategy module combines building topology and material location data from the laboratory's digital twin model library with real-time on-site data from the multi-dimensional sensing module. Through a strategy engine, it dynamically generates emergency response strategies adapted to the current scenario, breaks them down into clear task instructions, and pushes them to relevant personnel via a mobile interactive assistant. Simultaneously, it provides augmented reality navigation to guide personnel to quickly locate emergency supplies, evacuation routes, or disposal areas, solving the shortcomings of slow response and difficult material location in traditional static plans, ensuring efficient and accurate emergency response. Furthermore, the ledger registration module automates the recording of ledger information, providing data support for compliance traceability and emergency response. The central processing unit ensures real-time data linkage between all modules, ultimately forming a closed-loop management system from risk prediction and process supervision to emergency response, comprehensively improving the safety management level of hazardous chemical laboratories. Attached Figure Description

[0080] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0081] Figure 1 This is a schematic diagram of a hazardous chemical laboratory monitoring and emergency response system disclosed in an embodiment of this application;

[0082] Figure 2 This is another schematic diagram of a hazardous chemical laboratory monitoring and emergency response system disclosed in an embodiment of this application. Detailed Implementation

[0083] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0084] The following section introduces the solution proposed in this application. The technical solution is as follows, please refer to the text below for details.

[0085] Figure 1 This is a schematic diagram of a hazardous chemical laboratory monitoring and emergency response system disclosed in an embodiment of this application.

[0086] like Figure 1 As shown, the system may include:

[0087] The central processing unit 1, and the multi-dimensional perception module 2, the ledger registration module 3, the laboratory digital twin model library 4, the cross-validation module 5, the behavior recognition and intervention module 6, the emergency response strategy module 7, and the mobile terminal interactive assistant 8, which are respectively connected to the central processing unit 1 in communication.

[0088] The multidimensional sensing module 2 is used to collect multidimensional data from the laboratory site;

[0089] The ledger registration module 3 is used to automate the recording of ledger information;

[0090] The laboratory digital twin model library 4 contains pre-stored laboratory model data including building topology, hazardous chemical storage data, experimental equipment data, and emergency material locations.

[0091] The cross-validation module 5 uses at least two dosage detection sensors and a data fusion analysis unit to perform real-time comparison and cross-validation of hazardous chemical consumption, and outputs the final dosage value and abnormal status diagnosis signal.

[0092] The behavior recognition and intervention module 6 analyzes the video stream from the multi-dimensional perception module to identify the experimenter's violation of regulations and trigger active intervention.

[0093] The emergency response strategy module 7 is used to activate the emergency mode in response to alarm signals. Based on the laboratory digital twin model library and the laboratory on-site multidimensional data, it generates dynamic emergency response strategies through the strategy engine and decomposes them into task instructions.

[0094] The mobile interactive assistant 8 is used to receive and push the task instructions to the user, and provide interactive guidance including augmented reality navigation.

[0095] Specifically, the central processing unit, as the core control hub of the entire system, establishes communication connections with the multi-dimensional perception module, the ledger registration module, the laboratory digital twin model library, the cross-validation module, the behavior recognition and intervention module, the emergency response strategy module, and the mobile terminal interactive assistant. These communication connections can be made wirelessly to ensure real-time data interaction and efficient transmission of instructions among the modules.

[0096] Among them, the multi-dimensional sensing module 2 is used to collect multi-dimensional data in the laboratory. Specifically, it may include gas sensors, temperature and humidity sensors, flame detectors, infrared cameras and equipment status sensors deployed in various areas of the laboratory to achieve comprehensive perception of the environment, equipment and personnel activities.

[0097] The ledger registration module 3 is used to automate the recording of ledger information. It can be linked with the hazardous chemical procurement system, the warehousing barcode scanning equipment and the cross-verification module 5 to automatically record the full life cycle data of hazardous chemicals, such as name, specifications, quantity, procurement date, warehousing time, personnel who received the hazardous chemicals, quantity received, return time and scrapping information, replacing the traditional manual recording method and ensuring the accuracy and timeliness of ledger information.

[0098] The laboratory digital twin model library 4 pre-stores digital twin model data of the laboratory, specifically including the scaled-down building topology (such as room layout, corridor direction, door and window positions, etc.), hazardous chemical storage data (such as storage cabinet locations, types and quantities of hazardous chemicals in the cabinets, etc.), experimental equipment data (such as instrument models, installation locations, operating parameter thresholds, etc.), and emergency supplies locations (such as the precise coordinates of fire extinguishers, eyewash stations, first aid kits, emergency sprinkler systems, etc.). Moreover, the model data can be dynamically updated through real-time data from the multi-dimensional sensing module.

[0099] The cross-validation module 5 includes at least two dosage detection sensors and a data fusion analysis unit. The data fusion analysis unit compares the dosage data collected by the two sensors in real time using a preset algorithm. When the deviation between the two is within a preset threshold, the weighted average value is taken as the final dosage value. When the deviation exceeds the threshold, it is determined to be an abnormal state and a diagnostic signal is output. This signal is synchronously transmitted to the central processing unit.

[0100] The behavior recognition intervention module 6 has a built-in behavior recognition algorithm. After receiving the video stream data transmitted by the multi-dimensional perception module, it extracts the operation action features of the experimenter through image recognition technology and compares them with the preset safety operation standard library. When a violation is detected, it immediately triggers active intervention measures, including sending a warning signal to the central processing unit, activating the on-site audible and visual alarm, and displaying operation specification prompts on the experimental operation table display screen.

[0101] The emergency response strategy module 7 is used to respond to alarm signals (which may come from gas concentration exceeding the standard signal from the multi-dimensional sensing module, abnormal status signal from the cross-validation module, high-risk violation signal from the behavior recognition and intervention module, etc.). When an alarm signal is received, the emergency mode is automatically activated. It calls static data such as building topology and emergency material locations from the laboratory digital twin model library, and combines them with dynamic data such as leak location, diffusion concentration, and personnel distribution collected in real time by the multi-dimensional sensing module. Through the built-in strategy engine, it generates a dynamic emergency response strategy adapted to the current scenario, such as delineating a warning area, planning the optimal evacuation route, assigning emergency material retrieval tasks, and assigning rescue tasks to personnel. The strategy is decomposed into specific task instructions, including the executing entity, operation content, time requirements, and location information.

[0102] The mobile interactive assistant 8 can be an application or other mobile device installed on the mobile terminals of managers and laboratory personnel, enabling augmented reality functionality. The mobile interactive assistant communicates with the central processing unit to receive task instructions decomposed by the emergency response strategy module and push them to the corresponding personnel via pop-ups, voice prompts, etc. Simultaneously, by utilizing the mobile terminal's camera and positioning module, combined with data from the laboratory's digital twin model library, it provides augmented reality navigation functionality. This involves overlaying virtual guidance markers, such as arrows indicating the location of emergency supplies, highlighted lines for evacuation routes, and selection boxes for target areas, onto the real-time captured scene, while simultaneously playing voice guidance information to achieve interactive and precise guidance for personnel.

[0103] Under the coordinated control of the central processing unit 1, each module forms an organic whole with interconnected data and linked functions, realizing intelligent management of the entire process of hazardous chemical laboratories from daily supervision to emergency response.

[0104] As can be seen from the above technical solutions, the hazardous chemical laboratory supervision and emergency system provided in this application embodiment takes the central processing unit as the core and integrates seven major modules: multi-dimensional perception, ledger registration, laboratory digital twin model library, cross-validation of use, behavior recognition and intervention, emergency response strategy and mobile terminal interactive assistant, to build a complete closed-loop management system from hazardous chemical use compliance verification, personnel operation safety monitoring to intelligent emergency response.

[0105] In the compliance verification stage of hazardous chemical handling, the handling cross-validation module uses at least two types of dosage detection sensors paired with a data fusion analysis unit to compare and verify the consumption of hazardous chemicals in real time. This not only controls the dosage monitoring error to an extremely low range and accurately outputs the final dosage value that matches the actual consumption, but also quickly identifies violations such as over-disposal and abnormal loss and issues diagnostic signals. This completely solves the problems of inaccurate detection and missed risk assessment by single sensors in existing technologies, ensuring compliance in the handling stage from the source. In the personnel operation safety monitoring stage, the behavior recognition and intervention module relies on video stream data collected by the multi-dimensional perception module to automatically identify violations by laboratory personnel through intelligent algorithms, and can trigger active intervention as soon as a violation is detected. Preventive measures replace the traditional, slow-moving manual inspection model, enabling real-time safety monitoring of the entire operation process and preventing accidents caused by human error. In the intelligent emergency response phase, after receiving alarm signals, the emergency response strategy module combines building topology and material location data from the laboratory's digital twin model library with real-time on-site data from the multi-dimensional sensing module. Through a strategy engine, it dynamically generates emergency response strategies adapted to the current scenario, breaks them down into clear task instructions, and pushes them to relevant personnel via a mobile interactive assistant. Simultaneously, it provides augmented reality navigation to guide personnel to quickly locate emergency supplies, evacuation routes, or disposal areas, solving the shortcomings of slow response and difficult material location in traditional static plans, ensuring efficient and accurate emergency response. Furthermore, the ledger registration module automates the recording of ledger information, providing data support for compliance traceability and emergency response. The central processing unit ensures real-time data linkage between all modules, ultimately forming a closed-loop management system from risk prediction and process supervision to emergency response, comprehensively improving the safety management level of hazardous chemical laboratories.

[0106] In some embodiments of this application, combined with Figure 2 The cross-validation module 5 will be described below.

[0107] The dosage detection sensor combination in the cross-validation module 5 includes a weighing sensor 51, a pressure sensor 52, and a temperature sensor 53.

[0108] The data fusion analysis unit 54 is used to perform:

[0109] Based on the readings of the pressure sensor and temperature sensor, the first dosage value is calculated according to the ideal gas law.

[0110] The second dosage value is calculated based on the change in the reading of the weighing sensor;

[0111] The first dosage value and the second dosage value are compared in real time to determine the dosage difference;

[0112] When the difference in dosage is within a preset reasonable error range, the weighted average value is output as the final dosage value;

[0113] When the difference in dosage exceeds the reasonable error range, and the second dosage value is greater than the first dosage value by more than a first preset value, it is determined to be an abnormal state and a diagnostic signal for container leakage is generated.

[0114] When the difference in dosage exceeds the reasonable error range, and the first dosage value is greater than the second dosage value by more than the second preset value, it is determined to be an abnormal state and a diagnostic signal for pressure sensor failure is generated.

[0115] Specifically, the dosage detection sensor assembly includes a weighing sensor, a pressure sensor, and a temperature sensor. The weighing sensor is a high-precision strain gauge sensor, installed on the bottom support structure of the hazardous chemical storage container, used to collect the total weight data of the container in real time. Its measurement accuracy meets the requirements for sensing minute changes in weight. The pressure sensor is an explosion-proof piezoelectric sensor, integrated at the valve interface of the container, used to detect the real-time pressure value of the medium inside the container, suitable for pressure monitoring of gaseous or gas-liquid mixed hazardous chemicals. The temperature sensor is a contact platinum resistance sensor, attached to the outer wall of the container or inserted into the medium, used to collect real-time temperature data of the hazardous chemicals, and its measurement range is adapted to the storage and reaction temperature range of common hazardous chemicals in the laboratory.

[0116] The data fusion analysis unit has a built-in microprocessor and preset algorithms. Its specific execution logic is as follows: First, for gaseous or gas-liquid mixed hazardous chemicals, the data fusion analysis unit calls the real-time pressure value collected by the pressure sensor and the real-time temperature value collected by the temperature sensor, combines them with the pre-stored fixed volume of the container and the gas constant of the corresponding hazardous chemical, calculates the amount of substance of the medium based on the ideal gas law, and then converts the molar mass of the hazardous chemical into a first dosage value in the form of mass or volume; Second, the data fusion analysis unit simultaneously acquires the initial weight reading and real-time weight reading of the weighing sensor, and calculates the second dosage value by the difference between the two; Third, the first dosage value and the second dosage value are compared in real time, and the absolute difference and relative difference between the two are calculated; Fourth, when the dosage difference is within the preset reasonable error range, the range is... The system is pre-calibrated based on sensor accuracy and hazardous chemical characteristics. The data fusion analysis unit calculates the weighted average of the two values ​​according to preset weights, and outputs this value as the final dosage value to the central processing unit. The weights are dynamically allocated based on the measurement stability of the two values. Fifth, when the dosage difference exceeds the reasonable error range, and the magnitude of the difference between the second and first dosage values ​​exceeds the first preset value, the first preset value is set according to the container sealing standard and the volatility characteristics of the hazardous chemical, and it is determined that the container is leaking, and a diagnostic signal for container leakage is generated. Sixth, when the dosage difference exceeds the reasonable error range, and the magnitude of the difference between the first and second dosage values ​​exceeds the second preset value, the second preset value is set according to the accuracy level of the pressure sensor and the system fault tolerance threshold, and it is determined that the pressure sensor has a reading deviation or malfunction, and a diagnostic signal for pressure sensor malfunction is generated.

[0117] Once the aforementioned abnormal diagnostic signals are generated, they are all transmitted to the central processing unit in real time, which triggers the corresponding early warning mechanism to ensure the accuracy of dosage monitoring and the timeliness of abnormal handling.

[0118] Furthermore, the dosage detection sensor assembly in the cross-validation module also includes an acoustic fingerprint sensor 55 installed at the container valve or pipeline;

[0119] The data fusion and analysis unit is also used to perform:

[0120] The audio stream collected by the acoustic sensor is received and input into a pre-trained acoustic flow analysis model. The acoustic flow analysis model is used to identify specific acoustic patterns that characterize minute gas leaks, and to calculate a third dosage value by identifying the acoustic features of gas flow and establishing a mapping relationship with gas flow.

[0121] The first dosage value, the second dosage value, and the third dosage value are compared in real time.

[0122] When the difference between any two of the first dosage value, the second dosage value and the third dosage value is within the corresponding preset reasonable error range, the weighted average value is output as the final dosage value.

[0123] When the difference between the first dosage value and the second dosage value exceeds the preset reasonable error range, but there is a reasonable item in the first dosage value and the second dosage value whose difference with the third dosage value is within the reasonable error range, then the weighted average of the third dosage value and the reasonable item is used as the final dosage value, and a diagnostic signal for another sensor fault that is not included in the calculation of the final dosage value is generated.

[0124] When the acoustic flow analysis model identifies the specific acoustic pattern, it determines it to be an abnormal state and generates a leakage warning signal.

[0125] Specifically, the dosage detection sensor combination using the cross-validation module also includes an acoustic sensor. This acoustic sensor uses a high-sensitivity microphone array and is integrated and installed in key parts that are prone to leakage or flow changes, such as valve interfaces of hazardous chemical containers and flanges of conveying pipelines. Its sampling frequency is adapted to the acoustic signal frequency band generated by gas flow and minor leaks, and can collect audio flow data of medium flow or leakage in containers and pipelines in real time.

[0126] The data fusion analysis unit adds the following execution logic to the data collected by the acoustic sensor: First, after receiving the audio stream transmitted by the acoustic sensor, the data fusion analysis unit performs noise reduction and feature extraction, such as spectrum analysis and sound pressure level calculation, and inputs the processed acoustic features into a pre-trained acoustic flow analysis model. This acoustic flow analysis model is generated based on a large amount of sample data, which covers the acoustic features of different types of hazardous chemicals under different flow rates and different leakage states (such as micro-slit leaks, valve loosening leaks, etc.). The model learns specific acoustic patterns representing micro-leaks of gas through deep learning algorithms, and establishes a mapping relationship between acoustic feature parameters (such as sound pressure level change rate, peak frequency of the spectrum, etc.) and actual gas flow rate. It can calculate a third dosage value based on the input acoustic features, that is, the dosage based on the flow rate of the acoustic sensor. Second, the data fusion analysis unit performs a real-time comparison of the first dosage value, the second dosage value, and the third dosage value, and calculates the dosage difference between each pair. Third, when the difference in usage between any two of the three parameters is within their respective preset reasonable error ranges (the error range for different sensor combinations is calibrated based on their collaborative accuracy), the data fusion analysis unit calculates the weighted average of the three parameters according to preset weights. The weight allocation is dynamically adjusted based on the measurement stability of the three sensors, and this value is output as the final usage value. Fourth, when the difference between the first and second usage values ​​exceeds the preset reasonable error range, but the difference between one of them and the third usage value is within the reasonable error range, the data fusion analysis unit uses the weighted average of the third usage value and the reasonable value as the final usage value. At the same time, it generates a fault diagnosis signal for the other parameter not included in the calculation, indicating that the sensor may have a measurement deviation. Fifth, when the acoustic flow analysis model identifies a preset specific acoustic pattern from the audio stream, regardless of whether the usage difference is within the reasonable range, the data fusion analysis unit directly determines it as an abnormal state and generates a leakage warning signal. This signal has a higher priority than the normal usage abnormal signal and can trigger a more urgent warning response.

[0127] By introducing an acoustic fingerprint sensor and an acoustic flow analysis model, and using a cross-validation module, tripartite cross-validation of usage monitoring was achieved, which further improved the accuracy of usage calculation and enhanced the early identification capability of minor leaks, reducing the probability of missed risk detection.

[0128] In addition, the cross-validation module may also include a usage compliance verification unit 56;

[0129] The usage compliance verification unit is used to perform:

[0130] Before the hazardous chemical handling operation begins, receive the pre-use amount input by the experimenter;

[0131] After the hazardous chemical handling operation is completed, obtain the final usage value output by the handling cross-validation module;

[0132] If the deviation between the pre-used amount and the final usage value exceeds a preset deviation threshold, it is determined to be an abnormal state, the current operation record is locked, and a verification report is generated.

[0133] Specifically, before the hazardous chemical handling operation begins, the usage compliance verification unit receives pre-entered usage information from the laboratory personnel through an input interface integrated with the mobile interactive assistant or laboratory operating console. This information includes the name of the target hazardous chemical, the expected mass or volume to be handled, and is linked to the laboratory personnel's identity information and authorized scope of operation. After the handling operation is completed, the usage compliance verification unit automatically obtains the final usage value from the data fusion analysis unit and calculates the absolute and relative deviations between the pre-entered and final usage values. A preset deviation threshold is dynamically set based on the hazard level of the hazardous chemical. When the actual deviation exceeds this threshold, the usage compliance verification unit immediately determines it as an abnormal state of non-compliance. On the one hand, it locks the entire record of the current operation, including the operator, time, pre-entered usage, actual usage, and operation video clips; on the other hand, it automatically generates a verification report containing deviation details, anomaly level, and review suggestions, and simultaneously pushes it to the central processing unit and the administrator account of the mobile interactive assistant, awaiting manual review and confirmation.

[0134] Based on this, the behavior recognition intervention module and the cross-validation module form a linkage triggering mechanism, and the specific interaction logic is as follows:

[0135] When the behavior recognition intervention module detects that the experimenter has started to handle hazardous chemicals, it sends a start command to the handling cross-validation module to trigger the handling cross-validation module to perform dosage cross-validation.

[0136] When the cross-validation module determines that the operation is in an abnormal state, it sends a behavior analysis signal to the behavior recognition intervention module, triggering the behavior recognition intervention module to perform video stream analysis on the current operation area and improve the sensitivity of the identification of the violation.

[0137] Specifically, when the behavior recognition intervention module analyzes the video stream from the multi-dimensional perception module and identifies the initial actions of the experimenter in handling hazardous chemicals, such as unlocking the hazardous chemical storage cabinet, approaching the container with a handheld tool, or opening a valve, it immediately sends a start command to the handling cross-validation module. This start command includes information such as the coordinates of the operation area and the target container identifier. Upon receiving the command, the handling cross-validation module immediately activates the weighing sensor, pressure sensor, temperature sensor, and acoustic sensor to enter high-frequency acquisition mode, and simultaneously triggers the data fusion analysis unit to start the real-time calculation process, ensuring that the entire usage monitoring and cross-validation process begins from the start of the handling operation.

[0138] When the cross-validation module determines an abnormal state, such as excessive usage difference, leakage warning, or compliance deviation exceeding the standard, it simultaneously sends a behavior analysis signal to the behavior recognition and intervention module. This signal carries the anomaly type, occurrence time, and associated operation area information. Upon receiving this signal, the behavior recognition and intervention module immediately focuses its video stream analysis on the current operation area. It strengthens monitoring by increasing the sampling frame rate of the camera in that area and enhancing the accuracy of image feature extraction. At the same time, it adjusts the sensitivity of the built-in behavior recognition algorithm, such as lowering the threshold for judging illegal operations and increasing the feature weight of abnormal actions, to achieve refined tracking and risk identification of the current operation, ensuring timely detection and intervention of illegal behaviors related to abnormal usage.

[0139] In some embodiments of this application, combined with Figure 2 The emergency response strategy module 7 will be introduced.

[0140] The emergency response strategy module 7 includes a dynamic risk assessment unit 71;

[0141] The dynamic risk assessment unit is used to perform:

[0142] Construct a multidimensional feature vector for risk assessment. The dimensions of the multidimensional feature vector for risk assessment include at least the hazard index of the leaked substance, the estimated leakage intensity based on the output of the cross-validation module, the ventilation level of the laboratory space, and the number of trapped personnel in real time obtained through the behavior recognition and intervention module.

[0143] The risk assessment multidimensional feature vector is input into a pre-trained risk level assessment model, which is generated based on historical accident data and simulated data, and outputs a quantified comprehensive risk level.

[0144] The strategy engine triggers a warning information multicast protocol that matches the overall risk level. The warning information multicast protocol defines the target notification department, personnel priority set, and information content template.

[0145] Specifically, the emergency response strategy module, centered on the dynamic risk assessment unit, achieves precise quantification of emergency risks and targeted delivery of early warning information through multi-dimensional data fusion and intelligent model analysis. The dynamic risk assessment unit first constructs a multi-dimensional feature vector for risk assessment. Data for each dimension of the vector comes from real-time feedback from other modules within the system, ensuring data authenticity and timeliness. Among them, the "Leaked Substance Hazard Index" dimension combines parameters such as the substance's toxicity, flammability, corrosivity, and reactivity, and generates a quantitative index through a preset weighting algorithm. The higher the index, the stronger the inherent hazard of the substance. The "Leakage Intensity" dimension estimates the leakage amount per unit time based on the real-time dosage difference output by the cross-validation module, the leakage duration, and the leakage concentration gradient data collected by the sensor, forming a graded quantitative value. The "Laboratory Space Ventilation Level" dimension is divided into multiple levels and assigned corresponding quantitative values ​​by connecting to the laboratory ventilation system's operating status sensor and calculating the air exchange rate based on the space volume. The lower the level, the faster the leakage substance spreads. The "Real-time Number of Trapped Personnel" dimension relies on the video stream analysis results of the behavior recognition intervention module, and accurately counts the number of personnel in the leakage area and the surrounding affected area through human contour detection and personnel counting algorithms, while marking the personnel's location to assist in subsequent evacuation decisions.

[0146] After the feature vector is constructed, the dynamic risk assessment unit inputs it into the pre-trained risk level assessment model. The training data of this model covers two parts: first, historical accident data, such as the type of substance, leakage amount, on-site environment, casualties and handling results of past hazardous chemical leaks; second, data from different scenarios simulated by digital twin technology, such as the risk evolution process under different leakage intensities, ventilation conditions and personnel densities. The model is trained using deep learning algorithms, and through repeated iterations to optimize parameters, it finally has the ability to output a quantitative comprehensive risk level based on the input multi-dimensional feature vector. The comprehensive risk level is usually presented as a level 1-5, with level 1 being low risk and level 5 being extremely high risk. Each level corresponds to a clear risk impact range. For example, level 1 only affects a single experimental platform, while level 5 may affect the entire laboratory floor.

[0147] Once the overall risk level is determined, the strategy engine immediately triggers the matching early warning information multicast protocol. This protocol is a pre-configured set of standardized rules. For example, the "Target Notification Department" is divided according to risk level: low risk (levels 1-2) only notifies the internal management department of the laboratory (such as laboratory administrators and safety specialists), medium risk (level 3) additionally notifies the school / enterprise safety management department, and high risk (levels 4-5) simultaneously triggers the notification mechanism of external emergency departments (such as local fire and rescue agencies and medical emergency centers). The "Personnel Priority Set" is prioritized according to responsibilities and distance. For example, level 1 risk prioritizes notifying on-site laboratory personnel and administrators, level 5 risk prioritizes notifying the emergency response team and on-site supervisors, and then notifies ordinary laboratory personnel and personnel in the surrounding area in turn. The "Information Content Template" dynamically adjusts the level of detail according to the risk level. The low-risk template includes the name of the leaked substance, the location of the leak, and preliminary handling suggestions. In addition to basic information, the high-risk template also needs to add the real-time diffusion range, evacuation routes, location of emergency supplies, and contact person's phone number. All templates support simultaneous push through multiple channels such as mobile interactive assistants, laboratory on-site sound and light alarm systems, and SMS platforms to ensure that relevant personnel obtain accurate early warning information in the shortest possible time, buying time for subsequent emergency response.

[0148] Through the above logic, the dynamic risk assessment unit automates the entire process from risk data collection and quantitative assessment to early warning push, avoiding the template-based early warning mode in traditional emergency plans and significantly improving the accuracy and efficiency of emergency response.

[0149] In response to a widespread leak alert, the execution process of the policy engine 72 includes:

[0150] The hazardous chemicals storage data and experimental equipment data in the laboratory digital twin model library are called up, and based on the predefined experimental equipment asset value weight and hazardous chemical hazard weight, an equipment priority sequence and a hazardous chemical priority sequence are constructed.

[0151] Based on the multidimensional data simulation and prediction from the laboratory site, leakage diffusion simulation data is obtained. Combined with the equipment priority sequence and the hazardous chemical priority sequence, a dynamic emergency response strategy including equipment transfer sequence and hazardous material isolation scheme is generated.

[0152] The dynamic emergency response strategy is converted and decomposed into task instructions, which are then pushed to the mobile interactive assistants of the corresponding personnel. The mobile interactive assistants provide the optimal action path to the target equipment location or the target hazardous chemical storage location.

[0153] Specifically, the strategy engine first calls upon pre-stored hazardous chemical storage data and experimental equipment data in the laboratory digital twin model library to initiate a priority ranking algorithm. The construction of the equipment priority sequence is based on predefined equipment asset value weights, which comprehensively consider the equipment's acquisition cost, relevance to the research task, and irreplaceability. A value score is generated for each piece of equipment, and these scores are ranked from highest to lowest to form the equipment priority sequence. The hazardous chemical priority sequence is constructed based on the hazardous chemical hazard weights. These weights reference the GHS classification standards and combine the substance's toxicity level, flammability and explosiveness level, reactivity level, and diffusion rate after leakage to calculate the hazard score for each hazardous chemical. These scores are then ranked from highest to lowest to form the hazardous chemical priority sequence, ensuring that high-risk materials are handled first.

[0154] Subsequently, the strategy engine receives real-time data collected by the multi-dimensional perception module, including the location of the leak source, the current leak concentration, ventilation system operating parameters, and indoor airflow direction. Combined with the building topology in the laboratory's digital twin model library, including room partitions, passage widths, and door and window distribution, it uses a built-in fluid dynamics simulation algorithm to predict leak diffusion simulation data for the next 5-30 minutes, including diffusion boundaries, concentration gradient distribution, and the time points for arrival at each area. Based on this simulation data, the strategy engine maps the equipment priority sequence and the hazardous chemical priority sequence to the diffusion path: for equipment, priority is given to equipment located at the front of the diffusion path (i.e., the equipment that will be reached first and has a high priority), generating an equipment transfer sequence that specifies the transfer order, such as transferring core precision instruments first, followed by conventional equipment; the target location for transfer, such as a safe storage area far from the leak source; and the maximum capacity for a single transfer. For hazardous materials, for high-priority hazardous chemicals that have not leaked but are within the range of diffusion impact, a hazardous material isolation plan is generated, including isolation methods, such as activating explosion-proof isolation enclosures and closing area isolation valves; isolation sequence, such as isolating highly reactive substances first and then isolating highly toxic substances; and the required isolation tools, such as the location information of chemical protective gloves and sealing caps.

[0155] Finally, the strategy engine categorizes dynamic emergency response strategies into equipment transfer, material isolation, and path guidance groups based on task type, and decomposes them into structured task instructions. Each instruction includes the executing entity, matched with personnel skill tags (e.g., assigning personnel with chemical protection training qualifications to isolate high-risk materials); the object of operation, i.e., the unique identifier of the specific equipment or hazardous chemical; the operation steps, such as "disconnect the equipment connection line → use a special cart to transfer to safety zone 3"; the completion deadline, which is based on the time nodes of the diffusion simulation; and safety precautions, such as the protective equipment that must be worn. These task instructions are simultaneously pushed to the corresponding personnel's mobile interactive assistant through the central processing unit. The mobile interactive assistant calls on real-time data from the laboratory's digital twin model library, combines the current personnel location with the boundary of the hazardous area in the diffusion simulation, and generates the optimal action path through a path planning algorithm. This path avoids high-concentration diffusion areas, prioritizes wide passages and emergency passages, and overlays path guidance signs, such as arrows and highlighted routes, in the form of AR navigation on the real-time screen. At the same time, it updates the risk changes on the path in real time, such as prompting "Concentration increases 5 meters ahead, detour recommended," to ensure that personnel execute the response tasks efficiently and safely.

[0156] The process by which the emergency response strategy module assigns the task instructions to the execution of each mobile terminal interactive assistant includes:

[0157] Determine the task type, required skills, and task location corresponding to the task instruction, and match the optimal executor from among the online personnel. The matching algorithm takes into account the personnel's skill level, the distance between the real-time location and the task point, and the current task load.

[0158] The task instructions are pushed to the mobile interactive assistant of the optimal executor;

[0159] The system receives the task execution status of the task instruction returned by the mobile interactive assistant of the optimal executor, and re-triggers the optimal executor matching process when the task execution times out or fails.

[0160] Specifically, the strategy engine first performs structured parsing of task instructions to clarify the core attributes corresponding to the task. Task types are divided into categories based on the operation content, such as equipment transfer, hazardous material isolation, personnel evacuation guidance, and emergency material retrieval. Different types correspond to different skill requirements. For example, equipment transfer tasks require precision instrument operation skills, while hazardous material isolation tasks require chemical protection operation qualifications. The required skills are divided into three levels of proficiency: basic, intermediate, and professional. Professional-level skills are only matched with personnel who have undergone specialized training and certification. The task location is converted into precise spatial positioning information, such as the number of a certain experimental platform or the area identifier of a certain storage cabinet, through the coordinate system of the laboratory digital twin model library.

[0161] Based on the above attributes, the strategy engine initiates the optimal executor matching algorithm. This algorithm filters candidates from the system's online personnel database (including the real-time status of all experimenters, administrators, and emergency response personnel). The filtering process integrates three core indicators: personnel skill level (assigned based on the match between the required skill level and the personnel's actual skill level; higher match rate, higher score); distance between real-time location and task location (obtained through the location function of the mobile interactive assistant, calculating the straight-line distance and reachable path length to the task location; shorter distance, higher score); and current task load (counting the number of tasks the personnel have received but not completed; lower load, higher score). The three indicators are weighted according to a preset score, and the candidate with the highest score is determined as the optimal executor.

[0162] The strategy engine then pushes the task instructions in encrypted form to the mobile interactive assistant of the optimal executor. Push methods include pop-up notifications, vibration alerts, and voice announcements to ensure timely awareness. In addition to the task's operational requirements, the instructions also include task priority, estimated time, associated safety notices, and emergency contact information.

[0163] After receiving instructions, the mobile assistant provides real-time feedback on the task execution status to the strategy engine. The strategy engine has a built-in status monitoring timer with preset timeout thresholds based on task complexity. When the timer reaches the threshold and the task is still in the "in execution" state, or when a "failed execution" feedback is received, the strategy engine immediately re-triggers the optimal personnel matching process. At this point, the algorithm excludes the original personnel and adjusts the candidate pool based on real-time conditions, such as expanding it to include staff in adjacent laboratories. It then recalculates the comprehensive matching score to determine a new optimal personnel, ensuring that task instructions are always effectively executed and preventing interruptions in emergency response due to single-point failures.

[0164] Furthermore, the emergency response strategy module also includes a protective equipment planning unit 73, which uses knowledge graph association and spatial algorithm planning to ensure that emergency response personnel are equipped with appropriate protective equipment and tools before performing their tasks.

[0165] The protective equipment planning unit is used to execute:

[0166] When generating the task instruction, the material task association knowledge graph is accessed, which defines the standard protective equipment and handling tools required to handle different police situations.

[0167] Based on the real-time location of personnel, the nearest target material point with the required standard protective equipment and disposal tools is dynamically calculated from the emergency material points in the digital twin model library using the nearest neighbor search algorithm.

[0168] The target material point is inserted as a preceding task node into the personnel's original task sequence to obtain a composite task chain, and a globally optimal action route is generated for the composite task chain.

[0169] Specifically, the protective equipment planning unit starts working simultaneously with the task instructions generated by the strategy engine, first accessing a pre-built knowledge graph of material task associations. This knowledge graph uses the type of emergency as its core node, mapping the standard protective equipment and handling tools required to handle that type of emergency through semantic associations. For example, for an emergency involving acidic hazardous chemicals, the knowledge graph defines standard protective equipment including chemical protective suits, acid and alkali resistant gloves, and goggles; and handling tools including neutralizing agent sprayers, absorbent cotton, and explosion-proof shovels. For an emergency involving flammable gas leaks, it associates equipment such as antistatic suits, gas masks, and gas detectors, as well as tools such as explosion-proof wrenches and fire blankets. The knowledge graph also includes descriptions of the applicable scenarios for the equipment (e.g., the protection level of chemical protective suits is adapted to different concentrations of leaks) and the operational compatibility of the tools (e.g., a specific model of wrench corresponds to a specific valve specification), ensuring the accuracy of the associated content.

[0170] Subsequently, the protective equipment planning unit obtains the real-time location information of the personnel executing the task through the central processing unit. This location information is provided by the location module of the mobile interactive assistant, and simultaneously accesses emergency supply location data from the laboratory's digital twin model library. This location data includes the precise spatial coordinates of each emergency supply storage point, as well as the type, quantity, and condition of the protective equipment and handling tools currently stored at each point (updated in real-time by supply sensors). Based on this data, the protective equipment planning unit initiates a nearest neighbor search algorithm. During the calculation process, the algorithm considers not only straight-line distance but also the building topology in the laboratory's digital twin model library to filter reachable paths. It also verifies whether the point stores the complete set of standard protective equipment and handling tools required for the task, ultimately dynamically determining the target supply point that is closest to the personnel's current location and has all the necessary supplies.

[0171] After identifying the target material location, the protective equipment planning unit inserts it as a prerequisite task node into the personnel's original task sequence, forming a composite task chain of "receiving protective equipment → executing core tasks." For example, if the original task sequence is "transferring the precision instruments on Experiment 3 to the safe area," after inserting the prerequisite node, the composite task chain becomes "go to the emergency cabinet in Area A to retrieve chemical protective equipment → transfer the precision instruments on Experiment 3 to the safe area." Next, the unit invokes a path optimization algorithm, comprehensively considering the spatial locations of the target material location and the core task point, leakage diffusion simulation data, and passageway efficiency, to calculate the globally optimal action route covering the entire composite task chain. This route includes not only the path from the personnel's current location to the target material location but also the path from the target material location to the core task point. Key nodes are marked on the route, and the estimated time for each segment is generated simultaneously, ensuring that personnel complete the seamless operation of receiving protective equipment and executing core tasks in the shortest time and with the lowest risk.

[0172] Ultimately, the complex task chain and the globally optimal action route are pushed to the personnel's mobile interactive assistant through the central processing unit. The assistant overlays the route onto the real-time scene in the form of AR navigation and prompts the personnel with the list of equipment to be collected when approaching the target material point. This ensures that the personnel can efficiently complete the preparatory work before executing the core task, thereby improving the safety and standardization of emergency response.

[0173] In some embodiments of this application, combined with Figure 2 This section introduces the mobile interactive assistant 8.

[0174] In emergency mode, the mobile interactive assistant provides intuitive and safe action guidance to personnel through the collaboration of 3D modeling, precise positioning, dynamic path planning, and augmented reality navigation. The execution process of the mobile interactive assistant in emergency mode includes:

[0175] Load and render a 3D mesh model based on the aforementioned laboratory digital twin model library;

[0176] By integrating data from UWB positioning tags worn by personnel with data from inertial measurement units, real-time positioning and attitude mapping of personnel are achieved in the three-dimensional mesh model.

[0177] Based on the building topology in the digital twin model library, and with the dangerous areas detected by the multi-dimensional perception module being introduced in real time as dynamic obstacles, a path finding algorithm is used to calculate the optimal evacuation route in real time. The cost function of the path finding algorithm integrates the path length, the estimated travel time, and the real-time perceived environmental risk value.

[0178] The optimal evacuation route is overlaid onto the real-time image of the terminal camera in an augmented reality manner to form a visual navigation guide.

[0179] Specifically, after receiving the emergency mode command triggered by the emergency response strategy module, the mobile interactive assistant first retrieves a 3D mesh model matching the current laboratory scene from the laboratory digital twin model library. This model is constructed based on data such as building topology, equipment layout, and channel distribution, containing spatial details with millimeter-level precision. It is also optimized for mobile computing power through lightweight processing to ensure smooth rendering on the terminal. The model can display the location markers of static facilities in real time, while reserving dynamic data access interfaces to provide a basic framework for subsequent positioning and path overlay.

[0180] To achieve accurate mapping of personnel within the 3D mesh model, the mobile interactive assistant integrates two types of core data: first, real-time spatial coordinate data transmitted from a UWB positioning tag worn by the person. This tag, communicating with a UWB base station deployed in the laboratory, provides centimeter-level positioning accuracy, pinpointing the person's exact location in both the planar and vertical directions; second, motion data collected by the terminal's built-in inertial measurement unit (IMU), including acceleration, angular velocity, and magnetic field information, used to capture changes in posture such as walking stride frequency, turning angle, and climbing stairs. By fusing and calibrating these two types of data using a Kalman filter algorithm, the IMU data maintains positioning continuity even when UWB signals fluctuate due to obstruction. Ultimately, it generates a virtual posture mapping in the 3D mesh model that is completely synchronized with the person's actual actions, such as positional movement during walking and directional adjustments during turns.

[0181] In the path planning phase, the mobile assistant uses the building topology in a 3D mesh model as its basic framework, including static spatial information such as passage connectivity, door control status, and the locations of stairs and elevators. Simultaneously, it receives real-time hazardous area data pushed by a multi-dimensional perception module. This data, collected by devices such as gas sensors, flame detectors, and temperature sensors, is analyzed by the central processing unit and classified as dynamic obstacles. For example, an area is marked as a high-risk zone due to excessive gas concentration, and a passage is marked as impassable due to structural damage, with different attributes assigned according to risk level. Based on this information, the mobile assistant initiates a path-finding algorithm. Its core cost function integrates three key indicators: path length, estimated travel time, and real-time environmental risk value. The optimal evacuation route for the current scenario is generated through dynamic iterative calculation, and the path is automatically recalculated at preset intervals or when a change in hazardous area is detected to adapt to dynamic risks.

[0182] Ultimately, the mobile interactive assistant overlays the optimal evacuation route onto the real-time footage captured by the terminal's camera using augmented reality. Specifically, this manifests as: highlighted 3D lines overlaid on the real-time scene, with the line width dynamically adjusting to the width of the passageway; arrow markers displayed at key points such as turns and forks in the road, accompanied by voice prompts; estimated remaining distance and time to reach the safe zone indicated next to the path; and emergency supply points along the route highlighted as floating icons. Simultaneously, the system adjusts the position and size of the AR overlay in real-time based on the movement speed and perspective of the personnel, ensuring that the navigation guidance is always precisely aligned with the actual scene. Even with rapid movement or frequent turns, the path direction can be clearly identified, effectively improving the efficiency and safety of emergency evacuation.

[0183] The mobile terminal interactive assistant has a built-in multimodal interaction engine 81, which includes at least a voice interaction unit and a video stream scheduling unit.

[0184] The voice interaction unit is used to parse user voice commands and recognize user feedback data on site, and at the same time broadcast the task commands and navigation prompts to the user.

[0185] The video stream scheduling unit is used to invoke and present the real-time video stream of a specific camera in the multi-dimensional perception module according to the user's voice command or the task command.

[0186] Based on this, the emergency response strategy module is also used to execute:

[0187] Receive the user's on-site feedback data uploaded by the mobile interactive assistant;

[0188] When the user's on-site feedback data is inconsistent with the assumptions of the current dynamic emergency response strategy, the strategy adjustment mechanism is triggered;

[0189] The strategy adjustment mechanism, based on the user's on-site feedback data, uses the strategy engine to make real-time corrections to the dynamic emergency response strategy and reassign updated task instructions.

[0190] Specifically, the voice interaction unit integrates a natural language processing module and a speech synthesis engine, enabling real-time parsing of user-issued voice commands. After a user initiates a request through colloquial expression, the unit first performs noise reduction and feature extraction on the speech signal, then matches the intent using a pre-trained command recognition model, transforming ambiguous expressions into structured commands. Simultaneously, the unit can recognize user feedback data, including confirmation of task execution status and reports of on-site anomalies, and converts this feedback into standardized data formats for uploading to the emergency response strategy module. Furthermore, the voice interaction unit uses real-time speech synthesis technology to broadcast information to the user, including key content of task commands, navigation prompts, and risk warnings. The broadcast volume and speed can be automatically adjusted according to ambient noise to ensure audibility in noisy environments.

[0191] The video stream scheduling unit, acting as a bridge between the multi-dimensional perception module and the mobile terminal, possesses precise camera access and video presentation capabilities. Upon receiving a user voice command or a task command pushed by the emergency response strategy module, the unit first matches the corresponding unique camera identifier in the laboratory digital twin model library based on the spatial identifier in the command, and then sends a video stream retrieval request to that camera through the central processing unit. During video stream transmission, the unit employs adaptive bitrate technology to dynamically adjust the video clarity based on the mobile terminal's network status and supports split-screen presentation of multi-camera video streams. For critical operational scenarios, the unit can also automatically activate the video recording function, synchronously uploading the footage to the system backend for storage as a basis for subsequent review.

[0192] The emergency response strategy module continuously receives on-site user feedback data uploaded by the mobile interactive assistant through the central processing unit. This data covers abnormal situations actively reported by users, obstacles in task execution, and subjective descriptions of environmental conditions, and is corroborated with sensor data collected by the system in real time.

[0193] When the strategy engine detects a conflict between the user's on-site feedback data and the assumptions of the current dynamic emergency response strategy, it immediately triggers the strategy adjustment mechanism. For example, the original strategy assumes that "the passage in area E is unobstructed" to plan the equipment transfer route, but the user reports that "area E is impassable due to the accumulation of items," or the original strategy formulates a leak handling plan based on "sufficient stock of a certain type of neutralizing agent," but the user reports that "the corresponding materials have been exhausted." These are all cases where the assumptions are inconsistent.

[0194] Once the adjustment mechanism is triggered, the strategy engine uses user feedback data as key input, re-invokes data such as building topology and material locations from the laboratory's digital twin model library, and combines this with real-time monitoring results from the multi-dimensional sensing module to make targeted corrections to the original dynamic emergency response strategy: if the conflict is related to path, a new path is replanned to avoid the congested area; if the conflict is related to materials, available alternative materials are replaced and the handling procedures are adjusted; if the environmental risk assessment is flawed, the comprehensive risk level is recalculated and the warning range is updated. After the corrected strategy is verified by the strategy engine, it is broken down into updated task instructions and redistributed to the mobile interactive assistants of relevant personnel through the central processing unit, along with explanations of the strategy adjustments, ensuring that personnel are promptly informed of the changes and execute the new instructions, achieving dynamic adaptation of the emergency response strategy to the actual situation on site.

[0195] In some embodiments of this application, combined with Figure 2 The following section introduces module 3 of the ledger registration module.

[0196] As the data core of hazardous chemicals' full lifecycle management, the ledger registration module, through the collaboration of the registration data collection unit and the electronic ledger generation unit, realizes the automated and structured recording of hazardous chemicals' usage information, completely replacing the inefficiency and errors of traditional manual ledgers. Its specific structure and execution logic are as follows:

[0197] The ledger registration module 3 includes a registration data acquisition unit 31 and an electronic ledger generation unit 32;

[0198] The registration data collection unit is used to obtain the final usage value from the cross-validation module, automatically obtain the operator's identity and operation time period from the behavior recognition and intervention module, and obtain the corresponding experimental project information from the enterprise database.

[0199] The electronic ledger generation unit is used to structurally integrate the obtained final usage value, the operator's identity, the operation time period, and the experimental project information to generate electronic ledger records and register them in the hazardous chemicals electronic ledger.

[0200] Specifically, the registration data acquisition unit serves as the data input terminal of the module. By establishing real-time communication links with multiple modules within the system and external databases, it enables fully automated collection of key information without the need for manual intervention. First, this unit maintains high-frequency data interaction with the cross-validation module, acquiring the final usage value corresponding to each hazardous chemical retrieval operation in real time. This includes not only the difference in usage before and after retrieval but also key parameters and abnormal status records during the usage calculation process, ensuring the traceability of usage data. Second, the unit automatically extracts operator identity and operation time period information from the behavior recognition intervention module: operator identity is obtained through video stream analysis by the behavior recognition intervention module, which determines a unique identity through facial feature comparison or work badge recognition; the operation time period accurately records the complete time interval from the behavior recognition intervention module's detection of the "retrieval start action" to the "retrieval termination action," while also marking whether there are any special nodes such as pauses or interruptions during the process. Third, the unit connects to the company's internal database through an interface, automatically matching and obtaining the corresponding experimental project information using "operator identity" or "operation time period" as key fields. This includes the experimental project name, project number, project leader, purpose of hazardous chemical retrieval, and the corresponding approval number, ensuring that the ledger information is directly linked to the experimental task and meeting regulatory audit requirements.

[0201] The electronic ledger generation unit is responsible for converting the collected, scattered information into standardized electronic ledger records and completing registration and storage. This unit first pre-defines a structured field system for the hazardous chemicals ledger, covering categories such as unique ledger identifier, basic hazardous chemical information, usage data, operator information, operation time information, experimental project association information, and data source identifier. Subsequently, the unit maps each type of information acquired by the registration data collection unit to its corresponding field according to the pre-defined field logic, automatically completing data integration. If an abnormal data format is detected, a re-collection request is immediately sent to the registration data collection unit to ensure the completeness and accuracy of the ledger records.

[0202] After structured integration, the electronic ledger generation unit registers each generated electronic ledger record to the hazardous chemicals electronic ledger database in real time. This database supports the creation of a search catalog based on multi-dimensional indexes, facilitating subsequent queries and statistics. Simultaneously, each record is accompanied by a data modification log, ensuring the authenticity and immutability of the ledger data and meeting the core regulatory requirements of the industry for traceability and auditability of hazardous chemicals usage records.

[0203] Furthermore, the ledger registration module 3 also includes an access control unit 33;

[0204] The permission linkage control unit is used to execute:

[0205] The system receives in real time the abnormal status diagnosis signal sent by the cross-validation module and the abnormal usage event sent by the usage compliance verification unit.

[0206] When any of the aforementioned abnormal status diagnostic signals or the aforementioned abnormal dosage event is received, the corresponding operator's identity is determined by querying the currently associated electronic ledger record;

[0207] Based on the operator's identity, an access freeze command is generated and executed, and an access freeze event log containing the abnormal status diagnostic signal or the usage abnormal event is generated.

[0208] Specifically, the access control unit maintains real-time communication with the cross-validation module and the usage compliance verification unit, continuously receiving two types of key anomaly information: first, anomaly status diagnostic signals sent by the cross-validation module, including anomalies related to usage monitoring such as container leaks and sensor malfunctions; second, usage anomaly events sent by the usage compliance verification unit, i.e., non-compliant usage situations where the deviation between the pre-used amount and the final usage value exceeds a preset threshold. These anomaly messages are all accompanied by a unique event identifier, the time of the anomaly, and the associated hazardous chemical container or operating area identifier, providing basic data for subsequent traceability.

[0209] When the access control unit receives any of the aforementioned abnormal signals or events, it immediately initiates the operator identification locking process. This unit uses the event identifier or timestamp in the abnormal information to perform a precise search within the hazardous chemicals electronic ledger maintained by the electronic ledger generation unit, matching the corresponding electronic ledger record. Since the electronic ledger record already stores the operator's identification identifier in a structured manner, this identifier can be directly extracted as the associated object, ensuring the accuracy of the locked object.

[0210] Based on the extracted operator identification, the access control unit generates an access freeze command. This command specifies the freeze scope, dynamically adjusted according to the severity of the anomaly: for minor anomalies (such as a single dosage deviation slightly exceeding a threshold), the operator's access to similar hazardous chemicals is frozen; for serious anomalies (such as container leaks or repeated dosage violations), access to all hazardous chemical storage areas, storage cabinet unlocking permissions, and laboratory workbench activation permissions are frozen. After the command is generated, it is synchronized to the laboratory's access control system via the central processing unit, taking effect immediately to prevent the operator from continuing potentially risky operations.

[0211] Simultaneously, the access control unit automatically generates an access freeze event log. The log contains complete anomaly details: anomaly type, precise time of occurrence, associated hazardous chemical information, the identity and access scope of the frozen operator, and the execution status of the freeze command. This log is stored in conjunction with the corresponding electronic ledger record. After the anomaly event has been manually reviewed and confirmed to be resolved, administrators can manually unfreeze it through the access control system. This unfreezing operation is also recorded in the event log, forming a complete access control trajectory.

[0212] In some embodiments of this application, combined with Figure 2 The behavior recognition and intervention module 6 will be introduced.

[0213] The behavior recognition and intervention module 6 includes:

[0214] The video analysis unit 61 is used to decode and perform behavior sequence analysis on the real-time video stream acquired by the multi-dimensional perception module in order to identify the operational behavior of the experimenters.

[0215] Violation Pattern Library 62 stores violation operation pattern characteristics defined based on historical violation cases;

[0216] The intervention triggering unit 63 is used to generate and execute an intervention instruction when the degree of matching between the identified behavior sequence and any of the features of the violation operation mode exceeds a preset threshold.

[0217] Specifically, the video analysis unit is responsible for end-to-end processing and behavior analysis of the real-time video stream acquired by the multi-dimensional perception module. This unit first decodes the incoming video stream and initiates a hierarchical behavior analysis process: the first layer is object detection, which uses a deep learning model to accurately extract the human contours, hand positions, and tool coordinates of the experimenter from the video frames, while filtering out background interference; the second layer is feature extraction, which tracks the target motion trajectory in consecutive frames, extracts the spatiotemporal features of the actions, and forms structured behavior sequence data; the third layer is behavior semanticization, which maps low-order action features to high-order operation behavior descriptions, providing understandable behavior labels for subsequent violation comparisons.

[0218] The violation pattern library serves as a benchmark for judging behavioral compliance, storing standardized violation operation pattern features built upon massive historical violation cases and industry safety standards. These features exist in structured data form, covering typical violation types under different operational scenarios: for example, protective equipment violation patterns include "facial features when not wearing goggles," "torso contour features when not wearing chemical protective clothing," and "hand posture features of improper glove wearing"; operation sequence violation patterns include "reverse action sequence features of strong acid being poured into water" and "step features of disassembling pipelines without first closing valves"; hazardous materials handling violation patterns include "container contact features of improperly mixed mutually exclusive chemicals" and "pouring angle and location features of carelessly discarded waste liquid." Each feature in the pattern library is accompanied by a matching weight and supports dynamic updates. When the system adds new violation cases, the new violation patterns are automatically added to the library through manual annotation and feature extraction algorithms, ensuring the ability to identify new violations.

[0219] The intervention triggering unit, acting as the execution terminal for behavior monitoring, is responsible for generating and executing targeted intervention instructions based on behavior matching results. This unit receives behavior sequences and semantic tags output by the video analysis unit in real time, compares them one by one with features in the violation pattern library using a similarity calculation algorithm, and outputs a quantified matching value. Preset thresholds are set according to the risk level of the violation. When the matching degree exceeds the corresponding threshold, the unit immediately initiates the intervention process: for general violations (such as improper wearing of protective equipment), a mild intervention instruction is generated, including triggering the audible and visual alarms in the operating area and displaying a diagram of the violation and correct operating instructions on the experimental operating table display screen; for high-risk violations (such as improper mixing of explosive chemicals or failure to dilute strong acids according to procedures), an emergency intervention instruction is generated. In addition to strengthening the audible and visual alarms, it also links relevant equipment through the central processing unit and simultaneously pushes emergency warning information to the laboratory administrator's mobile interactive assistant. All execution records of intervention instructions are synchronously stored in the system log.

[0220] Through the collaboration of three-level units, the behavior recognition and intervention module has achieved full automation of the process from video acquisition, behavior analysis, violation judgment to proactive intervention, effectively making up for the lag and subjectivity of manual inspection and significantly reducing the safety risks caused by operational violations.

[0221] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, system, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, system, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, system, article, or apparatus that includes said element.

[0222] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0223] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A hazardous chemical laboratory monitoring and emergency response system, characterized in that, It includes a central processing unit, and a multi-dimensional perception module, a ledger registration module, a laboratory digital twin model library, a cross-validation module, a behavior recognition and intervention module, an emergency response strategy module, and a mobile interactive assistant, all of which are respectively connected to the central processing unit in communication. The multidimensional sensing module is used to collect multidimensional data from the laboratory site; The ledger registration module is used to automate the recording of ledger information; The laboratory digital twin model library contains pre-stored laboratory model data including building topology, hazardous chemical storage data, experimental equipment data, and emergency material locations. The cross-validation module uses at least two dosage detection sensors and a data fusion analysis unit to perform real-time comparison and cross-validation of hazardous chemical consumption, and outputs the final dosage value and abnormal status diagnosis signal. The behavior recognition and intervention module analyzes the video stream from the multi-dimensional perception module to identify the experimenter's violation of regulations and trigger active intervention. The emergency response strategy module is used to activate the emergency mode in response to alarm signals. Based on the laboratory digital twin model library and the laboratory on-site multidimensional data, it generates dynamic emergency response strategies through the strategy engine and decomposes them into task instructions. The mobile interactive assistant is used to receive and push the task instructions to the user, and provide interactive guidance including augmented reality navigation; The dosage detection sensor combination in the cross-validation module includes a weighing sensor, a pressure sensor, and a temperature sensor; The data fusion and analysis unit is used to perform: Based on the readings of the pressure sensor and temperature sensor, the first dosage value is calculated according to the ideal gas law. The second dosage value is calculated based on the change in the reading of the weighing sensor; The first dosage value and the second dosage value are compared in real time to determine the dosage difference; When the difference in dosage is within a preset reasonable error range, the weighted average value is output as the final dosage value; When the difference in dosage exceeds the reasonable error range, and the second dosage value is greater than the first dosage value by more than a first preset value, it is determined to be an abnormal state and a diagnostic signal for container leakage is generated. When the difference in dosage exceeds the reasonable error range, and the first dosage value is greater than the second dosage value by more than the second preset value, it is determined to be an abnormal state and a diagnostic signal for pressure sensor failure is generated. The dosage detection sensor assembly in the cross-validation module also includes an acoustic fingerprint sensor installed at the container valve or pipeline. The data fusion and analysis unit is also used to perform: The audio stream collected by the acoustic sensor is received and input into a pre-trained acoustic flow analysis model. The acoustic flow analysis model is used to identify specific acoustic patterns that characterize minute gas leaks, and to calculate a third dosage value by identifying the acoustic features of gas flow and establishing a mapping relationship with gas flow. The first dosage value, the second dosage value, and the third dosage value are compared in real time. When the difference between any two of the first dosage value, the second dosage value and the third dosage value is within the corresponding preset reasonable error range, the weighted average value is output as the final dosage value. When the difference between the first dosage value and the second dosage value exceeds the preset reasonable error range, but there is a reasonable item in the first dosage value and the second dosage value whose difference with the third dosage value is within the reasonable error range, then the weighted average of the third dosage value and the reasonable item is used as the final dosage value, and a diagnostic signal for another sensor fault that is not included in the calculation of the final dosage value is generated. When the acoustic flow analysis model identifies the specific acoustic pattern, it determines it to be an abnormal state and generates a leakage warning signal.

2. The system according to claim 1, characterized in that, The cross-validation module also includes a usage compliance verification unit; The usage compliance verification unit is used to perform: Before the hazardous chemical handling operation begins, receive the pre-use amount input by the experimenter; After the hazardous chemical handling operation is completed, obtain the final usage value output by the handling cross-validation module; If the deviation between the pre-used amount and the final usage value exceeds a preset deviation threshold, it is determined to be an abnormal state, the current operation record is locked, and a verification report is generated.

3. The system according to claim 2, characterized in that, When the behavior recognition intervention module detects that the experimenter has started to handle hazardous chemicals, it sends a start command to the handling cross-validation module to trigger the handling cross-validation module to perform dosage cross-validation. When the cross-validation module determines that the operation is in an abnormal state, it sends a behavior analysis signal to the behavior recognition intervention module, triggering the behavior recognition intervention module to perform video stream analysis on the current operation area and improve the sensitivity of the identification of the violation.

4. The system according to claim 1, characterized in that, The emergency response strategy module includes a dynamic risk assessment unit; The dynamic risk assessment unit is used to perform: Construct a multidimensional feature vector for risk assessment. The dimensions of the multidimensional feature vector for risk assessment include at least the hazard index of the leaked substance, the estimated leakage intensity based on the output of the cross-validation module, the ventilation level of the laboratory space, and the number of trapped personnel in real time obtained through the behavior recognition and intervention module. The risk assessment multidimensional feature vector is input into a pre-trained risk level assessment model, which is generated based on historical accident data and simulated data, and outputs a quantified comprehensive risk level. The strategy engine triggers a warning information multicast protocol that matches the overall risk level. The warning information multicast protocol defines the target notification department, personnel priority set, and information content template.

5. The system according to claim 1, characterized in that, The execution process of the mobile interactive assistant in the emergency mode includes: Load and render a 3D mesh model based on the aforementioned laboratory digital twin model library; By integrating data from UWB positioning tags worn by personnel with data from inertial measurement units, real-time positioning and attitude mapping of personnel are achieved in the three-dimensional mesh model. Based on the building topology in the digital twin model library, and with the dangerous areas detected by the multi-dimensional perception module being introduced in real time as dynamic obstacles, a path finding algorithm is used to calculate the optimal evacuation route in real time. The cost function of the path finding algorithm integrates the path length, the estimated travel time, and the real-time perceived environmental risk value. The optimal evacuation route is overlaid onto the real-time image of the terminal camera in an augmented reality manner to form a visual navigation guide.

6. The system according to claim 1, characterized in that, In response to a widespread leak alert, the execution process of the policy engine includes: The hazardous chemicals storage data and experimental equipment data in the laboratory digital twin model library are called up, and based on the predefined experimental equipment asset value weight and hazardous chemical hazard weight, an equipment priority sequence and a hazardous chemical priority sequence are constructed. Based on the multidimensional data simulation and prediction of the laboratory site, leakage diffusion simulation data is obtained. Combined with the equipment priority sequence and the hazardous chemical priority sequence, a dynamic emergency response strategy including equipment transfer sequence and hazardous material isolation scheme is generated. The dynamic emergency response strategy is converted and decomposed into task instructions, which are then pushed to the mobile interactive assistants of the corresponding personnel. The mobile interactive assistants provide the optimal action path to the target equipment location or the target hazardous chemical storage location.

7. The system according to claim 1 or 6, characterized in that, The emergency response strategy module also includes a protective equipment planning unit; The protective equipment planning unit is used to execute: When generating the task instruction, the material task association knowledge graph is accessed, which defines the standard protective equipment and handling tools required to handle different police situations. Based on the real-time location of personnel, the nearest target material point that has the required standard protective equipment and disposal tools is dynamically calculated from the emergency material points in the digital twin model library using the nearest neighbor search algorithm. The target material point is inserted as a preceding task node into the personnel's original task sequence to obtain a composite task chain, and a globally optimal action route is generated for the composite task chain.

8. The system according to claim 1 or 6, characterized in that, The emergency response strategy module assigns the task instructions to the execution processes of each of the mobile terminal interactive assistants, including: Determine the task type, required skills, and task location corresponding to the task instruction, and match the optimal executor from among the online personnel. The matching algorithm takes into account the personnel's skill level, the distance between the real-time location and the task point, and the current task load. The task instructions are pushed to the mobile interactive assistant of the optimal executor; The system receives the task execution status of the task instruction returned by the mobile interactive assistant of the optimal executor, and re-triggers the optimal executor matching process when the task execution times out or fails.

9. The system according to claim 2, characterized in that, The ledger registration module includes a registration data acquisition unit and an electronic ledger generation unit; The registration data collection unit is used to obtain the final usage value from the cross-validation module, automatically obtain the operator's identity and operation time period from the behavior recognition and intervention module, and obtain the corresponding experimental project information from the enterprise database. The electronic ledger generation unit is used to structurally integrate the obtained final usage value, the operator's identity, the operation time period, and the experimental project information to generate electronic ledger records and register them in the hazardous chemicals electronic ledger.

10. The system according to claim 9, characterized in that, The ledger registration module also includes an access control unit; The permission linkage control unit is used to execute: The system receives in real time the abnormal status diagnosis signal sent by the cross-validation module and the abnormal usage event sent by the usage compliance verification unit. When any of the aforementioned abnormal status diagnostic signals or the aforementioned abnormal dosage event is received, the corresponding operator's identity is determined by querying the currently associated electronic ledger record; Based on the operator's identity, an access freeze command is generated and executed, and an access freeze event log containing the abnormal status diagnostic signal or the usage abnormal event is generated.

11. The system according to claim 1, characterized in that, The mobile interactive assistant has a built-in multimodal interaction engine, which includes at least a voice interaction unit and a video stream scheduling unit. The voice interaction unit is used to parse user voice commands and recognize user feedback data on site, and at the same time broadcast the task commands and navigation prompts to the user. The video stream scheduling unit is used to invoke and present the real-time video stream of a specific camera in the multi-dimensional perception module according to the user's voice command or the task command.

12. The system according to claim 11, characterized in that, The emergency response strategy module is also used to execute: Receive the user's on-site feedback data uploaded by the mobile interactive assistant; When the user's on-site feedback data is inconsistent with the assumptions of the current dynamic emergency response strategy, the strategy adjustment mechanism is triggered; The strategy adjustment mechanism, based on the user's on-site feedback data, uses the strategy engine to make real-time corrections to the dynamic emergency response strategy and reassign updated task instructions.

13. The system according to claim 1, characterized in that, The behavior recognition and intervention module includes: The video analysis unit is used to decode and perform behavior sequence analysis on the real-time video stream acquired by the multi-dimensional perception module in order to identify the operational behavior of the experimenters. The violation pattern library stores the characteristics of violation operation patterns defined based on historical violation cases; An intervention triggering unit is used to generate and execute an intervention instruction when the degree of matching between the identified behavioral sequence and any of the features of the violation operation mode exceeds a preset threshold.

Images