Emergency training drilling method and system based on three-dimensional simulation

Abstract

The invention relates to an emergency training drilling method and system based on three-dimensional simulation. The method comprises the following steps: based on a physical environment state parameter and a user operation instruction set of a previous period, analyzing the occurrence condition of an emergency of the current period, and obtaining emergency identification information of the current period; obtaining a three-dimensional drilling scene of the current period based on the physical environment state parameters of the previous period, the user operation instruction set, preset current environment parameters and emergency identification information of the current period; acquiring a user operation instruction set and a current physiological signal when the three-dimensional drilling scene of the current period is executed; based on the emergency identification information of the current period, the user operation instruction set, the physical environment state parameters and the current physiological signals, the emergency operation condition of the user is evaluated, and an operation matching result and a cognitive load evaluation result of the current period are obtained. By adopting the method, actual combat pertinence and training effectiveness of drilling can be improved.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional simulation, and in particular relates to an emergency training and drill method and system based on three-dimensional simulation. Background Technology

[0002] With the development of 3D simulation and virtual reality technologies, their high immersion, repeatability, and high safety have enabled the widespread application of 3D simulation-based emergency training and drills, gradually forming a traditional drill mode driven by pre-set scripts.

[0003] In traditional methods, before an exercise begins, technical support personnel pre-compile an executable digital script based on a paper-based emergency plan, outlining fixed disaster evolution steps, standard response procedures, and a limited number of random branches. During the exercise, the 3D scenario is deduced and presented according to this script, following a timeline or event tree logic. The participants' actions are captured by the system and compared with the preset "standard answers" in the script. The main evaluation focuses on the correctness and completeness of the timing of their actions, ultimately generating a score or conclusion based on the correctness of their actions.

[0004] However, the current pre-script-driven approach inherently presents a deterministic or limitedly random disaster evolution path. Participants are easily led to test-taking mentality, failing to simulate the nonlinear evolution and black swan risks arising from complex system coupling, human error, and random environmental disturbances in real emergencies. Furthermore, its evaluation system focuses solely on judging the right or wrong of external operational actions, failing to identify and quantify the commanders' cognitive load under pressure, their decision-making logic, and their team collaboration mental models—deeper shortcomings. This results in insufficient relevance to real-world scenarios and ineffective training, hindering the genuine improvement of emergency decision-making and adaptability in extremely complex situations. Summary of the Invention

[0005] Therefore, it is necessary to provide an emergency training exercise method and system based on 3D simulation that can improve the practical relevance and training effectiveness of the exercises, addressing the aforementioned technical issues.

[0006] Firstly, this application provides an emergency training and drill method based on three-dimensional simulation, including:

[0007] Obtain the physical environment status parameters and user operation command set of the previous cycle;

[0008] Based on the physical environment status parameters and user operation instruction set of the previous cycle, analyze the occurrence of emergencies in the current cycle and obtain the emergency identification information of the current cycle.

[0009] Based on the physical environment state parameters of the previous cycle, the user operation instruction set, the preset current environment parameters, and the emergency event identification information of the current cycle, the physical environment state parameters of the current cycle are obtained, and the three-dimensional exercise scenario of the current cycle is generated based on the physical environment state parameters of the current cycle.

[0010] When executing the current cycle of the 3D simulation scenario, acquire the user operation instruction set and current physiological signals for the current cycle;

[0011] Based on the emergency event identification information, user operation instruction set, physical environment status parameters, and current physiological signals of the current period, the user's emergency operation situation is evaluated to obtain the operation matching result and cognitive load assessment result of the current period. The operation matching result is used to characterize the degree of conformity between the user's emergency operation and emergency logic, and the cognitive load assessment result is used to characterize the user's psychological load level.

[0012] Furthermore, based on the physical environment state parameters and user operation command set of the previous cycle, the occurrence of emergencies in the current cycle is analyzed to obtain the emergency identification information for the current cycle, including:

[0013] The physical environment state parameters and user operation instruction set of the previous cycle are concatenated to obtain the enhanced scenario state vector of the previous cycle; and the enhanced scenario state vector is input into the preset disaster strategy network to obtain the emergency event identification information of the current cycle.

[0014] The disaster response strategy network includes:

[0015] The input layer receives the enhanced scene state vector, performs standardization processing on the enhanced scene state vector to obtain a standardized state vector, and then transmits the standardized state vector to the feature encoding layer.

[0016] The feature encoding layer is used to perform at least one fully connected transformation on the normalized state vector, extract the high-order features of the normalized state vector, obtain the scene state encoding vector, and transmit the scene state encoding vector to the policy output layer.

[0017] The policy analysis layer is used to perform a fully connected transformation on the scene state encoding vector to obtain a logical value vector, and then transmit the logical value vector to the output layer.

[0018] The output layer is used to map the logic value vector through a preset non-linear activation function to obtain the probability distribution vector of the sudden event, and output the sudden event identification information of the current period based on the probability distribution vector of the sudden event.

[0019] Furthermore, the disaster response strategy network was obtained through the following methods:

[0020] Acquire preset historical exercise data, which includes historical enhanced scenario state vectors, historical emergency event identification information, and historical operation difficulty scores corresponding to historical emergency event identification information; and construct an initial policy network based on preset policy network initialization parameters.

[0021] Based on the initial policy network, the state vectors of historical augmented scenarios are processed to obtain the predicted emergency event identification information corresponding to the state vectors of historical augmented scenarios; and the predicted emergency event identification information is matched with the historical emergency event identification information in the historical exercise records to determine the historical operation difficulty score corresponding to the predicted emergency event identification information.

[0022] Based on a preset reward function, predicted emergency event identifiers, and the corresponding historical operational difficulty scores, the network prediction reward value is calculated; the expression for the reward function is:

[0023]

[0024] in, It is the network's predicted reward value. It is the first weighting coefficient. It is the second weighting coefficient. It is the third weighting coefficient. It is a predictive information for emergencies. It is an indicator function. It is a warning information for predicting emergencies. The historical operational difficulty score, It is the first evaluation threshold. It is the second evaluation threshold. It is a diverse reward function;

[0025] Based on the current reward value, the policy loss of the initial policy network is calculated, and the initial policy network is iteratively trained based on the policy loss until the initial policy network meets the preset iteration termination condition, thus obtaining the disaster policy network.

[0026] Furthermore, based on the physical environment state parameters of the previous cycle, the user operation instruction set, the preset current environment parameters, and the emergency event identification information of the current cycle, the physical environment state parameters of the current cycle are obtained, including:

[0027] Based on the physical environment state parameters of the previous cycle, the user operation instruction set, and the preset current environment parameters, the basic physical environment state parameters of the current cycle are obtained by calculation through the preset hybrid physical simulation model.

[0028] Based on the current period's emergency event identification information and the preset emergency time environment parameter adjustment rule set, determine the correction instructions for the basic physical environment state parameters;

[0029] The basic physical environment state parameters for the current period are corrected based on the correction instructions to obtain the physical environment state parameters for the current period.

[0030] Furthermore, based on the current period's emergency event identification information, user operation command set, physical environment state parameters, and current physiological signals, the user's emergency operation status is assessed to obtain the current period's operation matching results and cognitive load assessment results, including:

[0031] Based on a preset set of emergency response operation rules, the emergency identification information of the current period is mapped to obtain the standard emergency response operation set corresponding to the emergency identification information.

[0032] Quantify the differences between the standard emergency response operation set and the user operation instruction set in the current cycle to obtain the operation matching results;

[0033] Feature extraction is performed on the current physiological signals to obtain physiological feature data;

[0034] Physiological characteristic data are input into a preset cognitive load prediction model to obtain initial cognitive load values;

[0035] Based on the preset load calibration coefficient, the initial cognitive load value is numerically corrected to obtain the cognitive load assessment result.

[0036] Furthermore, the method also includes:

[0037] Obtain the operation matching results, user operation command set, and physical environment status parameters for each cycle;

[0038] The average value of the operation matching results for each cycle is calculated to obtain the traditional operation score;

[0039] Based on a preset set of key operation physical state mapping rules, the user operation instruction set and physical environment state parameters of each cycle are analyzed to construct a decision path topology graph.

[0040] By comparing the differences between the decision path topology graph and the pre-defined expert decision path knowledge graph, decision bias information is obtained.

[0041] Based on traditional operational scores and decision-making bias information, an evaluation report of the entire exercise process is generated.

[0042] Secondly, this application also provides an emergency training and drill system based on three-dimensional simulation, including:

[0043] The data acquisition module is used to acquire the physical environment status parameters and user operation command set of the previous cycle;

[0044] The emergency event determination module is used to analyze the occurrence of emergencies in the current period based on the physical environment status parameters and user operation instruction set of the previous period, and obtain the emergency event identification information of the current period.

[0045] The 3D scene construction module is used to obtain the physical environment state parameters of the current period based on the physical environment state parameters of the previous period, the user operation instruction set, the preset current environment parameters, and the emergency event identification information of the current period, and to generate the 3D exercise scene of the current period based on the physical environment state parameters of the current period.

[0046] The execution data module is used to acquire the user operation instruction set and current physiological signals for the current cycle when executing the 3D training scenario of the current cycle;

[0047] The data analysis module is used to assess the user's emergency response based on the current period's emergency event identification information, user operation command set, physical environment status parameters, and current physiological signals, and to obtain the current period's operation matching result and cognitive load assessment result. The operation matching result is used to characterize the degree of conformity between the user's emergency operation and emergency logic, and the cognitive load assessment result is used to characterize the user's psychological load level.

[0048] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the emergency training and drill methods based on three-dimensional simulation described in the first aspect of this application.

[0049] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the emergency training and drill methods based on three-dimensional simulation described in the first aspect of this application.

[0050] The aforementioned emergency training and drill method and system based on 3D simulation acquires the physical environment state parameters and user operation command set of the previous cycle; based on the physical environment state parameters and user operation command set of the previous cycle, it analyzes the occurrence of emergencies in the current cycle to obtain the emergency event identification information of the current cycle; based on the physical environment state parameters, user operation command set, preset current environment parameters, and emergency event identification information of the current cycle, it obtains the physical environment state parameters of the current cycle, and generates a 3D drill scenario for the current cycle based on the physical environment state parameters of the current cycle; when executing the 3D drill scenario of the current cycle, it acquires the user operation command set and current physiological signals of the current cycle; based on the emergency event identification information, user operation command set, physical environment state parameters, and current physiological signals of the current cycle, it evaluates the user's emergency operation situation to obtain the operation matching result and cognitive load assessment result of the current cycle; the operation matching result is used to characterize the degree of conformity between the user's emergency operation and emergency logic, and the cognitive load assessment result is used to characterize the user's psychological load level. This system upgrades traditional static script-driven drills into a dynamic closed-loop system. By introducing an intelligent event generation mechanism based on reinforcement learning, it breaks the deterministic limitations of pre-set scripts, enhancing the unpredictability and practical challenge of drills. Simultaneously, by integrating multimodal physiological and operational data, the assessment dimensions are extended from superficial operational correctness to deep cognitive load states, achieving a leap from behavioral assessment to cognitive shaping, thereby improving the relevance and effectiveness of emergency training. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A flowchart illustrating an emergency training and drill method based on three-dimensional simulation, provided as an embodiment of this application;

[0053] Figure 2 This is a schematic diagram of the structure of an emergency training and drill system based on three-dimensional simulation, provided as an embodiment of this application. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0056] In one embodiment, such as Figure 1 As shown, an emergency training and drill method based on 3D simulation is provided. This embodiment illustrates the method by applying it to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps S101-S105, wherein:

[0057] S101, obtain the physical environment status parameters and user operation instruction set of the previous cycle.

[0058] Specifically, the terminal divides a complete 3D virtual emergency training exercise into multiple consecutive and equally long analysis cycles. Let the total duration of the entire emergency training exercise be... The duration of each analysis cycle is Then the total number of periods Use integer subscripts. ( These cycles are identified by ) This is the initial cycle. The current analysis cycle refers to the cycle that the terminal is currently processing, with the sequence number [number missing]. The cycle. The previous analysis cycle is the one immediately preceding the current cycle, with the sequence number [missing information]. The period ( The terminal acquires the physical environment state parameters and user operation command set from the previous cycle. The physical environment state parameters are a multi-dimensional data set describing the overall physical state of the 3D virtual simulation scenario at the end of the previous analysis cycle. The mathematical form of the physical environment state parameters from the previous cycle can be represented as a high-dimensional vector. ,in This refers to the number of physical parameters included in the physical state characteristics. Specific physical parameters include, but are not limited to: temperature, pressure, combustible gas concentration, smoke concentration, and equipment on / off status (e.g., valve opening, pump start / stop) at key locations within the exercise scenario (e.g., storage tanks, pipelines, equipment). Physical environment state parameters can be obtained through simulation calculations of the previous cycle's 3D virtual exercise scenario. Upon initial execution, the physical environment state parameters are initialized according to the preset exercise plan. The user operation instruction set is the set of all control commands issued by the user in the previous cycle through human-computer interaction devices (e.g., VR controllers, touchscreens, keyboards). The mathematical form of the user operation instruction set in the previous cycle can be expressed as... ,in It is the number of user roles, where each element User In the cycle The operation instructions can be in the form of ,in It is the user's unique identifier. It is an action type identifier used to represent the type of action (such as "close the valve" or "start the water pump"). The target object identifier of the operation. These are action-related parameters (such as closing torque and water flow rate).

[0059] S102, based on the physical environment state parameters and user operation instruction set of the previous cycle, analyze the occurrence of emergencies in the current cycle and obtain the emergency identification information of the current cycle.

[0060] Specifically, the terminal uses the physical environment state parameters from the previous cycle. and user operation instruction set The system analyzes the data to determine whether and what type of emergency should be triggered within the current period, and outputs the emergency event identification information for the current period. Emergency event identification information is a string or code used to uniquely identify a specific black swan event or unexpected situation. The emergency event identification information for the current period... The specific form can be "secondary_leak_at_valve_B01", "communication_node_C_failure", or "wind_sudden_increase". The emergency event identifier information indicates the type of unexpected event that occurred within the current period, exceeding the preset exercise script, or was dynamically generated. The value range of the emergency event identifier information can be set according to the types of emergencies that may occur in the actual exercise, and the total number of emergency event types can be recorded as follows: .

[0061] S103: Based on the physical environment state parameters of the previous cycle, the user operation instruction set, the preset current environment parameters, and the emergency event identification information of the current cycle, the physical environment state parameters of the current cycle are obtained, and the three-dimensional exercise scenario of the current cycle is generated based on the physical environment state parameters of the current cycle.

[0062] Specifically, the preset current environmental parameters refer to the external environmental conditions that take effect in each cycle according to the preset exercise script. Their mathematical form can be expressed as: ,in It is the total number of periods, for any element. It is the first External environmental condition parameters for each cycle, It is the first Three-dimensional wind direction and speed vectors for each period. It is the first The ambient temperature for each cycle, It is the first Humidity in each cycle, It is the first The atmospheric pressure for each cycle, and the preset current environmental parameters can be set according to the preset exercise script and cycle division in actual work. The terminal is based on the physical environment state parameters of the previous cycle. User operation instruction set and preset current environment parameters Combined with the current period's emergency identification information Simulate the physical environment of the current cycle to obtain the physical environment state parameters of the current cycle. It employs a 3D graphics rendering engine (such as Unity3D or Unreal Engine) and adjusts the rendering based on physical environment parameters. Drive the digital twin scene update to obtain the current period's 3D simulation scene, which is then output to the user's display device.

[0063] S104: When executing the three-dimensional training scenario of the current cycle, acquire the user operation instruction set and current physiological signals of the current cycle.

[0064] Specifically, when executing the 3D simulation scenario of the current cycle, the terminal synchronously collects the user operation instruction set of the current cycle. and current physiological signals The user operation instruction set for the current cycle. The mathematical form of the instruction set of the previous cycle Similarities refer to the set of new control commands issued by the user after viewing the 3D simulation scene, which can be captured and encoded in real time by the human-computer interaction interface. Current physiological signals It is data on the physiological responses of users when dealing with the current 3D scene, and its mathematical form can be represented as a multivariate time series. ,in It is the first Heart rate data for each cycle, It is the first Skin conductance data over a period of time It is the first Eye-tracking data for each cycle (including 2D gaze coordinates and pupil diameter). All data are timestamped and aligned with the frame rendering time of the 3D training scene and the timestamps of the user's operation command set. The central velocity data and skin conductance data can be acquired through wristband or chest strap sensors, while the eye-tracking data can be acquired through an eye-tracking module integrated into the VR headset.

[0065] S105, based on the emergency event identification information, user operation instruction set, physical environment state parameters and current physiological signals of the current period, assess the user's emergency operation situation and obtain the operation matching result and cognitive load assessment result of the current period; the operation matching result is used to characterize the degree of conformity between the user's emergency operation and emergency logic, and the cognitive load assessment result is used to characterize the user's psychological load level.

[0066] Specifically, the terminal is based on the identified emergency event identifiers within the current period. User-generated real-time operation instruction sets Current physical environment status parameters and the collected current physiological signals The system performs multi-dimensional real-time evaluations of users' emergency response performance to obtain the operation matching results for the current period. and cognitive load assessment results Among them, the operation matching result Information used to characterize emergencies in the current period. and current physical environment state parameters The user's actions are consistent with standard logical actions under the same circumstances. (Action matching result) It can be a scalar score between 0 and 1, with a higher score indicating a higher degree of compliance. (Cognitive load assessment results) Used for identifying sudden events in the current cycle. and current physical environment state parameters In this situation, the level of psychological load experienced by the user. Cognitive load assessment results. It can be a continuous value, and the larger the value, the higher the level of psychological conformity.

[0067] This embodiment provides an emergency training and drill method based on 3D simulation. It acquires the physical environment state parameters and user operation command set from the previous cycle, analyzes and intelligently generates the emergency event identifier for the current cycle. Then, based on the physical environment state parameters, user operation command set, preset current environment parameters, and emergency event identifier information from the previous cycle, the simulation model deduces the physical environment state parameters for the current cycle that conform to physical laws and intelligent intervention results, and drives the generation of a 3D drill scenario. During the execution of the 3D drill scenario, the latest user operation command set and current physiological signals are simultaneously collected. Finally, by integrating the emergency event identifier information, user operation command set, physical environment state parameters, and current physiological signals from the current cycle, a quantitative matching assessment of the logical consistency of the user's emergency operations in the current cycle is achieved, as well as a real-time cognitive assessment of the user's psychological load level, resulting in the operation matching result and cognitive load assessment result for the current cycle. By introducing a reinforcement learning-based intelligent event generation mechanism, the deterministic limitations of pre-set scripts are broken, enhancing the unpredictability and practical challenge of drills. At the same time, by integrating multimodal physiological and operational data, the assessment dimensions are extended from superficial operational correctness to deep cognitive load status, thereby improving the pertinence and effectiveness of emergency training.

[0068] In one embodiment, based on the physical environment state parameters and user operation instruction set of the previous cycle, the occurrence of emergencies in the current cycle is analyzed to obtain the emergency identification information of the current cycle, including:

[0069] S201, the physical environment state parameters and user operation instruction set of the previous cycle are concatenated to obtain the enhanced scenario state vector of the previous cycle; and the enhanced scenario state vector is input into the preset disaster strategy network to obtain the emergency event identification information of the current cycle.

[0070] Specifically, the terminal will use the physical environment status parameters from the previous cycle. and user operation instruction set Data is stitched together to construct an enhanced scene state vector. Specifically, we can start with multi-dimensional physical environment state parameters. Convert to a one-dimensional vector through the flatten operation. At the same time, a structured set of user operation instructions will be implemented. Encoding is performed using a pre-trained embedding layer to encode the components of each instruction (such as...). , The vectors of all instructions from the previous cycle are converted into dense vectors of fixed dimensions. Then, the vectors of all instructions from the previous cycle are merged using an aggregation function (such as summation, averaging, or weighted summation based on attention weights) to form the operation behavior vector. Finally, the physical state vector With operation behavior vector Concatenate the vectors to obtain the enhanced scene state vector. The terminal inputs the enhanced scenario state vector from the previous cycle into the preset disaster response strategy network to obtain the emergency event identification information for the current cycle. The pre-trained disaster response strategy network is a deep neural network. Its function is to receive the enhanced state vector from the previous cycle as input, perform a series of internal nonlinear transformations and feature processing, and finally output an identifier representing the type of emergency event that should be triggered in the current cycle, i.e., the emergency event identification information for the current cycle. .

[0071] The disaster response strategy network includes:

[0072] S2011, Input Layer: The input layer receives the enhanced scene state vector, performs standardization processing on the enhanced scene state vector to obtain a standardized state vector, and transmits the standardized state vector to the feature encoding layer.

[0073] Specifically, the input layer of the terminal disaster response strategy network is its first layer, which is used to receive and preprocess the enhanced scenario state vector from external input. The input layer first undergoes standardization to eliminate the influence of different physical units and numerical ranges, ensuring the stability and convergence speed of network training. Specifically, the Z-score standardization method can be used, and its calculation formula is... ,in and These are pre-defined mean and standard deviation vectors. They can be calculated during network training from the augmented scene state vectors of the entire training dataset, representing the mean and standard deviation of each feature dimension, respectively. The resulting standardized state vectors... Each feature dimension is transformed into a distribution with a mean of 0 and a standard deviation of 1. Subsequently, the input layer standardizes this state vector. It is transmitted to the next layer of the network, namely the feature coding layer.

[0074] S2012, Feature Encoding Layer: The feature encoding layer is used to perform at least one fully connected transformation on the normalized state vector, extract the high-order features of the normalized state vector, obtain the scene state encoding vector, and transmit the scene state encoding vector to the policy output layer.

[0075] Specifically, the feature encoding layer of the disaster response strategy network consists of at least one fully connected layer (also known as a dense layer). Its core function is to transform the standardized state vector through a series of nonlinear transformations. This allows for the extraction of higher-order, more abstract semantic features, leading to a deeper understanding of the current complex exercise situation. The operations performed by each fully connected layer can be represented as... ,in This represents the index of the current fully connected layer (for the first layer, ), This is the weight matrix of this layer. It is the bias vector of this layer. It is a pre-defined non-linear activation function (such as ReLU, Tanh). Weight matrix and bias vector The dimension of the output vector of this layer is determined by the dimension of the output vector. These parameters are learned through optimization algorithms during the network training phase. Through at least one such transformation, the original, high-dimensional normalized state vector is mapped to a scene state encoding vector containing key situational information (such as the fire development trend, the effectiveness of user intervention, and system weaknesses). Scene state encoding vector It is then transmitted to the next layer, namely the strategy analysis layer.

[0076] S2013, the policy analysis layer, is used to perform a fully connected transformation on the scene state encoding vector to obtain a logical value vector, and then transmit the logical value vector to the output layer.

[0077] Specifically, the policy analysis layer of the disaster response strategy network receives the scene state encoding vector from the feature encoding layer. The policy analysis layer is a fully connected layer whose function is to encode the scene state vector. This is mapped to a logical value space corresponding to all possible actions related to unforeseen events. The specific operational formula is as follows: ,in It is the weight matrix of the strategy analysis layer. It is the bias vector of the strategy analysis layer. This represents the total number of preset emergency event action types. (Calculation result) That is, a logical value vector, where each element in the vector... Corresponding to the This is a logical value representing the type of emergency, where a higher value reflects the degree to which the disaster response strategy network is inclined to trigger that type of emergency under the current condition. The strategy analysis layer vectorizes this logical value. Transmitted to the final output layer.

[0078] S2014, the output layer, is used to map the logic value vector through a preset non-linear activation function to obtain the probability distribution vector of sudden events, and output the sudden event identification information of the current period based on the probability distribution vector of sudden events.

[0079] Specifically, the output layer of the disaster response strategy network receives the logical value vector from the strategy analysis layer. And through a preset non-linear activation function—the Softmax function— Mapping is performed. The formula for calculating the Softmax function is: for a logical value vector... Each element in Calculate the corresponding probability value ,in This is the total number of preset emergency event action types. It is a natural constant. This calculation yields the probability distribution vector of the sudden event. The vector satisfies and Each This represents the triggering of the first time in the current cycle. The predicted probability of sudden events. Finally, the output layer is based on this probability distribution vector. Output the event identifier information for the current period. Specifically, the output layer directly selects the event type with the highest probability. The final output They are all specific identifiers from a predefined set of events, such as the string "pump_A_failure".

[0080] This embodiment provides an emergency training and drill method based on 3D simulation. By deeply integrating the physical environment state parameters of the previous cycle with the user operation command set, an enhanced scenario state vector that comprehensively reflects historical states and human intervention is constructed. This enhanced scenario state vector is then input into a preset disaster strategy network to obtain the emergency event identification information for the current cycle. The generation of black swan events is upgraded from traditional scripts or random triggers to intelligent decision-making based on deep learning and tightly coupled with historical states and user behavior. This enables the dynamic and adaptive generation of the most challenging emergency events for training based on the real-time progress of the drill (physical state) and the specific operations of the users. This breaks through the deterministic and linear limitations of preset scripts in traditional drills, enhancing the unpredictability, adversarial nature, and realism of the drills. It provides crucial technical support for cultivating emergency personnel's ability to make on-the-spot decisions and adapt to systems in complex, dynamic, and high-pressure environments.

[0081] In one embodiment, the disaster response strategy network is obtained through the following method:

[0082] S301, acquire preset historical exercise data, which includes historical enhanced scenario state vectors, historical emergency event identification information, and historical operation difficulty scores corresponding to historical emergency event identification information; and construct an initial policy network based on preset policy network initialization parameters.

[0083] Specifically, the terminal acquires preset historical exercise data. This preset historical exercise data includes a large number of historical exercise records, which can be expressed mathematically as follows: ,in It is the total number of historical drill records. It is any historical exercise record. It is a record of historical drills. The historical enhanced scene state vector is constructed in the same way as the enhanced scene state vector construction method described in step S201. It is obtained by splicing and standardizing the physical environment state parameters and user operation instruction set of the historical exercise. It is a record of historical drills. The historical emergency identification information has the same value range as the emergency identification information. The records document the types of emergencies that actually occurred or that experts deemed should be triggered in the historical exercise log. It is a record of historical drills. Historical emergency identification information The corresponding historical operation difficulty score. The historical operation difficulty score is a preset scalar value used to quantify the challenge posed to users by this type of emergency. It can be derived from a post-event comprehensive score by experts based on factors such as the complexity of handling, resource consumption, and time pressure caused by the historical emergency's identifier information, or it can be calculated statistically based on indicators such as the user's operational error rate and task completion time after the event in historical drill records. Its value ranges from [1, 10], with a higher value indicating greater difficulty. The higher the difficulty of handling sudden events, the more complex the representation becomes. After acquiring data, the terminal constructs an initial policy network based on preset policy network initialization parameters. These preset policy network initialization parameters include the initial values ​​of the weight matrices and bias vectors for each layer of the network (input layer, feature encoding layer, policy analysis layer, and output layer). These initial values ​​can be obtained using standard initialization methods in the field of deep learning, such as Xavier initialization or He initialization. The specific process is as follows: based on the number of neurons in the input and output of each layer, from a specific probability distribution (e.g., mean 0, variance 0.05%)... Random sampling is performed on the truncated normal distribution to generate the initial weight matrix; the terminal constructs a deep neural network model with a hierarchical structure (input layer - feature encoding layer - policy analysis layer - output layer) as described in S201, but with untrained parameters, as the initial policy network. ,in This represents the set of parameters used for its initialization.

[0084] S302, based on the initial policy network, the historical augmented scenario state vector is processed to obtain the predicted emergency event identification information corresponding to the historical augmented scenario state vector; and the predicted emergency event identification information is matched with the historical emergency event identification information in the historical exercise record to determine the historical operation difficulty score corresponding to the predicted emergency event identification information.

[0085] Specifically, the terminal uses the initial policy network constructed in step S301. For each historical exercise record in the preset historical exercise data Historical Enhanced Scene State Vector Forward propagation is performed to obtain the network's predicted output for the input state under the current parameters. The specific processing procedure is as follows: The input is fed into the input layer of the initial policy network, and sequentially undergoes normalization, feature encoding, policy analysis, and a Softmax transform in the output layer. The specific computation process is deterministic; for each input... The initial policy network will output a probability distribution vector. ,in This represents the preset total number of emergency event types. This indicates that the initial policy network predicts the state. Next trigger The probability of a sudden event, and satisfying the following conditions. . That is, the historical enhanced scene state vector The corresponding probability distribution of predicted emergencies. And based on the predicted probability distribution. A greedy sampling method is used to select the event type with the highest probability as the identifier information for predicting sudden events. The terminal will display the predicted emergency identification information. The system matches historical emergency event identifiers with the pre-set historical exercise data in each historical exercise record to obtain predicted emergency event identifiers. Corresponding historical operation difficulty score Specifically, when predicting emergency identification information... and historical exercise records Historical emergency identification information If they are the same, then the historical emergency identification information will be used. Corresponding historical operation difficulty score Identified as an indicator for predicting emergencies Corresponding historical operation difficulty score In other cases, information identifying predicted emergencies can be extracted from pre-set historical drill data. Historical emergency identification information Same historical exercise record And calculate the identification information of each historical emergency. Corresponding historical operation difficulty score The arithmetic mean is used as a marker for predicting sudden events. Corresponding historical operation difficulty score .

[0086] S303, based on the preset reward function, predicted emergency event identifier information, and the historical operation difficulty score corresponding to the predicted emergency event identifier information, the network prediction reward value is calculated; the expression of the reward function is:

[0087]

[0088] in, It is the network's predicted reward value. It is the first weighting coefficient. It is the second weighting coefficient. It is the third weighting coefficient. It is a predictive information for emergencies. It is an indicator function. It is a warning information for predicting emergencies. The historical operational difficulty score, It is the first evaluation threshold. It is the second evaluation threshold. It is a diversity reward function.

[0089] Specifically, for each historical exercise record input into the initial policy network The terminal calculates the reward function formula based on the current initial policy network's performance against historical training records. The reward value that should be obtained from the prediction This reward function is the core incentive mechanism driving the network to learn and evolve towards the desired policy. Specifically, the network predicts the reward value. It is a scalar. First weighting coefficient. Second weighting coefficient and the third weighting coefficient These are preset hyperparameters used to adjust the relative importance of the three components in the reward function. They can be set according to actual work requirements; the default setting is [insert setting here]. When the input historical exercise record is a historical exercise record. At the same time, predict emergency event identification information It is the predicted emergency identification information calculated by S302. Predicting emergency identification information Historical operation difficulty score This can be obtained from S302. Indicator function. Used for logical judgment; the function value is 1 when the logical condition within the parentheses is true, and 0 otherwise. First evaluation threshold. Second evaluation threshold It is a preset scalar value used to divide the difficulty score interval, and satisfies First evaluation threshold Second evaluation threshold The settings can be customized according to the actual work requirements; the default setting is... Diverse reward functions It is an information system for predicting emergencies. The input function encourages the network to explore different types of events, preventing the policy from becoming rigid too early. Specifically, it can take the form of a reward based on action selection counts. ,in Predict event identifiers within the current training round (or a sliding window). The total number of times the network selected the initial policy. It is a preset scaling factor. During calculation, the terminal first determines... Is it true? If true, then the value of the first term is... Otherwise, it is 0. Then proceed with the judgment. Is it true? If true, then the value of the second term is... Otherwise, it is 0. Then calculate. Finally, summing the three terms yields the result for the historical exercise record. Reward value This reward value means that if the network chooses a very low difficulty level (…), then… The event indicates that its choice lacks challenge, thus warranting a positive penalty (+). ), encouraging networks to avoid such choices; if a network chooses a very difficult ( The event demonstrates that its choice is extremely challenging, resulting in a negative reward (i.e., punishment). The game encourages online players to apply appropriate pressure when facing skilled opponents rather than simply increasing the difficulty; the variety of reward items encourages online players to choose event types that they have not frequently selected recently.

[0090] S304. Based on the current reward value, calculate the policy loss of the initial policy network, and iteratively train the initial policy network based on the policy loss until the initial policy network meets the preset iteration termination condition, thus obtaining the disaster policy network.

[0091] Specifically, the terminal calculates a batch of historical exercise records based on step S304. Reward value The Proximal Policy Optimization (PPO) algorithm is used to calculate the policy loss of the initial policy network. PPO is a reinforcement learning policy gradient method whose core idea is to constrain the difference between the new and old policies when updating network parameters, avoiding excessively large update steps that could lead to training instability. For each historical training record... The terminal uses the PPO algorithm to calculate the corresponding policy loss. The network parameters are then updated using stochastic gradient descent (SGD) or its variants, such as the Adam optimizer. ,in This is the learning rate. This process (parameter updates from S302 to S305) is repeated, i.e., iterative training, until the initial policy network meets the preset iteration termination condition, thus obtaining the disaster policy network. The preset iteration termination condition is used to determine whether training is complete. The predicted sudden event identifier information can be: 1) The total number of training iterations reaches a preset maximum value. ;2) Strategy loss The change amplitude is less than a preset threshold over multiple consecutive iteration cycles. This indicates that convergence has been achieved, and the preset iteration termination condition can be set according to the actual work.

[0092] This embodiment provides an emergency training and drill method based on 3D simulation. It acquires pre-set historical drill data and constructs an initial policy network using a standard initialization method. This initial network is then used to perform forward prediction of historical states, obtaining corresponding predicted emergency event identifiers. By matching the predicted emergency event identifiers with the pre-set historical drill data, the corresponding historical operational difficulty score is determined. Furthermore, a reward function that comprehensively considers the difficulty of the selected events and the diversity of policy exploration is used to calculate the network prediction reward value that the initial policy network should obtain. Finally, based on this network prediction reward value, the proximal policy optimization (PPO) algorithm is used to calculate the policy gradient loss, and the optimizer iteratively updates the network parameters, thereby obtaining a mature disaster response policy network. This transforms the construction of the disaster response policy network from a black-box process into an interpretable and reproducible data-driven process. By combining historical difficulty scores with a multi-objective reward function, the network is explicitly guided to learn and generate intelligent emergencies with both diversity and difficulty, ensuring the training is targeted and dynamically matched with the user's operational level.

[0093] In one embodiment, the physical environment state parameters for the current period are obtained based on the physical environment state parameters of the previous period, the user operation instruction set, the preset current environment parameters, and the emergency event identification information of the current period, including:

[0094] S401 calculates the basic physical environment state parameters for the current cycle based on the physical environment state parameters of the previous cycle, the user operation instruction set, and the preset current environment parameters through a preset hybrid physical simulation model.

[0095] Specifically, the terminal uses the physical environment state parameters from the previous cycle. User operation instruction set and preset current environment parameters External environmental condition parameters in the current cycle They are all input into a pre-defined hybrid physics simulation model. In this process, calculations are performed to deduce the physical environment state that should exist at the end of the current cycle without any unforeseen intervention, i.e., the basic physical environment state parameters for the current cycle. The pre-defined hybrid physics simulation model is a computational model that integrates physical mechanisms and data-driven technologies. Specifically, it can be constructed by training a neural network using historical state transition data generated from high-fidelity CFD simulations, and constraining the neural network output to conform to physical laws through a hybrid loss function. This model can quickly and faithfully predict the basic physical environment state of the next moment based on the input state, operation, and environmental parameters, achieving a balance between simulation accuracy and computational efficiency. It is the core computational engine supporting the dynamic evolution of the scene. During real-time operation, it utilizes the physical environment state parameters of the previous cycle, the user operation command set, and the external environmental condition parameters of the current cycle. The input is the model's output. This characterizes the physical state to which the scene should evolve, considering only user actions and the influence of the natural environment. The specific form can be a collection of physical fields (such as temperature field, concentration field grid) and device state variables.

[0096] S402, based on the current period's emergency event identification information and the preset emergency time environment parameter adjustment rule set, determines the correction instruction for the basic physical environment state parameters.

[0097] Specifically, the terminal uses the emergency event identification information already generated in the current cycle. Query a preset set of rules for adjusting environmental parameters during emergencies. This determines the necessary parameters of the basic physical environment. The specific modifications applied, i.e., the correction instructions. Preset set of rules for adjusting environmental parameters during emergencies. It is a structured knowledge base, built upon expert experience and analysis of historical accident cases. Each rule establishes a mapping from emergency identification information to a set of specific physical parameter modification operations, which can be expressed mathematically as follows: ,in It is an emergency event identifier (such as "pump_A_failure"). This is the corresponding correction command. Correction command This can be a structured list or script that explicitly specifies the object to be modified, its attributes, and the new values. For example, for the event identifier "toxic_gas_leak_at_valve_V101", the corresponding correction instructions... It may include: 1) in three-dimensional coordinates At (corresponding to valve V101), create a gas source; the gas type is "hydrogen sulfide"; the leakage rate is... The preset set of rules for adjusting environmental parameters during emergencies can be set according to the type of emergency identified by the emergency event identification information in actual work and the corresponding changes in physical parameters.

[0098] S403, based on the correction instruction, corrects the basic physical environment state parameters of the current cycle to obtain the physical environment state parameters of the current cycle.

[0099] Specifically, the terminal executes the correction instruction obtained in step S402. And its specific effects are applied to the basic physical environment state parameters of the current cycle. This generates the physical environment state parameters for the current cycle. Among them, regarding the correction instructions Each sub-instruction in the code corresponds to a basic physical environment state parameter. This refers to an operation that modifies a specific physical field or device state variable. For example, if a correction instruction... To "add a heat source of intensity I at location P", the correction operation is performed on the basic physical environment state parameters. In the temperature field grid, the current temperature value of the grid node corresponding to position P is... Updated to ,in It is the time step. It is the spatial influence weighting function. (Correction instruction) After execution, the basic physical environment state parameters It is then updated to the physical environment state parameters of the current period. .

[0100] This embodiment provides an emergency training and drill method based on 3D simulation. First, physical environment state parameters, user operation command sets, and preset current environment parameters are input into a preset hybrid physical simulation model to calculate the basic physical environment state parameters for the current period, considering only the influence of normal interactions. Then, based on the emergency event identification information for the current period, a preset emergency time environment parameter adjustment rule set is queried to determine the correction instructions needed to modify the basic state, thus obtaining the correction instructions for the basic physical environment state parameters for the current period. Finally, by executing these correction instructions, the final physical environment state parameters are obtained. This method organically combines deterministic simulation based on physical laws with the injection of non-deterministic events based on artificial intelligence, achieving dynamic and intelligent evolution of disaster situations. It ensures that the virtual environment has a realistic physical response to user operations, while breaking the rigid settings of event timing and type in traditional scripts, thus improving the realism, complexity, and challenge of the drill scenario.

[0101] In one embodiment, based on the current period's emergency event identification information, user operation command set, physical environment state parameters, and current physiological signals, the user's emergency operation status is assessed to obtain the current period's operation matching result and cognitive load assessment result, including:

[0102] S501, based on a preset set of emergency response operation rules, maps the emergency identification information of the current period to obtain the standard emergency response operation set corresponding to the emergency identification information.

[0103] Specifically, the pre-defined emergency response operation rule set is a structured knowledge base that can be derived from industry emergency plan standards, the handling experience of domain experts, or the summary of numerous historical successful cases. The pre-defined emergency response operation rule set establishes a system based on each emergency type identifier. The mapping relationship to standard operating procedures, i.e. .in, It is an ordered sequence of operations, which can be expressed mathematically as: Each operation It is a structured tuple that defines the actions to be performed under ideal emergency logic. The terminal will display the emergency event identification information for the current period. Mapping is performed to obtain the event identification information for the current period. Corresponding standard emergency response operation set .

[0104] S502 quantifies the differences between the standard emergency response operation set and the user operation instruction set for the current period to obtain the operation matching result.

[0105] Specifically, the terminal calculates the user operation instruction set for the current cycle. The standard emergency response operation set obtained in step S501 The degree of difference between them, and transform this degree of difference into a measurable operation matching result. The terminal can use the sequence edit distance method to measure the user's operation command set. With standard emergency response operation set The degree of difference between the sequences. The sequence edit distance method is a classic algorithm for quantifying the differences between two sequences. It measures the difference by calculating the minimum number of editing operations (including insertion, deletion, and replacement) required to transform one sequence into another. The terminal uses the sequence edit distance method, first configuring the standard emergency response operation set for the current period. and user operation instruction set The operation is abstracted into two sequences of operation identifiers. Each operation is encoded as a unique symbol based on its action type and target object. The algorithm calculates the minimum edit distance between these two operation sequences using dynamic programming. The larger the distance value, the greater the difference between the user's operation and the standard procedure. Finally, a pre-defined normalization function (such as...) is used to... Convert this distance into an operation matching result between 0 and 1. The higher the value, the higher the degree of compliance.

[0106] S503 extracts features from the current physiological signal to obtain physiological feature data.

[0107] Specifically, the terminal processes the collected current physiological signals. Real-time signal processing and feature engineering are performed to extract quantitative indicators that effectively reflect changes in users' psychological load, thereby obtaining physiological characteristic data. The feature extraction process is performed separately for each type of signal: For heart rate data, filtering is first applied, followed by calculation of its frequency domain features, specifically the ratio of low-frequency power to high-frequency power (LF / HF) using Fast Fourier Transform (FFT) to obtain heart rate variability (HRV). For electrodermal response (EDS) data, smoothing is first performed, followed by feature extraction, including at least the average or trend of skin conductance level (SCL). For eye movement (EMG) data, regions of interest are first divided based on fixation coordinates, followed by feature extraction, including at least the total number of fixations and the average fixation duration. The terminal combines all these statistical, spectral, and time-frequency features calculated from the current physiological signals into a multidimensional physiological feature dataset. ,in It is the total dimension of the features.

[0108] S504, input physiological characteristic data into the preset cognitive load prediction model to obtain the initial cognitive load value.

[0109] Specifically, the terminal inputs physiological characteristic data into a preset cognitive load prediction model to obtain an initial cognitive load value. The preset cognitive load prediction model... It is a supervised learning-trained machine learning model whose goal is to establish a mapping from physiological characteristics to cognitive load levels. Specifically, it can be constructed using Support Vector Regression (SVR), Random Forest Regression, or Deep Neural Network algorithms, based on datasets from actual psychological analysis work (including physiological characteristic data from historical analysis processes and corresponding ground truth cognitive load values). The resulting initial cognitive load values... It is a continuous scalar value that reflects the original prediction of the user's current psychological load level by the preset cognitive load prediction model.

[0110] S505, based on a preset load calibration coefficient, performs numerical correction processing on the initial cognitive load value to obtain the cognitive load assessment result.

[0111] Specifically, the preset load calibration coefficient is used to correct the deviation of the preset cognitive load prediction model. It can be obtained by statistically analyzing the average deviation between the predicted values ​​output during the construction of the preset cognitive load prediction model and the true cognitive load values, and can be denoted as... The terminal calibrates according to the preset load calibration coefficient. Using formula Obtain the cognitive load assessment results for the current cycle.

[0112] This embodiment provides an emergency training and drill method based on 3D simulation. It determines a standard emergency response operation set through a pre-set set of emergency response rules. Then, using a refined sequence comparison and multi-feature similarity quantification algorithm, it calculates the differences between the standard emergency response operation set and the user's operation instruction set, thereby obtaining the operation matching result. Simultaneously, it performs in-depth time-domain, frequency-domain, and spatial feature extraction on synchronously acquired multimodal physiological signals to obtain physiological feature data. This physiological feature data is input into a pre-set cognitive load prediction model to obtain an initial cognitive load value for the user's psychological load. Finally, using a load calibration coefficient pre-calibrated based on an independent calibration dataset, it systematically corrects the deviation of this initial cognitive load value to obtain the cognitive load assessment result. This achieves a comprehensive quantitative evaluation of the user's emergency performance. It breaks through the limitations of traditional drills that only focus on the superficial assessment of correct or incorrect operational steps. By introducing a data-driven cognitive load calculation model, it can provide real-time insight into the commander's psychological resource consumption and attention allocation under high-pressure and complex environments, extending the assessment depth to the cognitive level and improving the accuracy, personalization, and final effectiveness of training.

[0113] In one embodiment, the method further includes:

[0114] S601, obtains the operation matching results, user operation instruction set and physical environment status parameters for each cycle.

[0115] Specifically, the terminal acquires the operation matching results, user operation command sets, and physical environment status parameters for all cycles throughout the entire 3D virtual emergency training exercise.

[0116] S602 calculates the average value of the operation matching results for each cycle to obtain the traditional operation score.

[0117] Specifically, the terminal matches the operation results for each cycle. The average value is calculated to obtain the traditional operation score. The specific calculation formula is as follows: . It is the total number of cycles.

[0118] S603, based on a preset set of key operation physical state mapping rules, analyzes the user operation instruction set and physical environment state parameters for each cycle and constructs a decision path topology map.

[0119] Specifically, the preset key operation physical state mapping rule set is a structured rule base, which can be denoted as... The preset set of key operation physical state mapping rules is used to characterize the mapping relationship between physical environment state parameters and various operation commands. Each rule defines: when the physical environment state parameters meet the physical environment state parameter conditions (For example, when "the temperature of storage tank T001 is >150°C and the pressure is <safety threshold"), the corresponding operating instructions are... What is it (e.g., "Start the T001 tank top spray system"). The terminal iterates through each cycle. Data pairs ,according to and Identify the operation instructions to be executed in the current cycle. and in The system searches for instructions that match these critical operations. If a match is found, the "state-operation" pair is defined as a decision node. ,in For period The time identifier is used to add this node to the node set of the graph. After traversing all cycles and extracting all decision nodes, the terminal determines the nodes according to their temporal order (by...). Directed edges are established based on the decision and logical causal relationship (derived from the state transition logic in the rule set). For example, if a node... Operations in The node satisfies Physical environment state parameters Then in and Establish a directed edge between them , indicating instructions This led to subsequent instructions. The needs of all nodes. Ultimately, all nodes... and edge This forms a directed graph, namely a decision path topology graph. .

[0120] S604. Compare the differences between the decision path topology graph and the preset expert decision path knowledge graph to obtain decision bias information.

[0121] Specifically, the terminal will use the decision path topology map constructed by S603. With a pre-defined expert decision-making path knowledge graph By performing comparative analysis and calculating the differences between the two in terms of structure, nodes, and edges, detailed information on decision bias is obtained. Pre-defined expert decision-making path knowledge graph It is a graph-structured knowledge base representing standard emergency decision-making processes. This knowledge base can be obtained through domain expert interviews, historical success case reviews, and structured emergency plans. Nodes in the graph typically include detailed scenario descriptions and standard operations, while edges are labeled with transition conditions or temporal constraints. The comparison process employs graph matching or graph similarity algorithms, specifically through subgraph isomorphism detection and difference enumeration: the terminal compares the decision path topology graph... With the pre-defined expert decision-making path knowledge graph Perform matching and alignment. For matched nodes and edges, check if their attributes (such as operation parameters and execution timing) are consistent. Unmatched parts are identified as deviations. Decision deviation information. It can be a structured list containing, but is not limited to, the following types: 1) Missing critical operations: Decision nodes that exist in the expert graph but are completely missing from the actual graph. 2) Incorrect operation order: The temporal or causal relationship between two nodes in the actual graph contradicts the logical order defined in the expert graph. 3) Operation-state mismatch: The operation actually performed in a certain state is inconsistent with the standard operation defined for that state in the expert graph. 4) Redundant or invalid operations: Nodes that exist in the actual graph but do not exist in the expert graph and cannot be associated with any standard decision logic.

[0122] S605 generates a full-process evaluation report of the exercise based on traditional operational scores and decision-making deviation information.

[0123] Specifically, the terminal integrates the traditional operation score calculated by S602 and the decision deviation information analyzed by S604, and automatically synthesizes a structured evaluation report of the entire exercise process according to a preset report template. The preset report template can be set according to the template requirements of actual work reports.

[0124] This embodiment provides an emergency training and drill method based on 3D simulation. It summarizes the operation matching results, user operation instructions, and physical environment state parameters for all cycles; averages the operation matching scores for each cycle to obtain a traditional operation score; then, based on a preset set of key operation physical state mapping rules, it analyzes the user operation instruction sets and physical environment state parameters for each cycle to identify key decision points and their logical connections during the actual drill, constructing a decision path topology map; subsequently, it performs a refined graph structure comparison with an expert decision path knowledge graph, automatically identifying specific decision logic deviations, including omitted operations, incorrect sequences, and operation mismatches, obtaining decision deviation information; and combines the traditional operation score with the decision deviation information to generate a full-process evaluation report for the drill. This achieves a qualitative leap in drill evaluation from single-point operation scoring to global decision logic reconstruction and diagnosis, intuitively revealing the deviations between the user's thinking path and the optimal path in actual handling, thereby improving the relevance and effectiveness of emergency training.

[0125] In the aforementioned emergency training and drill method based on 3D simulation, the following steps are taken: First, the physical environment state parameters and user operation command set of the previous cycle are acquired. Based on these parameters, the occurrence of emergencies in the current cycle is analyzed to obtain the emergency event identification information. Second, based on the physical environment state parameters, user operation command set, preset current environment parameters, and emergency event identification information of the previous cycle, the physical environment state parameters for the current cycle are obtained, and a 3D drill scenario for the current cycle is generated. Third, when executing the 3D drill scenario, the user operation command set and current physiological signals for the current cycle are acquired. Fourth, based on the emergency event identification information, user operation command set, physical environment state parameters, and current physiological signals, the user's emergency operation is evaluated to obtain the operation matching result and cognitive load assessment result for the current cycle. The operation matching result characterizes the degree of conformity between the user's emergency operation and emergency logic, while the cognitive load assessment result characterizes the user's psychological load level. This system upgrades traditional static script-driven drills into a dynamic closed-loop system. By introducing an intelligent event generation mechanism based on reinforcement learning, it breaks the deterministic limitations of pre-set scripts, enhancing the unpredictability and practical challenge of drills. Simultaneously, by integrating multimodal physiological and operational data, the assessment dimensions are extended from superficial operational correctness to deep cognitive load states, achieving a leap from behavioral assessment to cognitive shaping, thereby improving the relevance and effectiveness of emergency training.

[0126] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0127] Based on the same inventive concept, this application also provides a 3D simulation-based emergency training exercise system for implementing the aforementioned 3D simulation-based emergency training exercise method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more 3D simulation-based emergency training exercise system embodiments provided below can be found in the limitations of the 3D simulation-based emergency training exercise method described above, and will not be repeated here.

[0128] In one exemplary embodiment, such as Figure 2 As shown, an emergency training and drill system 200 based on three-dimensional simulation is provided, including:

[0129] The data acquisition module 201 is used to acquire the physical environment status parameters and user operation instruction set of the previous cycle;

[0130] The emergency event determination module 202 is used to analyze the occurrence of emergencies in the current period based on the physical environment status parameters and user operation instruction set of the previous period, and obtain the emergency event identification information of the current period.

[0131] The 3D scene construction module 203 is used to obtain the physical environment state parameters of the current period based on the physical environment state parameters of the previous period, the user operation instruction set, the preset current environment parameters and the emergency event identification information of the current period, and generate the 3D exercise scene of the current period based on the physical environment state parameters of the current period.

[0132] The execution data module 204 is used to acquire the user operation instruction set and current physiological signals of the current cycle when executing the three-dimensional training scenario of the current cycle;

[0133] The data analysis module 205 is used to evaluate the user's emergency operation based on the emergency event identification information, user operation instruction set, physical environment status parameters and current physiological signals in the current period, and to obtain the operation matching result and cognitive load assessment result for the current period. The operation matching result is used to characterize the degree of conformity between the user's emergency operation and emergency logic, and the cognitive load assessment result is used to characterize the user's psychological load level.

[0134] Furthermore, the emergency determination module can also be used for:

[0135] The physical environment state parameters and user operation instruction set of the previous cycle are concatenated to obtain the enhanced scenario state vector of the previous cycle; and the enhanced scenario state vector is input into the preset disaster strategy network to obtain the emergency event identification information of the current cycle.

[0136] The disaster response strategy network includes:

[0137] The input layer receives the enhanced scene state vector, performs standardization processing on the enhanced scene state vector to obtain a standardized state vector, and then transmits the standardized state vector to the feature encoding layer.

[0138] The feature encoding layer is used to perform at least one fully connected transformation on the normalized state vector, extract the high-order features of the normalized state vector, obtain the scene state encoding vector, and transmit the scene state encoding vector to the policy output layer.

[0139] The policy analysis layer is used to perform a fully connected transformation on the scene state encoding vector to obtain a logical value vector, and then transmit the logical value vector to the output layer.

[0140] The output layer is used to map the logic value vector through a preset non-linear activation function to obtain the probability distribution vector of the sudden event, and output the sudden event identification information of the current period based on the probability distribution vector of the sudden event.

[0141] Furthermore, the system also includes a disaster response strategy network construction module, which can be used for:

[0142] Acquire preset historical exercise data, which includes historical enhanced scenario state vectors, historical emergency event identification information, and historical operation difficulty scores corresponding to historical emergency event identification information; and construct an initial policy network based on preset policy network initialization parameters.

[0143] Based on the initial policy network, the state vectors of historical augmented scenarios are processed to obtain the predicted emergency event identification information corresponding to the state vectors of historical augmented scenarios; and the predicted emergency event identification information is matched with the historical emergency event identification information in the historical exercise records to determine the historical operation difficulty score corresponding to the predicted emergency event identification information.

[0144] Based on a preset reward function, predicted emergency event identifiers, and the corresponding historical operational difficulty scores, the network prediction reward value is calculated; the expression for the reward function is:

[0145]

[0146] in, It is the network's predicted reward value. It is the first weighting coefficient. It is the second weighting coefficient. It is the third weighting coefficient. It is a predictive information for emergencies. It is an indicator function. It is a warning information for predicting emergencies. The historical operational difficulty score, It is the first evaluation threshold. It is the second evaluation threshold. It is a diverse reward function;

[0147] Based on the current reward value, the policy loss of the initial policy network is calculated, and the initial policy network is iteratively trained based on the policy loss until the initial policy network meets the preset iteration termination condition, thus obtaining the disaster policy network.

[0148] Furthermore, the 3D scene building module can also be used for:

[0149] Based on the physical environment state parameters of the previous cycle, the user operation instruction set, and the preset current environment parameters, the basic physical environment state parameters of the current cycle are obtained by calculation through the preset hybrid physical simulation model.

[0150] Based on the current period's emergency event identification information and the preset emergency time environment parameter adjustment rule set, determine the correction instructions for the basic physical environment state parameters;

[0151] The basic physical environment state parameters for the current period are corrected based on the correction instructions to obtain the physical environment state parameters for the current period.

[0152] Furthermore, the data analysis module can also be used for:

[0153] Based on a preset set of emergency response operation rules, the emergency identification information of the current period is mapped to obtain the standard emergency response operation set corresponding to the emergency identification information.

[0154] Quantify the differences between the standard emergency response operation set and the user operation instruction set in the current cycle to obtain the operation matching results;

[0155] Feature extraction is performed on the current physiological signals to obtain physiological feature data;

[0156] Physiological characteristic data are input into a preset cognitive load prediction model to obtain initial cognitive load values;

[0157] Based on the preset load calibration coefficient, the initial cognitive load value is numerically corrected to obtain the cognitive load assessment result.

[0158] Furthermore, the system also includes a full-cycle analysis module, which can be used for:

[0159] Obtain the operation matching results, user operation command set, and physical environment status parameters for each cycle;

[0160] The average value of the operation matching results for each cycle is calculated to obtain the traditional operation score;

[0161] Based on a preset set of key operation physical state mapping rules, the user operation instruction set and physical environment state parameters of each cycle are analyzed to construct a decision path topology graph.

[0162] By comparing the differences between the decision path topology graph and the pre-defined expert decision path knowledge graph, decision bias information is obtained.

[0163] Based on traditional operational scores and decision-making bias information, an evaluation report of the entire exercise process is generated.

[0164] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the emergency training exercise method based on three-dimensional simulation as described above.

[0165] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0166] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0167] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. An emergency training and drill method based on three-dimensional simulation, characterized in that, The method includes: Obtain the physical environment status parameters and user operation command set of the previous cycle; Based on the physical environment state parameters and user operation instruction set of the previous cycle, the occurrence of emergencies in the current cycle is analyzed to obtain the emergency identification information of the current cycle. Based on the physical environment state parameters of the previous cycle, the user operation instruction set, the preset current environment parameters, and the emergency event identification information of the current cycle, the physical environment state parameters of the current cycle are obtained, and a three-dimensional exercise scenario for the current cycle is generated based on the physical environment state parameters of the current cycle. When executing the three-dimensional training scenario of the current cycle, the user operation instruction set and current physiological signals of the current cycle are acquired; Based on the emergency event identification information, user operation instruction set, physical environment state parameters, and current physiological signals of the current period, the user's emergency operation situation is evaluated to obtain the operation matching result and cognitive load assessment result of the current period. The operation matching result is used to characterize the degree of conformity between the user's emergency operation and emergency logic, and the cognitive load assessment result is used to characterize the user's psychological load level.

2. The method according to claim 1, characterized in that, Based on the physical environment state parameters and user operation instruction set of the previous period, the occurrence of emergencies in the current period is analyzed to obtain the emergency identification information of the current period, including: The physical environment state parameters and the user operation instruction set from the previous cycle are concatenated to obtain the enhanced scenario state vector of the previous cycle; and the enhanced scenario state vector is input into the preset disaster strategy network to obtain the emergency event identification information of the current cycle. The disaster response strategy network includes: An input layer is used to receive the enhanced scene state vector, perform normalization processing on the enhanced scene state vector to obtain a normalized state vector, and transmit the normalized state vector to the feature encoding layer. The feature encoding layer is used to perform at least one fully connected transformation on the standardized state vector, extract the high-order features of the standardized state vector, obtain the scene state encoding vector, and transmit the scene state encoding vector to the policy output layer. The strategy analysis layer is used to perform a fully connected transformation on the scene state encoding vector to obtain a logical value vector, and then transmit the logical value vector to the output layer. The output layer is used to map the logic value vector through a preset nonlinear activation function to obtain a probability distribution vector of sudden events, and output the sudden event identification information of the current period based on the probability distribution vector of sudden events.

3. The method according to claim 2, characterized in that, The disaster response strategy network was obtained through the following methods: Obtain preset historical exercise data, which includes historical enhanced scenario state vectors, historical emergency event identification information, and historical operation difficulty scores corresponding to the historical emergency event identification information. And based on the preset policy network initialization parameters, an initial policy network is constructed; Based on the initial policy network, the historical enhanced scene state vector is processed to obtain the predicted sudden event identification information corresponding to the historical enhanced scene state vector; The predicted emergency event identification information is matched with the historical emergency event identification information in the historical exercise record to determine the historical operation difficulty score corresponding to the predicted emergency event identification information. Based on a preset reward function, predicted emergency event identifiers, and the historical operation difficulty scores corresponding to the predicted emergency event identifiers, the network predicted reward value is calculated; the expression for the reward function is: in, It is the network's predicted reward value. It is the first weighting coefficient. It is the second weighting coefficient. It is the third weighting coefficient. It is a predictive information for emergencies. It is an indicator function. It is a predictive information for emergencies. The historical operational difficulty score, It is the first evaluation threshold. It is the second evaluation threshold. It is a diverse reward function; Based on the current reward value, the policy loss of the initial policy network is calculated, and the initial policy network is iteratively trained based on the policy loss until the initial policy network meets the preset iteration termination condition, thus obtaining the disaster policy network.

4. The method according to claim 1, characterized in that, The process of obtaining the physical environment state parameters for the current period based on the physical environment state parameters of the previous period, the user operation instruction set, the preset current environment parameters, and the emergency event identification information of the current period includes: Based on the physical environment state parameters of the previous cycle, the user operation instruction set, and the preset current environment parameters, the basic physical environment state parameters of the current cycle are calculated using a preset hybrid physical simulation model. Based on the current period's emergency event identification information and the preset emergency time environment parameter adjustment rule set, the correction instruction for the basic physical environment state parameters is determined; Based on the correction instruction, the basic physical environment state parameters for the current period are corrected to obtain the physical environment state parameters for the current period.

5. The method according to claim 1, characterized in that, The system assesses the user's emergency response based on the current period's emergency event identification information, user operation command set, physical environment state parameters, and current physiological signals, obtaining the current period's operation matching result and cognitive load assessment result, including: Based on a preset set of emergency response operation rules, the emergency identification information of the current period is mapped to obtain a standard emergency response operation set corresponding to the emergency identification information. The difference between the standard emergency response operation set and the user operation instruction set in the current period is quantified to obtain the operation matching result; Feature extraction is performed on the current physiological signal to obtain physiological feature data; The physiological characteristic data is input into a preset cognitive load prediction model to obtain an initial cognitive load value; Based on a preset load calibration coefficient, the initial cognitive load value is numerically corrected to obtain the cognitive load assessment result.

6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Obtain the operation matching results, user operation instruction set, and physical environment status parameters for each cycle; The average value of the operation matching results for each cycle is calculated to obtain the traditional operation score; Based on a preset set of key operation physical state mapping rules, the user operation instruction set and the physical environment state parameters of each cycle are analyzed to construct a decision path topology graph. By comparing the decision path topology with the preset expert decision path knowledge graph, decision bias information is obtained; Based on the traditional operation score and the decision deviation information, an evaluation report of the entire exercise process is generated.

7. An emergency training and drill system based on three-dimensional simulation, characterized in that, The system includes: The data acquisition module is used to acquire the physical environment status parameters and user operation command set of the previous cycle; The emergency event determination module is used to analyze the occurrence of emergencies in the current period based on the physical environment state parameters and the user operation instruction set of the previous period, and obtain the emergency event identification information of the current period. The three-dimensional scene construction module is used to obtain the physical environment state parameters of the current period based on the physical environment state parameters of the previous period, the user operation instruction set, the preset current environment parameters and the emergency event identification information of the current period, and generate the three-dimensional exercise scene of the current period based on the physical environment state parameters of the current period. The execution data module is used to acquire the user operation instruction set and current physiological signals for the current cycle when executing the three-dimensional training scenario of the current cycle; The data analysis module is used to assess the user's emergency response based on the emergency event identification information, user operation instruction set, physical environment state parameters, and current physiological signals in the current period, and to obtain the operation matching result and cognitive load assessment result for the current period. The operation matching result is used to characterize the degree of conformity between the user's emergency operation and emergency logic, and the cognitive load assessment result is used to characterize the user's psychological load level.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images