Nuclear emergency psychological training system and method

Abstract

The invention discloses a nuclear emergency psychological training system and method, and relates to the technical field of data processing. A training preparation instruction is received, and reference physiological data is collected according to the training preparation instruction; a training starting instruction is sent to the main system for simulation training; the method comprises the following steps: collecting time sequence motion data and time sequence physiological data of a target person in real time through wearable collection equipment, processing the time sequence physiological data to obtain target physiological data, and processing the time sequence motion data to obtain target motion data; inputting the target physiological data and the target motion data into a target model to obtain a training score; and if the training score is smaller than the score threshold, determining that the target person is unqualified, and performing training plan updating on the target person. Physiological index monitoring and action analysis are combined, quantitative evaluation of nuclear emergency psychological training is achieved, and compared with traditional psychological training, evaluation precision is improved, training scientificity is improved, and training is more targeted and efficient.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a nuclear emergency psychological training system and method. Background Technology

[0002] With the widespread application of nuclear energy technology, the importance of nuclear emergency psychological training has become increasingly prominent. In recent years, nuclear emergency psychological training techniques have continuously developed, evolving from traditional theoretical explanations and simulations to today's use of high-tech methods such as virtual reality and psychological testing systems. These methods have become more scientific and diverse, effectively improving the psychological qualities and coping abilities of emergency personnel.

[0003] Publication No. CN118253060A discloses a VR-based fire emergency rescue simulation training system. The system comprises a virtual environment generation layer, a training control layer, an intelligent analysis layer, and an interactive teaching layer. The virtual environment generation layer provides realistic fire scenarios; the training control layer manages the training process and interactive operations; the intelligent analysis layer provides behavioral analysis and guidance suggestions; and the interactive teaching layer provides teaching resources and online communication functions. The system hardware includes a VR headset display, VR controllers and trackers, a computer, and a sound output device. The system database includes a scenario database, a firefighter database, and a teaching resource database. This system uses 3D modeling and physics engine technology to construct realistic fire scenarios through the virtual environment generation layer. This highly realistic simulation environment not only allows firefighters to conduct real-world simulations under safe conditions but also enables them to fully experience the urgency and complexity of a fire scene. The training control layer can set diverse rescue tasks, such as firefighting, evacuation, and search and rescue, according to different training needs. Simultaneously, it can control the time flow of the training process, simulating the emergency time pressure in real-world situations, thereby helping firefighters improve their response capabilities in various scenarios.

[0004] In nuclear emergency psychological training, traditional methods are difficult to accurately quantify and evaluate the training effect, resulting in a lack of accuracy and efficiency in the training. Summary of the Invention

[0005] The purpose of this invention is to solve the problem mentioned in the background art that traditional methods are difficult to achieve accurate quantitative evaluation of training effects, resulting in a lack of accuracy and efficiency in training, and to propose a nuclear emergency psychological training system and method.

[0006] A first aspect of this invention provides a nuclear emergency psychological training method, wherein the target personnel wear training simulation equipment, the training simulation equipment including virtual reality equipment and wearable data collection equipment, the method comprising:

[0007] The system receives training preparation instructions from the main system and collects baseline physiological data of the target personnel according to the training preparation instructions; the main system is used to control each training simulation device.

[0008] Send a training start command to the main system so that the virtual reality device can receive the start command and perform simulation training;

[0009] Wearable data acquisition devices are used to collect real-time temporal motion data and temporal physiological data of target personnel. The temporal physiological data is processed to obtain target physiological data, and the temporal motion data is processed to obtain target motion data.

[0010] The target physiological data and the target motion data are input into the target model to obtain a training score;

[0011] If the training score is less than the scoring threshold, the target personnel are deemed unqualified, and the training plan for the target personnel is updated.

[0012] Optionally, the time-series physiological data is processed to obtain target physiological data, which includes audio data, skin conductivity data, and pupillary change data. The audio data includes heart rate data and respiratory data. The audio data processing procedure includes:

[0013] The audio data is filtered to obtain high-quality audio data, and the high-quality audio data is subjected to short-time Fourier transform to obtain the time-frequency matrix and phase information;

[0014] The time-frequency matrix is ​​decomposed to obtain a basis matrix and an activation matrix. Based on minimizing the cost function, the basis matrix and the activation matrix are iteratively updated using a multiplication iterative algorithm until a preset condition is met, at which point the target basis matrix and the target activation matrix are output.

[0015] The feature mask is calculated based on the basis matrix and the activation matrix. The feature mask is multiplied by the time-frequency matrix to obtain the feature components. The target physiological data is reconstructed based on the feature components and the phase information.

[0016] Optionally, the temporal motion data is processed to obtain the target motion data, including:

[0017] The time-series motion data is frame-sampling to obtain multiple motion images. The target motion image is segmented to obtain a set of region images. The region images in the set of region images are projected to obtain a set of region vectors. The target motion image is any one of the multiple motion images.

[0018] Add corresponding position coordinates to each region vector in the region vector set to obtain a feature vector set, assign weights to each feature vector in the feature vector set to obtain a weight vector set, and combine the weight vectors in the weight vector set whose weights are greater than the weight threshold to obtain a target weight vector set.

[0019] Linear projection is performed on the target weight vectors in the target weight vector set to obtain query vector, key vector, and value vector. Attention features are calculated based on the query vector, key vector, and value vector. A feature set is obtained by combining the attention features. The target weight vector is any one of the weight vectors in the target weight vector set.

[0020] The target motion data is obtained by fusing the feature vectors in the feature vector set and the attention features in the feature set according to the position coordinates.

[0021] Optionally, before inputting the target physiological data and the target motion data into the target model to obtain a training score, the update process of the target model includes:

[0022] Historical training data and a scoring comparison table are obtained, and historical physiological data and historical motion data at the same time are fused to obtain fused training data; the historical training data includes historical physiological data and historical motion data.

[0023] The fused training data and the scoring table are input into a preset model for training to obtain model update parameters. The model parameters of the preset model are then updated according to the model update parameters to obtain the target model.

[0024] Optionally, updating the training plan for the target personnel includes:

[0025] A training score sequence is obtained by acquiring training scores within a preset time period. The difference between training scores at adjacent times in the training score sequence is calculated, and the difference curve is obtained by fitting each training score difference.

[0026] The slope of the curve is calculated based on the difference curve. If the slope of the curve is greater than the slope threshold, the difficulty of the training plan will increase.

[0027] A second aspect of this invention provides a nuclear emergency psychological training system, the system comprising:

[0028] The instruction receiving module is used to receive training preparation instructions sent by the main system and collect baseline physiological data of the target personnel according to the training preparation instructions; the main system is used to control each training simulation device.

[0029] The instruction sending module is used to send training start instructions to the main system so that the virtual reality device can receive the start instructions and perform simulation training.

[0030] The data processing module is used to collect time-series motion data and time-series physiological data of the target person in real time through wearable acquisition devices, process the time-series physiological data to obtain target physiological data, and process the time-series motion data to obtain target motion data.

[0031] The model evaluation module is used to input the target physiological data and the target motion data into the target model to obtain a training score;

[0032] The plan update module is used to determine that the target personnel are unqualified if the training score is less than the score threshold, and to update the training plan for the target personnel.

[0033] Optionally, the data processing module includes:

[0034] The data transformation module is used to filter the audio data to obtain high-quality audio data, and to perform a short-time Fourier transform on the high-quality audio data to obtain a time-frequency matrix and phase information.

[0035] The matrix decomposition module is used to decompose the time-frequency matrix to obtain a base matrix and an activation matrix. Based on minimizing the cost function, the base matrix and the activation matrix are iteratively updated using a multiplication iterative algorithm until a preset condition is met, at which point the target base matrix and the target activation matrix are output.

[0036] The data reconstruction module is used to calculate a feature mask based on the basis matrix and the activation matrix, multiply the feature mask by the time-frequency matrix to obtain feature components, and reconstruct the target physiological data based on the feature components and the phase information.

[0037] Optionally, the data processing module further includes:

[0038] The region segmentation module is used to extract frames from the temporal motion data to obtain multiple motion images, segment the target motion image to obtain a region image set, and project the region images in the region image set to obtain a region vector set; the target motion image is any one of the multiple motion images.

[0039] The weight fusion module is used to add corresponding position coordinates to each region vector in the region vector set to obtain a feature vector set, assign weights to each feature vector in the feature vector set to obtain a weight vector set, and combine the weight vectors in the weight vector set whose weights are greater than the weight threshold to obtain a target weight vector set.

[0040] The vector projection module is used to perform linear projection on the target weight vector in the target weight vector set to obtain a query vector, a key vector, and a value vector; calculate attention features based on the query vector, key vector, and value vector; and combine the attention features to obtain a feature set; the target weight vector is any one of the weight vectors in the target weight vector set.

[0041] The feature fusion module is used to fuse the feature vectors in the feature vector set and the attention features in the feature set according to the position coordinates to obtain target motion data.

[0042] Optionally, the model evaluation module includes:

[0043] The model training module is used to acquire historical training data and a scoring comparison table, and to fuse historical physiological data and historical motion data at the same time to obtain fused training data; the historical training data includes historical physiological data and historical motion data.

[0044] The model update module is used to input the fused training data and the scoring comparison table into a preset model for training to obtain model update parameters, and update the model parameters of the preset model according to the model update parameters to obtain the target model.

[0045] Optionally, the planned update module includes:

[0046] The curve fitting module is used to obtain a training score sequence by acquiring training scores within a preset time period, calculate the training score difference between adjacent time points in the training score sequence, and fit each training score difference to obtain a difference curve.

[0047] The training plan difficulty update module is used to calculate the slope of the curve based on the difference curve. If the slope of the curve is greater than the slope threshold, the difficulty of the training plan is increased.

[0048] The beneficial effects of this invention are:

[0049] This invention proposes a method for nuclear emergency psychological training. It involves receiving a training preparation command and collecting baseline physiological data based on that command; sending a training start command to the main system for simulation training; using wearable data acquisition devices to collect real-time temporal and physiological data of the target personnel; processing the temporal physiological data to obtain target physiological data; and processing the temporal motion data to obtain target motion data; inputting the target physiological and motion data into a target model to obtain a training score; if the training score is lower than a threshold, the target personnel are deemed unqualified, and the training plan is updated. By combining physiological indicator monitoring with motion analysis, this method achieves quantitative evaluation of nuclear emergency psychological training. Compared to traditional psychological training, it improves evaluation accuracy, enhances the scientific rigor of the training, and makes the training more targeted and efficient. Attached Figure Description

[0050] Figure 1 A flowchart of a nuclear emergency psychological training method is provided as an embodiment of the present invention;

[0051] Figure 2 This invention provides a framework diagram of a nuclear emergency psychological training system. Detailed Implementation

[0052] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0053] This invention provides a method for psychological training in nuclear emergency response. See also... Figure 1 , Figure 1 A flowchart illustrating a nuclear emergency psychological training method provided in an embodiment of the present invention. The method includes the following steps:

[0054] S101 receives the training preparation instruction sent by the main system and collects the baseline physiological data of the target personnel according to the training preparation instruction;

[0055] S102, send a training start command to the main system so that the virtual reality device can receive the start command and perform simulation training;

[0056] S103: Real-time collection of temporal motion data and temporal physiological data of the target person through wearable data acquisition devices; processing of temporal physiological data to obtain target physiological data; processing of temporal motion data to obtain target motion data.

[0057] S104, Input the target physiological data and target motion data into the target model to obtain a training score;

[0058] S105. If the training score is less than the scoring threshold, the target personnel are deemed unqualified, and the training plan for the target personnel is updated.

[0059] The main system is used to control each training simulation device;

[0060] The nuclear emergency psychological training method provided by this invention combines physiological indicator monitoring with action analysis to achieve quantitative assessment of nuclear emergency psychological training. Compared with traditional psychological training, it improves assessment accuracy, enhances the scientific nature of training, and makes training more targeted and efficient.

[0061] In one implementation, the main system is used to control various training simulation devices. For example, the main system and multiple subsystems are used. The main system (computer host) is used to send instructions and receive data uploaded by the subsystems. The subsystems (training simulation devices) are used to provide a simulated three-dimensional environment and collect physiological data during the simulation training process. The simulated three-dimensional environment is a nuclear emergency scenario, such as simulating the impact of radiation alarm sound and light stimulation on the accuracy of operation, so as to train operators.

[0062] In one implementation, the training simulation device includes a virtual reality device and a wearable data acquisition device. The virtual reality device is used to provide a simulated three-dimensional environment, and the wearable data acquisition device is used to collect physiological data.

[0063] In one implementation, a main system and multiple subsystems work together to achieve precise monitoring and data collection of the nuclear emergency psychological training process. The main system, as the core control unit, can send training preparation and start commands, precisely control the operating status of each training simulation device, and ensure the orderly conduct of the training process. Virtual reality devices in the subsystems provide immersive, simulated 3D nuclear emergency scenarios, simulating various stimuli in a real nuclear emergency environment, such as radiation alarms and sound and light stimuli, thus providing operators with a highly realistic training environment, allowing them to receive psychological training under near-real-world conditions. Wearable data collection devices can collect real-time temporal physiological and motion data of the target personnel. This data, after processing, yields target physiological and motion data, providing a comprehensive and accurate data foundation for subsequent training scoring.

[0064] In one implementation, processed target physiological and motion data are input into the target model to obtain a training score. This achieves an objective and quantitative evaluation of the training effect on the target personnel. The data-driven scoring system avoids the drawbacks of traditional training that relies on subjective judgment, and can more accurately reflect the performance of the target personnel in nuclear emergency psychological training. The training score not only integrates data from both physiological and motion dimensions, but also ensures high scientific validity and reliability through algorithmic processing of the target model.

[0065] In one embodiment, time-series physiological data is processed to obtain target physiological data, which includes audio data, skin conductivity data, and pupillary change data. The audio data includes heart rate data and respiratory data. The audio data processing procedure includes:

[0066] The audio data is filtered to obtain high-quality audio data, and the high-quality audio data is subjected to short-time Fourier transform to obtain the time-frequency matrix and phase information;

[0067] The time-frequency matrix is ​​decomposed to obtain the basis matrix and activation matrix. Based on minimizing the cost function, the basis matrix and activation matrix are iteratively updated through a multiplication iterative algorithm until the preset conditions are met, at which point the target basis matrix and target activation matrix are output.

[0068] The feature mask is calculated based on the basis matrix and activation matrix. The feature mask is multiplied by the time-frequency matrix to obtain the feature components. The target physiological data is then reconstructed based on the feature components and phase information.

[0069] In one implementation, high-quality audio is obtained by filtering audio data. High-quality audio is obtained by filtering time-series physiological data through operations such as bandpass filtering, endpoint detection, and periodic screening. This high-quality audio still contains mixed components such as breathing, heartbeat, motion noise, and environmental noise, but the high-frequency environmental noise has been removed (higher frequencies than 8000Hz are removed), and physiological signals in the 500Hz-8000Hz frequency band are retained. Segments with obvious physiological characteristics are screened out through periodic analysis.

[0070] In one implementation, a short-time Fourier transform (STFT) is performed on the input audio signal to obtain a time-frequency matrix containing the spectral and temporal characteristics of all components in the mixed audio, as well as phase information, which is used for subsequent signal reconstruction.

[0071] In one implementation, the time-frequency matrix is ​​decomposed to obtain a basis matrix and an activation matrix. Minimizing the cost function, which is also the optimization objective, is used to achieve accurate separation of respiratory and heartbeat signals from noise. The cost function is minimized for example: 1. Using KL divergence (when β = 1) to measure the difference between the time-series physiological data and the target physiological data, ensuring that the reconstructed signal closely approximates the original mixed signal; 2. Applying L1 regularization to the activation matrix, forcing only a single signal component (such as respiration or heartbeat) to be activated at the same time, reducing mutual interference between different components; 3. Introducing unsupervised components and their corresponding activation matrices, handling undefined noise through additional error terms, improving the algorithm's adaptability to complex environments, and simultaneously using the hyperparameter λ... xt , λ hx , λ zs Control the regularization intensity of heart rate, respiration, and noise sub-components to avoid overfitting.

[0072] In one implementation, the activation matrix is ​​updated. Through molecules And denominator (W*Λ) β-1 The numerator (β) is adjusted by increasing the value of W, V, and Λ. The larger the difference, the larger the numerator, pushing the activation matrix to increase and reducing reconstruction error. Here, W is the basis matrix, V is the time-frequency matrix, Λ is the reconstruction matrix (Λ = WH, the time-frequency matrix reconstructed from W and H in the current iteration, used to approximate the initial time-frequency matrix V), β is the KL divergence (value 1), μ is the control parameter for L1 sparsity penalty (used to constrain the sparsity of H, ensuring that only a single signal component is activated at any given time), and H is the activation matrix. The operation is element-wise multiplication. The basis matrix is ​​updated by adjusting the V / WH ratio to make the spectral template of W closer to the characteristics of real breathing and heartbeat signals. In each iteration, the activation matrix is ​​updated first, then each submatrix of the basis matrix is ​​updated, and this process is repeated. When the number of iterations reaches the preset maximum value, or the reconstruction error tends to stabilize and no longer decreases significantly, the iteration is stopped, and the final optimization result is used as the target basis matrix and the target activation matrix.

[0073] In one implementation, a feature mask is calculated based on the basis matrix and the activation matrix: Feature Mask = (Target Basis Matrix * Target Activation Matrix) / (Basis Matrix * Activation Matrix). The feature mask is then multiplied by the time-frequency matrix to obtain feature components, such as respiratory or heartbeat features. The target physiological data (e.g., respiration or heartbeat) is reconstructed using an inverse short-time Fourier transform, combining phase information and feature components.

[0074] In one implementation, skin conductivity data and pupil change data need to be cleaned of outliers and normalized. The physiological data includes audio data, skin conductivity data, and pupil change data. The processed data are aligned according to time series and then weighted and summed to obtain the target physiological data.

[0075] In one embodiment, processing temporal motion data to obtain target motion data includes:

[0076] Frames are extracted from the temporal motion data to obtain multiple motion images. The target motion image is segmented to obtain a set of region images. The region images in the set of region images are projected to obtain a set of region vectors. The target motion image is any one of the multiple motion images.

[0077] Add corresponding position coordinates to each region vector in the region vector set to obtain a feature vector set. Assign weights to each feature vector in the feature vector set to obtain a weight vector set. Combine the weight vectors in the weight vector set whose weights are greater than the weight threshold to obtain the target weight vector set.

[0078] Linear projection is performed on the target weight vectors in the target weight vector set to obtain the query vector, key vector, and value vector. Attention features are calculated based on the query vector, key vector, and value vector. The feature set is obtained by combining the attention features. The target weight vector is any weight vector in the target weight vector set.

[0079] The target motion data is obtained by fusing the feature vectors in the feature vector set and the attention features in the feature vector set based on the position coordinates.

[0080] In one implementation, the operation of the target person is captured by a camera; the temporal motion data is processed by frame extraction, segmentation and projection, which can transform continuous motion information into a structured region vector form, realize the preliminary extraction and quantification of key visual information in the motion process, provide a standardized input basis for subsequent feature processing, and ensure that the features of different motion images are processed in a unified dimension.

[0081] In one implementation, location coordinates are added to the region vector and weights are assigned before filtering. This can retain the spatial location information of the region while highlighting the region features that are more critical to motion analysis through weight adjustment and weakening the interference of secondary regions. Feature filtering can focus on the key regions of motion and weaken the influence of irrelevant regions, thereby improving the robustness to changes in illumination and partial occlusion without significantly increasing computational complexity.

[0082] In one implementation, the target weight vector is linearly projected to calculate attention features. This can model the global association between different regions through the interaction of query, key, and value vectors, making up for the shortcomings of local region features in expressing motion coherence and correlation. This allows the feature set to contain information not only from a single region, thus improving the accuracy of motion detection.

[0083] In one embodiment, before inputting target physiological data and target motion data into the target model to obtain a training score, the target model update process includes:

[0084] Historical training data and scoring tables are obtained, and historical physiological data and historical exercise data at the same time are fused to obtain fused training data; historical training data includes historical physiological data and historical exercise data.

[0085] The fused training data and scoring table are input into the preset model for training to obtain model update parameters. The model parameters of the preset model are then updated based on the model update parameters to obtain the target model.

[0086] In one implementation, fusing historical physiological data and historical motion data at the same moment before inputting them into the model for training can enhance the ability to capture the correlation between features. Physiological data (such as heart rate, respiratory rate, and skin conductance response) and motion data (such as joint angle and operation speed) often have a strong coupling relationship at the same moment. For example, sudden stress may simultaneously manifest as a sudden increase in heart rate and an increase in the amplitude of hand tremors. Through data fusion, the above-mentioned cross-modal features are integrated into a unified input, and the neural network can directly learn the joint distribution pattern of the two types of data, avoiding the feature fragmentation problem that may occur when inputting separately, improving the correlation between physiological data and motion data and psychological scores, and enhancing the robustness of score prediction.

[0087] In one implementation, the parameters used to drive model updates by fusing training data emphasize the interaction weights of cross-modal features. In addition to basic parameters such as inter-layer connection weights and bias terms in traditional neural networks, these parameters also include adjustment parameters for the fusion layer. For example, attention weights are used to balance the importance of physiological and motor features. In fine manipulation scenarios, the weights of motor data are increased. A transformation matrix is ​​used to map the two data feature dimensions, mapping the temporal features of physiological signals and the frequency features of motor signals to the same dimensional space. Through training iterations, these parameters enable the model to dynamically adapt to the correlation strength between physiological and motor data in different scenarios, thereby optimizing the scoring prediction logic.

[0088] In one embodiment, updating the training plan for the target personnel includes:

[0089] The training scores within a preset time period are obtained to obtain a training score sequence. The training score difference between adjacent time points in the training score sequence is calculated, and the difference curve is obtained by fitting each training score difference.

[0090] The slope of the curve is calculated based on the difference curve. If the slope of the curve is greater than the slope threshold, the difficulty of the training plan will increase.

[0091] In one implementation, by acquiring training scores within a preset time period and forming a training score sequence, and then calculating the difference between training scores at adjacent times and fitting it into a difference curve, the progress of the target personnel in the nuclear emergency psychological training process can be accurately quantified and analyzed. The difference curve reflects the trend of training scores over time, avoiding interference from score fluctuations caused by accidental factors on the evaluation of training effectiveness, thereby providing a basis for adjusting subsequent training plans, ensuring that the training plan always aligns with the actual level and needs of the target personnel, and improving the pertinence and effectiveness of the training.

[0092] Based on the same inventive concept, this invention also provides a nuclear emergency psychological training system. See [link to related document]. Figure 2 , Figure 2A schematic diagram of a nuclear emergency psychological training system provided in an embodiment of the present invention includes:

[0093] The instruction receiving module is used to receive training preparation instructions sent by the main system and collect baseline physiological data of the target personnel according to the training preparation instructions; the main system is used to control various training simulation devices.

[0094] The instruction sending module is used to send training start instructions to the main system so that the virtual reality device can receive the start instructions and perform simulation training.

[0095] The data processing module is used to collect real-time temporal motion data and temporal physiological data of the target personnel through wearable data collection devices, process the temporal physiological data to obtain target physiological data, and process the temporal motion data to obtain target motion data.

[0096] The model evaluation module is used to input target physiological data and target motion data into the target model to obtain a training score;

[0097] The plan update module is used to determine that the target personnel are unqualified if the training score is less than the score threshold, and to update the training plan for the target personnel.

[0098] The nuclear emergency psychological training system provided by this invention combines physiological indicator monitoring with action analysis to achieve quantitative assessment of nuclear emergency psychological training. Compared with traditional psychological training, it improves assessment accuracy, enhances the scientific nature of training, and makes training more targeted and efficient.

[0099] In one embodiment, the data processing module includes:

[0100] The data transformation module is used to filter audio data to obtain high-quality audio data, and to perform short-time Fourier transform on the high-quality audio data to obtain the time-frequency matrix and phase information.

[0101] The matrix decomposition module is used to decompose the time-frequency matrix to obtain the basis matrix and activation matrix. Based on minimizing the cost function, the basis matrix and activation matrix are iteratively updated through a multiplication iterative algorithm until the preset conditions are met, and then the target basis matrix and target activation matrix are output.

[0102] The data reconstruction module is used to calculate the feature mask based on the basis matrix and activation matrix, multiply the feature mask by the time-frequency matrix to obtain the feature components, and reconstruct the target physiological data based on the feature components and phase information.

[0103] In one embodiment, the data processing module further includes:

[0104] The region segmentation module is used to extract frames from temporal motion data to obtain multiple motion images, segment the target motion image to obtain a set of region images, and project the region images in the set of region images to obtain a set of region vectors; the target motion image is any one of the multiple motion images.

[0105] The weight fusion module is used to add corresponding position coordinates to each region vector in the region vector set to obtain a feature vector set, assign weights to each feature vector in the feature vector set to obtain a weight vector set, and combine the weight vectors in the weight vector set whose weights are greater than the weight threshold to obtain the target weight vector set.

[0106] The vector projection module is used to perform linear projection on the target weight vector in the target weight vector set to obtain the query vector, key vector, and value vector. Attention features are calculated based on the query vector, key vector, and value vector. A feature set is obtained by combining the attention features. The target weight vector is any weight vector in the target weight vector set.

[0107] The feature fusion module is used to fuse the feature vectors in the feature vector set and the attention features in the feature set according to the position coordinates to obtain the target motion data.

[0108] In one embodiment, the model evaluation module includes:

[0109] The model training module is used to acquire historical training data and a scoring comparison table, and to fuse historical physiological data and historical motion data at the same time to obtain fused training data; the historical training data includes historical physiological data and historical motion data.

[0110] The model update module is used to input the fused training data and the scoring table into the preset model for training to obtain model update parameters, and update the model parameters of the preset model according to the model update parameters to obtain the target model.

[0111] In one embodiment, the planned update module includes:

[0112] The curve fitting module is used to obtain the training score sequence within a preset time period, calculate the training score difference between adjacent time points in the training score sequence, and fit each training score difference to obtain the difference curve.

[0113] The training plan difficulty update module is used to calculate the slope of the curve based on the difference curve. If the slope of the curve is greater than the slope threshold, the difficulty of the training plan is increased.

[0114] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for psychological training in nuclear emergency response, characterized in that, The target personnel wear training simulation equipment, which includes virtual reality equipment and wearable data acquisition devices. The method includes: The system receives training preparation instructions from the main system and collects baseline physiological data of the target personnel according to the training preparation instructions; the main system is used to control each training simulation device. Send a training start command to the main system so that the virtual reality device can receive the start command and perform simulation training; Wearable data acquisition devices are used to collect real-time temporal motion data and temporal physiological data of target personnel. The temporal physiological data is processed to obtain target physiological data, and the temporal motion data is processed to obtain target motion data. The target physiological data and the target motion data are input into the target model to obtain a training score; If the training score is less than the scoring threshold, the target personnel are deemed unqualified, and the training plan for the target personnel is updated.

2. The nuclear emergency psychological training method according to claim 1, characterized in that, The time-series physiological data is processed to obtain target physiological data, which includes audio data, skin conductivity data, and pupillary change data. The audio data includes heart rate data and respiratory data. The audio data processing procedure includes: The audio data is filtered to obtain high-quality audio data, and the high-quality audio data is subjected to short-time Fourier transform to obtain the time-frequency matrix and phase information; The time-frequency matrix is ​​decomposed to obtain a basis matrix and an activation matrix. Based on minimizing the cost function, the basis matrix and the activation matrix are iteratively updated using a multiplication iterative algorithm until a preset condition is met, at which point the target basis matrix and the target activation matrix are output. The feature mask is calculated based on the basis matrix and the activation matrix. The feature mask is multiplied by the time-frequency matrix to obtain the feature components. The target physiological data is reconstructed based on the feature components and the phase information.

3. The nuclear emergency psychological training method according to claim 1, characterized in that, The target motion data is obtained by processing the temporal motion data, including: The time-series motion data is frame-sampling to obtain multiple motion images. The target motion image is segmented to obtain a set of region images. The region images in the set of region images are projected to obtain a set of region vectors. The target motion image is any one of the multiple motion images. Add corresponding position coordinates to each region vector in the region vector set to obtain a feature vector set, assign weights to each feature vector in the feature vector set to obtain a weight vector set, and combine the weight vectors in the weight vector set whose weights are greater than the weight threshold to obtain a target weight vector set. Linear projection is performed on the target weight vectors in the target weight vector set to obtain query vector, key vector, and value vector. Attention features are calculated based on the query vector, key vector, and value vector. A feature set is obtained by combining the attention features. The target weight vector is any one of the weight vectors in the target weight vector set. The target motion data is obtained by fusing the feature vectors in the feature vector set and the attention features in the feature set according to the position coordinates.

4. The nuclear emergency psychological training method according to claim 1, characterized in that, Before inputting the target physiological data and the target motion data into the target model to obtain a training score, the update process of the target model includes: Historical training data and a scoring comparison table are obtained, and historical physiological data and historical motion data at the same time are fused to obtain fused training data; the historical training data includes historical physiological data and historical motion data. The fused training data and the scoring table are input into a preset model for training to obtain model update parameters. The model parameters of the preset model are then updated according to the model update parameters to obtain the target model.

5. The nuclear emergency psychological training method according to claim 1, characterized in that, Updating the training plan for the target personnel includes: A training score sequence is obtained by acquiring training scores within a preset time period. The difference between training scores at adjacent times in the training score sequence is calculated, and the difference curve is obtained by fitting each training score difference. The slope of the curve is calculated based on the difference curve. If the slope of the curve is greater than the slope threshold, the difficulty of the training plan will increase.

6. A nuclear emergency psychological training system, characterized in that, The system includes: The instruction receiving module is used to receive training preparation instructions sent by the main system and collect baseline physiological data of the target personnel according to the training preparation instructions; the main system is used to control each training simulation device. The instruction sending module is used to send training start instructions to the main system so that the virtual reality device can receive the start instructions and perform simulation training. The data processing module is used to collect time-series motion data and time-series physiological data of the target person in real time through wearable acquisition devices, process the time-series physiological data to obtain target physiological data, and process the time-series motion data to obtain target motion data. The model evaluation module is used to input the target physiological data and the target motion data into the target model to obtain a training score; The plan update module is used to determine that the target personnel are unqualified if the training score is less than the score threshold, and to update the training plan for the target personnel.

7. A nuclear emergency psychological training system according to claim 6, characterized in that, The target physiological data includes audio data, skin conductivity data, and pupillary change data. The audio data includes heart rate data and respiratory data. The data processing module includes: The data transformation module is used to filter the audio data to obtain high-quality audio data, and to perform a short-time Fourier transform on the high-quality audio data to obtain a time-frequency matrix and phase information. The matrix decomposition module is used to decompose the time-frequency matrix to obtain a base matrix and an activation matrix. Based on minimizing the cost function, the base matrix and the activation matrix are iteratively updated using a multiplication iterative algorithm until a preset condition is met, at which point the target base matrix and the target activation matrix are output. The data reconstruction module is used to calculate a feature mask based on the basis matrix and the activation matrix, multiply the feature mask by the time-frequency matrix to obtain feature components, and reconstruct the target physiological data based on the feature components and the phase information.

8. A nuclear emergency psychological training system according to claim 6, characterized in that, The data processing module further includes: The region segmentation module is used to extract frames from the temporal motion data to obtain multiple motion images, segment the target motion image to obtain a region image set, and project the region images in the region image set to obtain a region vector set; the target motion image is any one of the multiple motion images. The weight fusion module is used to add corresponding position coordinates to each region vector in the region vector set to obtain a feature vector set, assign weights to each feature vector in the feature vector set to obtain a weight vector set, and combine the weight vectors in the weight vector set whose weights are greater than the weight threshold to obtain a target weight vector set. The vector projection module is used to perform linear projection on the target weight vector in the target weight vector set to obtain a query vector, a key vector, and a value vector; calculate attention features based on the query vector, key vector, and value vector; and combine the attention features to obtain a feature set; the target weight vector is any one of the weight vectors in the target weight vector set. The feature fusion module is used to fuse the feature vectors in the feature vector set and the attention features in the feature set according to the position coordinates to obtain target motion data.

9. A nuclear emergency psychological training system according to claim 6, characterized in that, The model evaluation module includes: The model training module is used to acquire historical training data and a scoring comparison table, and to fuse historical physiological data and historical motion data at the same time to obtain fused training data; the historical training data includes historical physiological data and historical motion data. The model update module is used to input the fused training data and the scoring comparison table into a preset model for training to obtain model update parameters, and update the model parameters of the preset model according to the model update parameters to obtain the target model.

10. A nuclear emergency psychological training system according to claim 6, characterized in that, The plan update module includes: The curve fitting module is used to obtain a training score sequence by acquiring training scores within a preset time period, calculate the training score difference between adjacent time points in the training score sequence, and fit each training score difference to obtain a difference curve. The training plan difficulty update module is used to calculate the slope of the curve based on the difference curve. If the slope of the curve is greater than the slope threshold, the difficulty of the training plan is increased.

Images