Intelligent evaluation method and system for emergency safety training effect

Abstract

The invention discloses an intelligent assessment method and system for an emergency safety training effect, relates to the technical field of emergency safety training effect assessment, and aims to solve the technical problems of inaccurate and incomplete assessment results in the prior art. Standardizing and normalizing a behavior sequence in the emergency safety rescue plan to obtain standard action data, dividing a training set and a verification set to determine a standard action standard value, and marking a standard sample library; s2, establishing an evaluation model, setting an emergency safety training effect evaluation model, dividing sample subsets, training the model by adopting a stochastic gradient descent method, and determining parameters; and S3, data processing and evaluation calculation: obtaining training drill behavior data, constructing a standard sample library and a model training library, training and correcting a model, and calculating a safety training normative score # imgabs0 #, a strain score # imgabs1 # and an efficiency score # imgabs2 #. The method has the advantage of improving the accuracy and comprehensiveness of emergency safety training effect evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of emergency safety training effect evaluation, and more particularly to an emergency safety training effect intelligent evaluation method and system. Background Art

[0002] In the field of emergency safety training, with the development of digital technology, some institutions have begun to experiment with using data to evaluate training effectiveness. Currently, common evaluation methods rely primarily on manual observation and simple data recording. Manual observation often involves experienced instructors scoring trainees' performance based on their subjective judgment. This method is significantly influenced by subjective factors, and evaluation criteria vary between instructors, resulting in a lack of accuracy and consistency in evaluation results. Simple data recording, such as basic information like training completion time and number of operations, fails to fully and thoroughly reflect trainees' performance in key areas such as standardization of movements and adaptability.

[0003] Existing data-based evaluation techniques suffer from numerous flaws when processing emergency safety training data. For one thing, data processing is insufficiently refined, failing to fully tap the potential value of behavioral sequences within emergency safety and rescue plans. In the process of converting plans into evaluation data, behavioral actions are not standardized or normalized, resulting in a lack of comparability between different action data and an inability to accurately construct a standard sample library for evaluation, severely impacting the training effectiveness and accuracy of the evaluation model. Furthermore, existing evaluation models often have simple structures and are unable to effectively extract key features from complex training data. For example, many models fail to simultaneously consider multi-dimensional evaluation metrics such as action standardization, adaptability, and training efficiency, making it difficult to fully reflect the actual effectiveness of emergency safety training. This results in trainers and organizers being unable to accurately identify training issues, which in turn impacts the quality improvement and achievement of training objectives. In light of this, we propose an intelligent evaluation method and system for emergency safety training effectiveness. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent evaluation method and system for emergency safety training effects, so as to solve the technical problem of inaccurate and incomplete evaluation results in the prior art.

[0005] To solve the above technical problems, the present invention provides the following technical solution: a method for intelligently evaluating the effectiveness of emergency safety training, comprising the following steps: S1. Basic data preparation: standardize and normalize the behavior sequences in the emergency safety and rescue plan to obtain standard action data, divide the training set and validation set to determine the standard action values, and mark the standard sample library; S2. Construct an evaluation model, set up an evaluation model for emergency safety training effectiveness, divide the sample subsets, and train the model using the stochastic gradient descent method to determine the parameters; S3. Data processing and evaluation calculation, obtain training exercise behavior data, construct a standard sample library and a model training library, train and correct the model, calculate the safety training compliance score , the emergency response score , the efficiency score , substitute into the evaluation model to calculate the final evaluation score .

[0006] Preferably, the method for standardizing and normalizing the behavior sequence in S1 is as follows: S1.1. Analyze the collected emergency safety rescue plans in detail, identify each action in them, construct a behavior sequence, for each action, clarify its key features and quantify them. By removing the dimensional differences between different action data, the behavior sequence is standardized. For the standardized emergency safety standard action data, use the min-max normalization method to map the data value range to the interval to make different action data comparable; S1.2. Divide the standard action data into a training set and a validation set, calculate the mean and variance of the action data in the training set and the validation set respectively. The mean reflects the average level of the data, and the variance reflects the degree of data dispersion. Combining the mean and variance of both, use the weighted average method to determine the standard value of the emergency safety standard action ; S1.3. Randomly group the emergency safety standard action data into multiple sub-datasets, further divide each sub-dataset into a training set and a validation set, calculate the mean deviation and variance deviation between the training set and the validation set. When both the mean deviation and variance deviation are less than the pre-set threshold, mark it as the emergency safety standard action standard sample library.

[0007] Preferably, the method for setting the evaluation model in S2 is as follows: S2.1. Design the evaluation model as a multi-layer neural network structure. The input layer receives the processed emergency safety training data, the hidden layer extracts data features through non-linear transformation, and the output layer outputs corresponding results according to the evaluation task. The hidden layer uses the ReLU activation function, and the output layer uses the Softmax function to convert the output into a probability distribution for classification decision-making; S2.2. Make a detailed division of the marked emergency safety standard action standard sample library to form multiple sample subsets containing action samples and their corresponding labels. Each action sample represents a specific action in emergency safety training, and the label clarifies information such as whether the action is correct or its category; S2.3. Train the set emergency safety training evaluation model using the stochastic gradient descent method. Use the cross-entropy loss function to measure the difference between the model prediction result and the true label. The model continuously adjusts the parameters to minimize the loss function, and the parameter update step size is controlled by the learning rate. At the same time, use the validation samples in the sample subset to validate the trained model, calculate the model accuracy rate. When the model accuracy rate reaches the pre-set threshold, it indicates that the model has learned the characteristics and rules of the emergency safety standard actions well, determine the model parameters and output the final evaluation model.

[0008] Preferably, in the above S3, the method for obtaining the training and drill behavior data, constructing the standard sample library and the model training library, and training and correcting the model is as follows: S3.1. During the emergency safety training and drill process, use the data acquisition device to obtain the behavior data information of the training personnel in real time and comprehensively. These data cover the action postures, time, and force of the training personnel, and can accurately reflect their actual performance in the drill. S3.2. Convert the emergency safety rescue plan into a behavior sequence diagram. According to the action process and logical relationship of the plan, divide it into several task nodes containing unique behavior actions. Each task node represents a specific action or operation step in the emergency safety training. Set the action threshold for the emergency safety standard action value for each task node to judge whether the training personnel's actions meet the specification requirements. According to the task nodes and action thresholds, construct the standard sample library and the model training library. The standard sample library provides correct reference examples, and the model training library stores the actual training data. S3.3. Initialize the emergency safety training evaluation model, input the data in the standard sample library into the model for training. To accelerate the model convergence speed and improve the training stability, use the batch normalization technique. Continuously adjust the model parameters according to the difference between the model output result and the label in the standard sample library, and repeat the training and correction process.

[0009] Preferably, the calculation method of the above safety training standardization score is as follows: Divide the to-be-evaluated emergency safety training data into different action units or time periods. Through the trained model, analyze the training data in each action unit or time period in detail, calculate the corresponding standard action value and action execution value of the specification, introduce a weight function to measure the difference degree between the two, and sum the weight function values of each action unit or time period to obtain the safety training standardization score. After dividing the to-be-evaluated data, calculate the standard action value of the specification and , the weight function , ; where, is a constant, is the fault tolerance coefficient. is the number of divided action units or time periods, is the weight function value of the th action unit or time period.

[0010] Preferably, the calculation method of the safety training response score is as follows: Deeply analyze the collected training data, evaluate it from three aspects: behavior accuracy, behavior effectiveness, and behavior matching degree. Behavior accuracy is measured by calculating the cosine similarity between the actions of the training personnel and the standard actions. Behavior effectiveness is evaluated according to whether the actions achieve the expected effects. Behavior matching degree is measured by calculating the edit distance between the behavior sequences of the training personnel and the emergency safety rescue plan behavior sequences. Weighted sum the scores of these three aspects to obtain the safety training response score; Calculate the behavior accuracy , the behavior matching degree , and calculate the score according to the formula , where is the action vector of the training personnel, is the standard action vector, and the behavior effectiveness is evaluated according to whether the action achieves the expected effect, is the edit distance, is the length of the behavior sequence, , , are weight parameters, and .

[0011] Preferably, the calculation method of the safety training efficiency score is as follows: Based on the collected training data, evaluate the training efficiency from two aspects: time and operation. The time score is calculated according to the ratio of the time required for the training personnel to complete the task to the standard time. The operation score is calculated according to the ratio of the number of operation mistakes of the training personnel to the total number of operations. Weighted sum the time score and the operation score to obtain the safety training efficiency score; Calculate the time score , the operation score , and calculate the score according to the formula , where is the time required for the training personnel to complete the task, is the standard time, is the number of operation mistakes, is the total number of operations, and are weight parameters, and .

[0012] Preferably, the emergency safety training evaluation score The calculation method is as follows: Substitute the calculated safety training standardization score, safety training adaptability score, and safety training efficiency score into the evaluation model, and perform weighted summation according to the weights to obtain the emergency safety training evaluation score; Substitute , , into the evaluation model to calculate the final score. Among them, , , are weight parameters, and .

[0013] An intelligent evaluation system for emergency safety training effects includes: Data acquisition module: Collect behavioral data of training personnel's action postures, time, and strength during training drills; Data processing and sample construction module: Process emergency safety rescue plan data, determine the standard values of standard actions, and construct a standard sample library and a model training library; Model construction and training module: Set a multi-layer neural network evaluation model, divide the sample subsets to train the model, and determine and output the model parameters; Evaluation calculation module: Train and correct the model, calculate the standardization, adaptability, and efficiency scores, and substitute them into the model to obtain the overall evaluation score.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. By standardizing and normalizing the behavior sequences in the emergency safety rescue plan, the present invention constructs a scientific and reasonable standard sample library, and at the same time designs an evaluation model with a multi-layer neural network structure, which can accurately extract data features. This improvement effectively solves the problems of inaccurate and incomplete evaluation results in the prior art, greatly improves the accuracy and comprehensiveness of emergency safety training effect evaluation, makes the evaluation results more valuable for reference, and helps training personnel and organizers to understand the training situation more accurately.

[0015] 2. Based on the accurate evaluation results, training personnel can clearly recognize their own advantages and disadvantages in terms of action standardization, adaptability, and training efficiency. For example, through the standardization score, training personnel can specifically improve non-standard actions; according to the adaptability score, they can strengthen the training of emergency handling capabilities. This enables training personnel to conduct more targeted training, improve the training effect, and further enhance the quality of emergency safety training, solving the problem in the prior art that accurate training guidance cannot be provided for training personnel.

[0016] 3. The comprehensive and accurate evaluation results of the present invention provide a strong basis for optimizing training programs. Organizers can adjust the training content, duration and methods according to the evaluation data of different trainees and reasonably allocate training resources. For example, the training duration can be increased for actions with generally poor standardization, and special training can be designed for links with weak adaptability. This helps to improve the utilization efficiency of training resources, solves the problems of lack of pertinence and waste of training resources in the existing technology, and makes emergency safety training more scientific and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the method flow of the present invention; Figure 2 Schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0018] To facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described with reference to the accompanying drawings.

[0019] Example 1, as Figure 1 As shown, the present invention provides an intelligent evaluation method for emergency safety training effects, comprising the following steps: S1. Basic data preparation: standardize and normalize the behavior sequences in the emergency safety and rescue plan to obtain standard action data, divide the training set and validation set to determine the standard action values, and mark the standard sample library; S2. Evaluation model construction: setting up the emergency safety training effect evaluation model, dividing the sample subsets, using the stochastic gradient descent method to train the model and determine the parameters; S3: Data processing and evaluation calculation, obtaining training and exercise behavior data, building a standard sample library and model training library, training and correcting the model, and calculating the safety training standardization score , strain score , efficiency score , substitute into the evaluation model to calculate the final evaluation score .

[0020] In an embodiment of the present invention, the method for standardizing and normalizing the behavior sequence in S1 is: S1.1. Analyze the collected emergency safety and rescue plans in detail, identify each action, and construct a behavior sequence. For each action, clarify its key characteristics (such as the start and end conditions, amplitude, speed, etc.) and quantify them. By removing the dimensional differences between different action data, standardize the behavior sequence. For the standardized emergency safety standard action data, use the minimum-maximum normalization method to map the data value range to the interval , making different action data comparable and avoiding the impact of data scale differences on subsequent model training; Let the behavior sequence in the emergency safety rescue plan be , and the standardized emergency safety specification action data be . Using the minimum-maximum normalization formula , we obtain the standardized action data with a value range of ; where represents the th behavior action in the behavior sequence, represents the th standardized emergency safety specification action data, represents the minimum value in the set , and represents the maximum value in the set ; S1.2. Divide the standardized action data into a training set and a validation set according to a pre-set ratio for effective validation and adjustment during model training, improving the model's generalization ability. Calculate the mean and variance of the action data for the training set and the validation set respectively. The mean reflects the average level of the data, and the variance reflects the degree of data dispersion. Combining the means and variances of both, use the weighted average method to determine the emergency safety specification action standard value, which serves as an important reference for subsequent evaluation of the training effect; Divide the standardized action data into a training set and a validation set , where is the pre-set division ratio, and its value range is ; Training set mean , training set variance ; Validation set mean , validation set variance ; Determine the emergency safety specification action standard value through the weighted average formula ; where is the number of data in the training set, is the th data in the training set, is the number of data in the validation set, is the th data in the validation set, is the weight coefficient, and its value range is ; S1.3. Randomly group the emergency safety standard action data into multiple sub - data sets. Each sub - data set has a certain degree of independence and representativeness to comprehensively cover different types of emergency safety standard actions. Further divide each sub - data set into a training set and a validation set, and calculate the mean deviation and variance deviation between the training set and the validation set. When both the mean deviation and the variance deviation are less than a pre - set threshold, it indicates that the consistency and stability of the training set and the validation set of this sub - data set are good, and mark it as the emergency safety standard action sample library for subsequent model training; Randomly group the emergency safety standard action data into sub - data sets , and each sub - data set is proportionally divided into a training set and a validation set ; Calculate the mean deviation , the variance deviation ; When and , mark this sub - data set as the emergency safety standard action sample library; Among them, is the number of sub - data sets, is the mean of the training set of the sub - data set , is the mean of the validation set of the sub - data set , is the variance of the training set of the sub - data set , is the variance of the validation set of the sub - data set , and are the set thresholds.

[0021] In the embodiment of the present invention, the setting method of the evaluation model in S2 is as follows: S2.1. Deeply study the specific requirements and objectives of the emergency safety rescue plan, comprehensively consider the action standardization, response ability, and training efficiency of emergency safety training, set up an emergency safety training effect evaluation model, design the evaluation model as a multi - layer neural network structure. The input layer receives the processed emergency safety training data, the hidden layer extracts data features through non - linear transformation, and the output layer outputs corresponding results according to the evaluation task. The hidden layer uses the ReLU activation function to alleviate the gradient disappearance problem and accelerate the model training speed. The output layer uses the Softmax function to convert the output into a probability distribution for convenient classification decision - making; The evaluation model is a multi - layer neural network structure, and the input layer has neurons, and the hidden layer has neurons, and the output layer has neurons, with the ReLU activation function , and the Softmax function ; Among them, is the output probability of the -th neuron in the output layer, is the input value of the -th neuron in the output layer, is the natural constant; S2.2. Subdivide the marked emergency safety specification action standard sample library in detail to form multiple sample subsets containing action samples and their corresponding labels. Each action sample represents a specific action in emergency safety training, and the label clarifies information such as whether the action is correct or its category, which helps improve the training efficiency and accuracy of the model and enables the model to better learn the characteristics of different types of actions; Divide the marked emergency safety specification action standard sample library into multiple sample subsets containing action samples and their corresponding labels , where each action sample is , and the corresponding label is ; Among them, is the number of sample subsets, represents the number of the sample subset, represents the number of the action sample in this sample subset; S2.3. Train the set emergency safety training evaluation model using the stochastic gradient descent method. Use the cross-entropy loss function to measure the difference between the model prediction result and the true label. The model continuously adjusts the parameters to minimize the loss function, and the parameter update step size is controlled by the learning rate. At the same time, use the validation samples in the sample subset to verify the trained model and calculate the model accuracy rate. When the model accuracy rate reaches the pre-set threshold, it indicates that the model has better learned the characteristics and rules of emergency safety specification actions, determine the model parameters and output the final evaluation model; Cross-entropy loss function , the parameter update formula is , calculate the model accuracy rate , when takes the value of 1, otherwise takes the value of 0. When reaches the threshold , determine the model parameters and output the model; Among them, is the number of training samples, are the parameters of the model, is the learning rate, is the gradient of the loss function with respect to the parameter , is the number of validation samples, is the predicted label of the model for the th validation sample, is the true label of the model for the th validation sample, is the indicator function.

[0022] In the embodiments of the present invention, in S3, the method for obtaining training drill behavior data, constructing a standard sample library and a model training library, and training and correcting the model is as follows: S3.1, during the emergency safety training drill process, use a variety of advanced data collection devices such as high-precision sensors, high-definition cameras, and professional recorders to obtain the behavior data information of the training personnel in real time and comprehensively. These data cover the action postures, time, and strength of the training personnel, and can accurately reflect their actual performance in the drill; Let the collected behavior data information be , where represents the th collected behavior data information; S3.2, convert the emergency safety rescue plan into a behavior sequence diagram. According to the plan action process and logical relationship, divide it into several task nodes containing unique behavior actions. Each task node represents a specific action or operation step in the emergency safety training, which is convenient for fine analysis of the training process. For the standard values of emergency safety standard actions, set action thresholds for each task node to determine whether the actions of the training personnel meet the specification requirements. According to the task nodes and action thresholds, construct a standard sample library and a model training library. The standard sample library provides correct reference examples, and the model training library stores actual training data to support model training and optimization; Convert the emergency safety rescue plan into a behavior sequence diagram and divide it into several task nodes , where represents the th task node. Combine the standard values of the standard actions to set action thresholds for each task node , and construct a standard sample library and a model training library accordingly; S3.3, initialize the emergency safety training evaluation model, input the data in the standard sample library into the model for training. To accelerate the model convergence speed and improve the training stability, use the batch normalization technique. According to the difference between the model output result and the standard sample library label, continuously adjust the model parameters, and repeat the training and correction process until the model accuracy meets the pre-set requirements. At this time, the model can better adapt to the actual situation of the emergency safety training and provide accurate prediction results for subsequent evaluation calculations; Initialize the model and adopt the batch normalization technique for the input data , the formula is , for the output data , the formula is , where is the normalized data, is the mean of the current batch of data, is the variance of the current batch of data, is a constant, and are learning parameters.

[0023] In the embodiment of the present invention, the calculation method of the safety training norm score is as follows: Divide the emergency safety training data to be evaluated into different action units or time periods. Through the trained model, analyze the training data in each action unit or time period in detail, calculate the corresponding standard action value and action execution value of the specification, introduce a weight function to measure the difference degree between the two, and sum the weight function values of each action unit or time period to obtain the safety training norm score. This score reflects the degree of action specification of the training personnel. The lower the score, the more compliant the action is with the specification requirements; After dividing the data to be evaluated, calculate the standard action value of the specification and , the weight function , ; where is a constant, is the fault tolerance coefficient, is the number of divided action units or time periods, is the th weight function value of the action unit or time period.

[0024] In the embodiment of the present invention, the calculation method of the safety training adaptability score is as follows: Deeply analyze the collected training data, and evaluate it from three aspects: behavior accuracy, behavior effectiveness, and behavior matching degree. The behavior accuracy is measured by calculating the cosine similarity between the actions of the training personnel and the standard actions. The behavior effectiveness is evaluated according to whether the actions achieve the expected effects. The behavior matching degree is measured by calculating the edit distance between the behavior sequences of the training personnel and the behavior sequences of the emergency safety rescue plan. Weighted sum the scores of these three aspects to obtain the safety training adaptability score, which reflects the ability of the training personnel to respond and handle problems in an emergency; Calculate the behavior accuracy , the behavior matching degree , and calculate the score according to the formula , where is the trainer's motion vector, is the standard action vector, the effectiveness of the behavior Evaluate whether the action achieves the expected effect. is the edit distance, is the length of the behavior sequence, 、 、 is the weight parameter, and .

[0025] In an embodiment of the present invention, the safety training efficiency score The calculation method is: Based on the collected training data, the training efficiency is evaluated from two aspects: time and operation. The time score is calculated based on the ratio of the time required for the trainee to complete the task to the standard time. The operation score is calculated based on the ratio of the number of operational errors of the trainee to the total number of operations. The time score and the operation score are weighted and summed to obtain the safety training efficiency score, which reflects the efficiency level of emergency safety training. Calculate time score , operation score , according to the formula Calculate the score, where is the time required for trained personnel to complete the task, It's standard time. is the number of operational errors, is the total number of operations, and is the weight parameter, and .

[0026] In an embodiment of the present invention, the emergency safety training evaluation score The calculation method is: Substitute the calculated safety training standardization score, safety training response score, and safety training efficiency score into the evaluation model and sum them up according to the weights to obtain the emergency safety training evaluation score. This score comprehensively reflects the overall effect of emergency safety training and provides a quantitative indicator for evaluating training quality. Will 、 、 Substitute into the evaluation model Calculate the final score, where 、 、 is the weight parameter, and .

[0027] Example 2, as Figure 2 As shown, an intelligent evaluation system for emergency safety training effects includes: Data acquisition module: Collect the behavioral data of the training personnel's movement postures, time, and strength during training and drills; Data processing and sample construction module: Process the emergency safety rescue plan data, determine the standard values of the standard actions, and construct the standard sample library and the model training library; Model construction and training module: Set up a multi-layer neural network evaluation model, divide the sample subsets to train the model, and determine and output the model parameters; Evaluation calculation module: Train and correct the model, calculate the scores of normativity, strain, and efficiency, and substitute them into the model to obtain the overall evaluation score.

[0028] The embodiments disclosed in the present invention are preferred embodiments, but not limited thereto. Those of ordinary skill in the art can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. However, as long as they do not depart from the spirit of the present invention, they are within the protection scope of the present invention.

Claims

1. An intelligent evaluation method for the effect of emergency safety training, characterized in that, It includes the following steps: S1. Basic data preparation: Standardize and normalize the behavior sequences in the emergency safety rescue plan to obtain standard action data, divide the training set and the validation set to determine the standard values of the standard actions, and mark the standard sample library; S2. Evaluation model construction: Set up an emergency safety training effect evaluation model, divide the sample subsets, and use the stochastic gradient descent method to train the model and determine the parameters; S3. Data processing and evaluation calculations, obtaining training and exercise behavior data, constructing a standard sample library and a model training library, training and correcting the model, calculating the safety training standardization score Response score Efficiency score , substituting into the evaluation model to calculate the final evaluation score .

2. The intelligent evaluation method for the effect of emergency safety training according to claim 1, wherein The method for standardizing and normalizing the behavior sequences in S1 is as follows: S1.

1. Analyze the collected emergency safety rescue plans in detail, identify each action in them, construct an action sequence, for each action, clarify its key features and quantify them. By removing the dimensional differences between different action data, standardize the action sequence. For the standardized emergency safety standard action data, use the min-max normalization method to map the data value range to the interval , making different action data comparable; S1.

2. Divide the standard action data into a training set and a validation set, and calculate the mean and variance of the action data in the training set and the validation set respectively. The mean reflects the average level of the data, and the variance reflects the degree of data dispersion. Combining the mean and variance of both, use the weighted average method to determine the standard value of the emergency safety specification actions ; S1.

3. Randomly group the emergency safety standard action data into multiple sub-datasets, further divide each sub-dataset into a training set and a validation set, calculate the mean deviation and variance deviation between the training set and the validation set. When both the mean deviation and the variance deviation are less than the pre-set threshold, mark it as the emergency safety standard action sample library.

3. An intelligent evaluation method for the effect of emergency safety training according to claim 1, characterized in that, The method for setting up the evaluation model in S2 is as follows: S2.

1. Design the evaluation model as a multi-layer neural network structure. The input layer receives the processed emergency safety training data, the hidden layer extracts data features through non-linear transformation, and the output layer outputs corresponding results according to the evaluation task. The ReLU activation function is used in the hidden layer, and the Softmax function is used in the output layer to convert the output into a probability distribution for classification decision-making; S2.

2. Make a detailed division of the marked emergency safety standard action sample library to form multiple sample subsets containing action samples and their corresponding labels. Each action sample represents a specific action in the emergency safety training, and the label clarifies information such as whether the action is correct or its category; S2.

3. Use the stochastic gradient descent method to train the set-up emergency safety training evaluation model. Use the cross-entropy loss function to measure the difference between the model prediction result and the true label. The model continuously adjusts the parameters to minimize the loss function, and the parameter update step size is controlled by the learning rate. At the same time, use the validation samples in the sample subsets to verify the trained model, calculate the model accuracy rate. When the model accuracy rate reaches the pre-set threshold, it indicates that the model has learned the emergency safety standard action features and rules well, determine the model parameters and output the final evaluation model.

4. An intelligent evaluation method for the effect of emergency safety training according to claim 1, characterized in that In S3, the method for obtaining the training drill behavior data, constructing the standard sample library and the model training library, and training and correcting the model is as follows: S3.

1. During the emergency safety training drill process, use data acquisition equipment to obtain the behavior data information of the training personnel in real time and comprehensively. These data cover the action postures, time, and strength of the training personnel, and can accurately reflect their actual performance in the drill; S3.

2. Convert the emergency safety rescue plan into a behavior sequence diagram. According to the action process and logical relationship of the plan, divide it into several task nodes containing unique behavior actions. Each task node represents a specific action or operation step in the emergency safety training. Set the action threshold for each task node to judge whether the actions of the training personnel meet the specification requirements. Based on the task nodes and the action thresholds, construct the standard sample library and the model training library. The standard sample library provides correct reference examples, and the model training library stores the actual training data; S3.

3. Initialize the emergency safety training evaluation model, input the data in the standard sample library into the model for training. To accelerate the model convergence speed and improve the training stability, the batch normalization technique is adopted. According to the difference between the model output result and the label in the standard sample library, continuously adjust the model parameters, and repeat the training and correction process.

5. The intelligent evaluation method for the effect of emergency safety training according to claim 4, wherein The calculation method of the safety training standardization score is as follows: Divide the emergency safety training data to be evaluated into different action units or time periods. Through the trained model, analyze the training data in each action unit or time period in detail, calculate the corresponding standard values of standard actions and action execution values, introduce a weight function to measure the difference degree between the two, and sum up the weight function values of each action unit or time period to obtain the safety training standardization score. After dividing the data to be evaluated, calculate the standard values of the standard actions and , the weight function , ; Among them, is a constant, is the fault tolerance coefficient, is the number of divided action units or time periods, is the weight function value of the th action unit or time period.

6. The intelligent evaluation method for the effect of emergency safety training according to claim 5, wherein The calculation method of the safety training score to be achieved is as follows: Deeply analyze the collected training data, and evaluate it from three aspects: behavior accuracy, behavior effectiveness, and behavior matching degree. The behavior accuracy is measured by calculating the cosine similarity between the actions of the training personnel and the standard actions. The behavior effectiveness is evaluated according to whether the action achieves the expected effect. The behavior matching degree is measured by calculating the edit distance between the behavior sequence of the training personnel and the behavior sequence of the emergency safety rescue plan. Weightedly sum up the scores of these three aspects to obtain the safety training adaptability score. Calculate the accuracy of behavior , the behavior matching degree , according to the formula calculate the score, where is the action vector of the trainer is the standard action vector, and the behavior effectiveness is evaluated according to whether the action achieves the expected effect is the edit distance is the length of the behavior sequence , , are weight parameters, and .

7. An intelligent evaluation method for the effect of emergency safety training according to claim 6, characterized in that The safety training efficiency score is calculated as follows: Based on the collected training data, evaluate the training efficiency from two aspects: time and operation. The time score is calculated according to the ratio of the time required for the training personnel to complete the task to the standard time. The operation score is calculated according to the ratio of the number of operation mistakes of the training personnel to the total number of operations. Weightedly sum up the time score and the operation score according to the weight to obtain the safety training efficiency score. Calculate the time score , operation score , and calculate the score according to the formula , where is the time required for the trainer to complete the task, is the standard time, is the number of operation errors, is the total number of operations, and are weight parameters, and .

8. An intelligent evaluation method for the effect of emergency safety training according to claim 7, characterized in that The calculation method of the emergency safety training assessment score is as follows: Substitute the calculated safety training standardization score, safety training adaptability score, and safety training efficiency score into the evaluation model, and weightedly sum them according to the weight to obtain the emergency safety training evaluation score. Substitute , , into the evaluation model to calculate the final score, where , , are weight parameters, and .

9. A system applied to an intelligent evaluation method for the effect of emergency safety training as described in claim 8, characterized in that, including: Data acquisition module: Collect the behavior data of the training personnel's action postures, time, and strength during the training drill. Data processing and sample construction module: Process the emergency safety rescue plan data, determine the standard values of standard actions, and construct the standard sample library and the model training library. Model construction and training module: Set the multi-layer neural network evaluation model, divide the sample subsets to train the model, and determine and output the model parameters. Evaluation calculation module: Train and correct the model, calculate the standardization, adaptability, and efficiency scores, and substitute them into the model to obtain the overall evaluation score.

Images