Intelligent assessment method for mine accident emergency risk avoiding

Abstract

The invention relates to the technical field of safety simulation training drilling, in particular to an intelligent assessment and evaluation method for mine accident emergency risk avoiding, which comprises the following steps of: S1, acquiring multi-dimensional data in an emergency risk avoiding capability drilling process of a student; s2, performing data preprocessing on the multi-dimensional data, and constructing an evaluation data set; s3, constructing an emergency risk avoiding capability evaluation system, wherein the system comprises evaluation items and distribution weights of the evaluation items; s4, establishing a weight evaluation model, and dynamically adjusting the distribution weight of the evaluation item in combination with the environmental data; and S5, scoring is completed according to the emergency risk avoiding capability evaluation system, and a personalized evaluation report is generated. According to the invention, comprehensive, objective and real-time evaluation of the emergency risk avoiding capability of the miner is realized, and the influence of human factors on the evaluation result can be reduced; the method can flexibly adapt to different environments and scenes through dynamic adjustment of the evaluation model, provides personalized evaluation reports and improvement suggestions, and effectively assists students in improving the emergency risk avoiding capability.

Description

Technical Field

[0001] The present invention relates to the technical field of safety simulation training and drills, and specifically to an intelligent assessment method for mine accident emergency refuge. Background Art

[0002] Mine accidents are characterized by suddenness and great harmfulness. The emergency refuge capabilities of miners and rescue teams have become key factors in ensuring life safety. Regularly carrying out emergency refuge training and drills can not only improve the reaction speed and self-help and mutual-help capabilities of miners in the face of sudden disasters, but also enhance the professional skills and coordinated combat levels of rescue teams, minimizing accident risks and casualties. During the training and drill process, by simulating real accident scenarios, the emergency refuge capabilities of trainees are evaluated and assessed, problems are discovered and improved, and it is ensured that trainees truly master emergency refuge skills.

[0003] Currently, the assessment methods for mine accident emergency refuge capabilities mostly stay at the levels of traditional theoretical examinations and on-site drills, relying on manual assessment. However, manual assessment has strong subjectivity, which may lead to inaccurate assessment results and cannot truly reflect the actual emergency refuge capabilities of miners. At the same time, the traditional manual assessment process usually includes multiple links such as on-site observation, recording performance, and analyzing data, which requires a large amount of time and human resources, resulting in low assessment efficiency. It may not be able to promptly discover the deficiencies of miners in emergency refuge capabilities, nor can it quickly adjust the training plan to meet actual needs. In addition, restricted by time and resources, manual assessment is difficult to cover all miners or all possible accident scenarios, cannot fully assess the emergency refuge capabilities of miners, and the lack of diverse scenario assessments also makes the assessment results incomplete.

[0004] Therefore, there is an urgent need for an intelligent and systematic assessment method for mine accident emergency refuge to achieve a comprehensive, objective, and real-time assessment of the emergency refuge capabilities of miners. Summary of the Invention

[0005] The present invention provides an intelligent assessment method for mine accident emergency refuge, which realizes a comprehensive, objective, and real-time assessment of the emergency refuge capabilities of miners. The assessment is based on intelligent algorithms to reduce the influence of human factors on the assessment results; by dynamically adjusting the assessment model, it can adapt to different environments and scenarios; it provides personalized assessment reports and improvement suggestions for drill personnel, which can effectively help trainees and managers improve their emergency refuge capabilities.

[0006] The present application provides the following technical solutions:

[0007] An intelligent assessment method for mine accident emergency refuge, comprising the following steps:

[0008] S1. Collect multi-dimensional data during the emergency avoidance ability drill of trainees;

[0009] S2. Conduct data preprocessing on the collected multi-dimensional data;

[0010] S3. Construct an emergency avoidance ability evaluation system, which is a tree structure and includes evaluation items and initial weights of the evaluation items;

[0011] S4. Establish a weight evaluation model based on machine learning algorithms, and dynamically optimize the initial weights of the evaluation items in combination with environmental data;

[0012] S5. Update the initial weights of the emergency avoidance ability evaluation system according to the results of S4, complete the assessment scoring, generate a personalized evaluation report and feedback the evaluation results to trainees and management personnel.

[0013] Furthermore, the S1 includes:

[0014] S11. Build an emergency avoidance drill environment based on VR technology;

[0015] S12. Trainees enter the emergency avoidance drill environment to complete the drill, and collect multi-dimensional data during the trainees' drill process.

[0016] Furthermore, the emergency avoidance drill environment is a mine three-dimensional model constructed based on VR technology. In the three-dimensional model, several three-dimensional roadways and through nodes are constructed according to various mine environment data collected on-site. The three-dimensional roadways include mine operation scene models and coal mine equipment and facility models; the emergency avoidance drill environment also includes a mine accident simulation function, which can simulate selected mine disaster scenarios in the selected three-dimensional roadways or nodes.

[0017] Furthermore, in the S12 step, trainees wear VR devices to enter the emergency avoidance drill environment and complete the emergency avoidance drill through the VR devices. The VR devices include VR glasses, VR handles and omnidirectional mobile platforms; the VR glasses provide visual input of the three-dimensional scene, and at the same time, the VR glasses communicate with the control background of the drill environment in real time and can receive the set disaster parameters in real time. The VR handles complete the positioning and action simulation of the trainees' hands, and the omnidirectional mobile platform provides real movement simulation of the trainees. Trainees also wear intelligent wearable devices for collecting physiological data.

[0018] Furthermore, multi-dimensional data is collected during the trainees' drill process. The multi-dimensional data includes action data, physiological data and environmental data.

[0019] The action data includes movement data and behavior data. Among them, the movement data includes the movement direction, movement speed, movement trajectory and movement posture of the trainees; the behavior data includes the action sequence of the trainees operating emergency equipment or wearing self-rescue devices.

[0020] The physiological data includes physiological indicators such as the heart rate, blood oxygen saturation, and body temperature of the trainee;

[0021] The environmental data is the environmental data set for the emergency evacuation drill environment;

[0022] The physiological data, action data of the trainee, and the environmental data set in the drill environment are stored in the control background according to a unified time stamp.

[0023] Furthermore, the action data of the trainee during the emergency evacuation process is collected through VR devices and cameras set in the drill environment; various physiological data of the trainee are collected in real time through intelligent wearable devices, and the physiological data, action data, and the environmental data set in the drill environment are stored in the control background of the drill environment according to a unified time stamp.

[0024] Furthermore, the intelligent wearable device is a smart watch or a smart bracelet.

[0025] Furthermore, the S2 includes the following steps:

[0026] S21. Remove the outliers in the data;

[0027] S22. Supplement the missing values in the data;

[0028] S23. Eliminate the noise existing in the data;

[0029] S24. Perform standardization processing on the multi-dimensional data.

[0030] Furthermore, in the S24, Z-score standardization is performed on physiological data such as heart rate and body temperature, and the physiological data is converted into a distribution with a mean of 0 and a standard deviation of 1. The specific process is as follows:

[0031]

[0032] where μ is the mean and σ is the standard deviation;

[0033] Min-Max normalization is performed on the environmental data, and the environmental data is linearly transformed into the range of [0, 1] or [-1, 1]. The specific process is as follows:

[0034]

[0035] where, X min is the minimum value, X max is the maximum value.

[0036] Furthermore, the emergency evacuation ability evaluation system is a tree-shaped hierarchical structure, including first-level evaluation indicators and second-level evaluation indicators.

[0037] Furthermore, the first-level evaluation indicators include the physiological state indicators, behavior normativity indicators, and environmental adaptability indicators of miners, where:

[0038] The physiological state indicators are used to evaluate the stability and endurance of miners' physical conditions during the emergency refuge process. The physiological state indicators can be decomposed into 3 secondary evaluation indicators, including heart rate stability, respiratory rate, and body temperature change;

[0039] The behavior normativity indicators are used to examine whether the operations of miners during the emergency refuge comply with safety regulations and standard procedures. The behavior normativity indicators can be decomposed into 3 secondary evaluation indicators, including refuge path selection, action completion time, and action normativity;

[0040] The environmental adaptability indicators are used to evaluate the adaptability of miners to complex mine environments and their ability to respond to sudden changes. The environmental adaptability indicators can be decomposed into 3 secondary evaluation indicators, including environmental perception ability, use of emergency equipment, and refuge route adjustment ability.

[0041] Furthermore, the emergency refuge ability evaluation system sets weights for each evaluation indicator, obtains the scores of the first-level evaluation indicators by weighted summation of the secondary evaluation indicators, and then obtains the final assessment score by weighted summation of the first-level evaluation indicators.

[0042] Furthermore, S4 includes the following steps:

[0043] S41. Construct an evaluation data set;

[0044] S42. Perform feature engineering on the evaluation data set to extract feature values from the original data;

[0045] S43. Construct an intelligent weight evaluation model;

[0046] S44. Train and validate the weight evaluation;

[0047] S45. Dynamically adjust and optimize the weight evaluation model to obtain optimized weights.

[0048] Furthermore, S41 includes the following steps:

[0049] S411. Classify the data sources;

[0050] S412. Develop data annotation rules according to the data classification and complete data annotation;

[0051] S413. Integrate multi-dimensional data and annotation information to establish an evaluation data set;

[0052] S414. Divide the evaluation data set into a training set, a validation set, and a test set.

[0053] Further, S42 includes:

[0054] Extract time-domain features and frequency-domain features from physiological data;

[0055] Extract statistical features from behavioral data, including frequency, duration, and trajectory features including path length and speed;

[0056] Extract statistical features from environmental data, including mean and variance;

[0057] Normalize or standardize the extracted feature values.

[0058] Further, the weight evaluation model includes a data input layer, a feature extraction layer, a base model layer, an integration layer, and an output layer, where:

[0059] The data input layer obtains input data from the evaluation dataset;

[0060] The feature extraction layer is used to extract hidden feature information of the input data, including a CNN model based on the convolutional neural network algorithm and an RNN model based on the recurrent neural network algorithm;

[0061] The base model layer respectively includes an SVM base model based on the support vector machine algorithm and an RF base model based on the random forest algorithm. The SVM base model is used to capture the non-linear relationship in the input data, and the RF base model improves the robustness and generalization ability of the model through the integration of multiple decision trees;

[0062] The integration layer is used to integrate the output results of multiple base models to generate a final prediction result;

[0063] The output layer outputs a weight evaluation result according to the calculated value of the integration layer.

[0064] Further, S45 includes the following steps:

[0065] S451. Dynamically update the online learning model;

[0066] S452. Model the emergency avoidance behavior of miners as a Markov decision process;

[0067] S453. Use reinforcement learning algorithms to optimize the avoidance strategy;

[0068] S454. Dynamically adjust the classification threshold and apply it to the online SVM or incremental random forest model;

[0069] S455. Further optimize the model parameters and strategies according to the environmental feedback.

[0070] The principle and advantages of the present invention are as follows:

[0071] A three-dimensional simulation exercise environment is established. Trainees complete the exercise in the three-dimensional simulation exercise environment through VR devices. Physiological data, heart rate data, and environmental data of the trainees are collected through VR devices and intelligent devices worn by the trainees. The multi-dimensional data is preprocessed and fused to establish an evaluation data set; an intelligent evaluation model and an evaluation index system for the risk avoidance ability are established, and the evaluation data set is analyzed to obtain the evaluation result of the risk avoidance ability and a personalized improvement report.

[0072] The technical solution of the present invention can comprehensively evaluate the emergency risk avoidance ability of miners by integrating multi-dimensional data; the exercise scenarios comprehensively cover different types of emergencies, which can ensure the comprehensiveness of the evaluation results and are applicable not only to single scenarios but also can be generalized to more emergency scenarios;

[0073] Automatically identify key features through intelligent algorithms to avoid subjective biases that may be introduced by manual intervention. The evaluation results are accurate and reliable, and the evaluation rules and scoring criteria are strictly consistent. The data-driven evaluation method can ensure the consistency and traceability of the evaluation results and ensure the credibility of the evaluation process;

[0074] The evaluation model is updated in real time through an online learning mechanism. Combined with the real-time update of environmental data, the evaluation model can quickly adapt to new data and new scenarios, with real-time performance: dynamically adjust the evaluation model to adapt to different environments and scenarios.

[0075] The present invention can provide detailed evaluation reports and improvement suggestions, which can help trainees improve their emergency risk avoidance ability. At the same time, it can provide a data analysis tool for management to help it comprehensively understand the real data of miners' emergency risk avoidance ability and assist it in formulating more practical emergency plans. Brief Description of the Drawings

[0076] Figure 1 It is a flowchart of an intelligent assessment method for emergency risk avoidance in mine accidents. Detailed Embodiments

[0077] The following is a further detailed description through specific embodiments:

[0078] Embodiment 1

[0079] This embodiment provides an intelligent assessment method for emergency risk avoidance in mine accidents, as Figure 1 shown, including the following steps:

[0080] S1. Collect multi-dimensional data during the exercise of the trainees' emergency risk avoidance ability, specifically including:

[0081] S11. Build an emergency risk avoidance exercise environment based on VR technology;

[0082] The emergency refuge drill environment is a 3D mine model constructed based on VR technology. In the 3D model, several 3D roadways and through nodes are constructed according to various mine environment data collected on-site. The 3D roadways contain different mine operation scene models, such as excavation and working faces, transportation roadways, ventilation roadways, refuge chambers, etc. In the mine operation scene models, corresponding coal mine equipment and facilities models are provided, including various coal mine operation equipment and self-rescue equipment.

[0083] The emergency refuge drill environment also includes a mine accident simulation function, which can simulate selected mine disaster scenarios in the selected 3D roadway or node, including mine fires, mine water hazards, gas disasters, roof accidents, dust disasters, heat hazards, etc. For different mine disaster scenarios, the disaster parameter settings can be flexibly adjusted. For example, when simulating a gas disaster scenario, different gas concentrations, oxygen concentrations, carbon monoxide concentrations, environmental temperatures, environmental humidities, etc. can be set; when simulating a mine water hazard scenario, different water inrush points, water inrush directions, water inrush speeds, water accumulation depths, etc. can be set.

[0084] S12. The trainees enter the emergency refuge drill environment to complete the drill, and collect multi-dimensional data during the trainees' drill process;

[0085] The trainees wear VR devices to enter the emergency refuge drill environment and complete the emergency refuge drill through the VR devices. The VR devices include VR glasses, VR handles, and omnidirectional movement platforms. The VR glasses provide visual input of the 3D scene, and at the same time, the VR glasses communicate with the control background of the drill environment in real time and can receive the set disaster parameters in real time. The VR handles complete the positioning and movement simulation of the trainees' hands, and the omnidirectional movement platform provides real movement simulation for the trainees. The trainees also wear intelligent wearable devices that can collect physiological data.

[0086] During the trainees' drill process, multi-dimensional data is collected, including:

[0087] The action data of the trainees during the emergency refuge process is collected through the VR devices and the cameras set in the drill environment. The action data includes movement data and behavior data. Among them, the movement data includes the movement direction, movement speed, movement trajectory, and movement posture of the trainees; the behavior data includes the action sequence of the trainees operating emergency equipment or wearing self-rescue devices.

[0088] A variety of physiological data of the trainees is collected in real time through the intelligent wearable devices. The intelligent wearable devices can be selected as intelligent watches or intelligent bracelets that can monitor physiological indicators such as the trainees' heart rate, blood oxygen saturation, and body temperature in real time. In this embodiment, preferably, a Huawei intelligent watch (Watch 4) is selected. The control background of the drill environment establishes communication with the intelligent watch and receives and stores the trainees' physiological data in real time.

[0089] The emergency evacuation drill environment sets and updates environmental data according to the preset environmental conditions and the actions of the trainees. For example, the drill environment presets a gas disaster scenario where roadway A is filled with 3% high-concentration gas for 10 m, and at the same time, warning lights and self-rescuer models are set in the scenario. When the trainees enter roadway A, they receive three-dimensional scene and gas concentration data information including self-rescuer and warning light models through the VR glasses they wear. If the trainees have completed the action of wearing the self-rescuer before entering roadway A, the oxygen reserve and gas concentration information are refreshed regularly according to the action time and route of the trainees; if the trainees have not completed the action of wearing the self-rescuer before entering roadway A, the oxygen reserve information is displayed when the trainees come into contact with the self-rescuer model, and the oxygen reserve display and gas concentration information are refreshed regularly after the trainees complete the correct action of wearing the self-rescuer.

[0090] The physiological data, action data of the trainees and the environmental data set in the drill environment are stored in the control background according to a unified timestamp.

[0091] S2. Perform data preprocessing on the collected multi-dimensional data.

[0092] Perform data cleaning, denoising and standardization on the collected multi-dimensional data to ensure the accuracy and consistency of the data, including the following steps:

[0093] S21. Remove outliers from the data;

[0094] During the drill, it is possible to generate some abnormal data due to unstable sensor operation or communication failures, and these abnormal data need to be processed. For example, the heart rate data of the trainees is 0 or significantly exceeds the normal range (such as <30 bpm or >200 bpm). In this embodiment, the threshold filtering method is used to delete the abnormal data.

[0095] At the same time, according to the device working status information, the collected data within the interval where the device working status information is abnormal is excluded.

[0096] S22. Supplement the missing values in the data:

[0097] After completing the operation of removing outliers, there may be some missing data points. Interpolation method, mean filling and other methods are used to fill in the missing values to ensure the integrity of the data.

[0098] S23. Eliminate the noise existing in the data;

[0099] During the exercise process, there are various noise sources, including environmental noise (such as electromagnetic interference, vibration, etc.), equipment noise (insufficient sensor accuracy), human noise (such as improper sensor wearing, data transmission errors, etc.), and biological noise (such as signal interference caused by muscle activity, respiratory movement), resulting in interference and distortion in some of the collected physiological data, which affects subsequent analysis. Therefore, it is necessary to denoise the collected data.

[0100] During the data denoising process, for slowly changing physiological data (such as heart rate, body temperature, etc.), using moving window averaging and smoothing filtering are common methods.

[0101] Using moving window averaging can smooth short-term fluctuations, and its operation steps include:

[0102] Select a suitable window length according to the data characteristics (for example, for heart rate data, a time range of several seconds to several minutes can be selected). The window length selection is based on a balanced consideration of smoothing short-term fluctuations and retaining sufficient details.

[0103] For each time point, take the data points within a certain range before and after it according to the window length, and calculate the average value of these points as the new value of this time point;

[0104] Gradually move the window backward step by step according to the preset step size and repeat the above calculation until all physiological data is covered.

[0105] At the same time, use smoothing filtering on physiological data to remove some frequency components to facilitate subsequent physiological data analysis. For example, heart rate variability (HRV) analysis is an important method to evaluate the function of the autonomic nervous system. By analyzing the changes in the RR interval, the activity states of the sympathetic and parasympathetic nerves can be reflected. In high-pressure or high-intensity environments such as mine operations or emergency shelter, the low-frequency component of 0.04 Hz - 0.15 Hz reflects sympathetic nerve activity and can be used to evaluate the psychological stress level or physical fatigue degree of miners, while the high-frequency component of 0.15 Hz - 0.4 Hz reflects parasympathetic nerve activity and is usually closely related to physiological phenomena such as the relaxation state. Moreover, the high-frequency component may be greatly interfered by external factors. Therefore, the low-frequency component is more significant for the heart rate variability analysis of emergency shelter.

[0106] By selecting a suitable type of low-pass filter (such as Butterworth filter, Chebyshev filter, etc.), selecting a suitable cut-off frequency according to the characteristics of the physiological signal, and applying the designed low-pass filter to the original physiological data, the denoised signal can be obtained.

[0107] S24. Standardize the multi-dimensional data.

[0108] Multidimensional data has different dimensions and ranges. It is necessary to convert various data types into a unified standard range for subsequent analysis and modeling. Different data types require different standardization methods, including:

[0109] Perform Z-score standardization on physiological data such as heart rate and body temperature, and convert the physiological data into a distribution with a mean of 0 and a standard deviation of 1. The specific process is as follows:

[0110]

[0111] Among them, μ is the mean and σ is the standard deviation.

[0112] Perform Min-Max normalization on environmental data, and linearly transform the environmental data into the range of [0,1] or [-1,1]. The specific process is as follows:

[0113]

[0114] Among them, X min is the minimum value, and X max is the maximum value.

[0115] S3. Construct an emergency shelter ability evaluation system:

[0116] The constructed emergency shelter ability evaluation system is a tree-like hierarchical structure, including first-level evaluation indicators and second-level evaluation indicators. The first-level evaluation indicators include miners' physiological state indicators, behavior normativity indicators, and environmental adaptability indicators. Specifically, it is shown in Table 1.

[0117] Table 1

[0118]

[0119] The physiological state indicators are used to evaluate the stability and tolerance of miners' physical conditions during the emergency shelter process, judge whether the physical reactions of miners in high-pressure or dangerous environments are normal, and ensure that they have the physical conditions to cope with emergencies. The physiological state indicators include 3 second-level evaluation indicators, namely heart rate stability, respiratory rate, and body temperature change.

[0120] Heart rate stability: Set the normal change range of heart rate (such as 60-100 bpm), count the proportion of data points within the normal range, and obtain the evaluation score of this item through percentage conversion. The score range of this item is 0-100;

[0121] Respiratory rate: Set the normal change range of respiratory rate (such as 12-20 times per minute), count the proportion of data points within the normal range, and obtain the evaluation score of this item through percentage conversion. The score range of this item is 0-100;

[0122] Body temperature change: Set the normal change range of body temperature (such as 36 - 37.5 °C), and comprehensively judge the stability of body temperature change by combining the standard deviation and the maximum fluctuation range. The score range for this item is 0 - 100.

[0123] The behavior norm index is used to test whether the operations of miners during emergency shelter are in line with safety regulations and standard procedures, including behaviors such as correctly using rescue equipment, following the escape route, and implementing the emergency plan, to ensure that miners can take scientific and reasonable actions in case of accidents and reduce secondary injuries. The behavior norm index includes 3 secondary evaluation indicators, namely, the selection of the shelter path, the completion time of actions, and the norm of actions.

[0124] Selection of the shelter path: Examine whether miners choose the shortest and safest shelter path. If the selected path is reasonable, the full score can be obtained, and corresponding deductions will be made for each deviation from the reasonable path. Finally, the total score is calculated as the score for this item. The score range for this item is 0 - 100 points.

[0125] Completion time of actions: Examine the time for miners to complete emergency shelter actions (such as wearing a self-rescuer, using emergency equipment). According to relevant regulations on emergency management, wearing a self-rescuer needs to be completed within 30 s. If the time for a miner to wear a self-rescuer exceeds 30 s, the score for this item is 0 points; if the time is within 20 - 30 s, the score for this item is 60 points; if the time is within 10 - 20 s, the score for this item is 80 points; if the time for wearing a self-rescuer is less than 10 s, the score for this item is 100 points.

[0126] Norm of actions: Examine whether the actions of miners during emergency shelter are in line with the standard procedures (such as equipment operation, escape posture). The score is given by analyzing the video recording data or the time data of relevant sensors. If the actions are standard, the full score is obtained. The score range for this item is 0 - 100.

[0127] The environmental adaptability index is used to evaluate the adaptability of miners to complex mine environments and their ability to respond to sudden changes. For example, in a dark, toxic gas, high-temperature or collapsed environment, whether miners can quickly adjust their states, accurately judge the situation and make effective decisions to ensure their own and others' safety. The environmental adaptability index includes 3 secondary evaluation indicators, namely, environmental perception ability, use of emergency equipment, and the ability to adjust the shelter route.

[0128] Environmental perception ability: Evaluate the reaction speed of trainees to environmental changes such as gas concentration, temperature, and oxygen content. The score is evaluated by calculating the reaction time from when trainees obtain environmental change information to when they take actions. The shorter the reaction time, the higher the score, and deductions are made if the reaction time exceeds the threshold. The score range for this item is 0 - 100 points.

[0129] The use of emergency equipment is used to evaluate whether trainees have mastered the correct usage methods of emergency equipment (such as self-rescuers, respirators). If the equipment is used correctly, it gets a full score. Errors are deducted item by item, and the score range for this item is 0 - 100 points.

[0130] The adjustment of the evacuation route is used to evaluate the trainees' ability to dynamically adjust the evacuation route according to environmental changes. The evacuation path is divided into several segments according to preset nodes, and the change data of environmental conditions are dynamically set at each node. Trainees choose the subsequent evacuation path based on the obtained environmental information. After separately integrating the scores of each segment of the path and then adding them up, the score for this item is obtained. If the selection of a certain segment of the path is incorrect, the score for that segment of the path will be deducted. The score range for this item is 0 - 100 points.

[0131] Initial weights are set for each evaluation index as shown in Table 1. The scores of the corresponding first-level evaluation indexes are obtained by weighted summation of the secondary evaluation indexes, and then the weighted summation of the first-level evaluation indexes is performed to obtain the assessment score.

[0132] S4. Establish an intelligent weight evaluation model based on machine learning algorithms, and dynamically optimize the initial weights of the evaluation items in combination with environmental data to ensure the adaptability of the evaluation results.

[0133] In order to ensure that the established emergency evacuation ability evaluation system can be widely applicable to diverse mine disaster scenarios, an intelligent weight evaluation model is established to dynamically adjust the initial weights.

[0134] S41. Construct an evaluation data set:

[0135] In this step, an evaluation data set is constructed based on the preprocessed multi-dimensional data, specifically including:

[0136] S411. Classify the data sources.

[0137] The classification of data sources includes physiological data (heart rate, body temperature, respiratory rate, etc.), behavioral data (movement trajectory, action standardization, evacuation path selection, etc.), environmental data (gas concentration, temperature, oxygen content, etc.), and historical data (records of evacuation behaviors in past drills or actual accidents).

[0138] S412. Develop data annotation rules according to the data classification and complete data annotation:

[0139] Data of different classifications have different data annotation rules. For example, when annotating gas concentration, according to the safety standard of gas concentration, the data are divided into the following three categories:

[0140] Low risk: Gas concentration ≤ 1.0% CH4;

[0141] Medium risk: 1.0% CH4 < gas concentration ≤ 3.0% CH4;

[0142] High risk: gas concentration > 3.0% CH4;

[0143] Data annotation is completed using expert evaluation, rule judgment, or semi-automatic methods, including behavior categories (correct risk avoidance, incorrect risk avoidance, hesitation, etc.) and risk levels (low risk, medium risk, high risk).

[0144] S413. Integrate multi-dimensional data and annotation information to establish an evaluation data set.

[0145] S414. Divide the evaluation data set into a training set, a validation set, and a test set. In this embodiment, 70% is divided into the training set, 15% is divided into the validation set, and 15% is divided into the test set to ensure balanced data distribution.

[0146] S42. Perform feature engineering on the evaluation data set to extract feature values from the original data:

[0147] The purpose of this step is to extract meaningful feature values from the original data so that the model can better understand and predict the target variable.

[0148] Select corresponding feature extraction methods according to the respective characteristics of physiological data, behavioral data, and environmental data.

[0149] Physiological data usually has time series characteristics or frequency domain characteristics. Therefore, features need to be extracted from both the time domain and the frequency domain. Mean and variance calculation methods can be used to extract time domain features, and Fourier transform or wavelet transform methods can be used to extract frequency domain features.

[0150] For behavioral data, mainly extract statistical features (such as frequency, duration) and trajectory features (such as path length, speed);

[0151] For environmental data, mainly extract statistical features (such as mean, variance).

[0152] Normalize (such as Z-score normalization) or standardize (such as Min-Max normalization) the extracted feature values (time domain features, frequency domain features, behavioral features, and environmental features) to ensure consistent dimensions of different features.

[0153] S43. Construct an intelligent weight evaluation model:

[0154] An ensemble learning idea is used to construct the weight evaluation model, which combines the advantages of support vector machines and random forest algorithms, can comprehensively consider the characteristics of multi-dimensional data, and provide accurate and reliable evaluation weight prediction results.

[0155] The weight evaluation model includes a data input layer, a feature extraction layer, a base model layer, an integration layer, and an output layer, where:

[0156] The data input layer obtains input data from the evaluation dataset, including physiological data, behavioral data, and environmental data. The input data undergoes standardized preprocessing to ensure the stability of model training.

[0157] The feature extraction layer is constructed based on deep learning models, including a CNN model based on the convolutional neural network algorithm and an RNN model based on the recurrent neural network algorithm, which extracts hidden feature information from the input data. The structure of the CNN model includes a convolutional layer, a pooling layer, and a fully connected layer, which is used to process data with spatial structures and extract high-level feature representations from the input data for output. The RNN model includes an input layer that receives sequence data, a hidden layer with a recurrent connection structure, and an output layer that outputs a feature vector. Through the recurrent connection of the hidden layer, the information from the previous time step is passed to the current time step, thereby capturing the dependencies in the time series and generating output information at the last time step, that is, a feature vector containing temporal information.

[0158] The base model layer contains an SVM base model based on the support vector machine algorithm and an RF base model based on the random forest algorithm respectively. The SVM base model is used to capture the non-linear relationships in the input data, and its kernel function can be selected as a linear kernel or an RBF kernel. The corresponding hyperparameters include the regularization coefficient and the kernel function parameters. The RF base model improves the robustness and generalization ability of the model through the integration of multiple decision trees. The corresponding hyperparameters include the number of trees, the maximum depth, the minimum number of samples to split a tree, etc. The feature data is input into the SVM base model and the RF base model respectively, and the output result is the predicted value.

[0159] The integration layer is used to integrate the output results of multiple base models to generate the final prediction result, which can effectively improve the stability and generalization ability of the model. The integration layer can adopt a stacked integration or a voting integration structure. In the stacked integration structure, the output results of the SVM base model and the RF base model are used as new features and input into the meta-model. The meta-model can adopt a logistic regression model or a gradient boosting decision tree model, and the output result is the final predicted value. In the voting integration structure, a vote is taken on the output results of the SVM base model and the RF base model, and the final predicted value is obtained by weighting according to the voting results. The integration layer in this embodiment adopts a voting integration structure.

[0160] The output layer outputs the weight evaluation result according to the calculated value of the integration layer.

[0161] S44. Train and validate the weight evaluation model to obtain the optimized weights:

[0162] Use the training set data to train the model, use the mean squared error or cross entropy as the loss function, and adjust the model parameters through backpropagation;

[0163] Regularly perform hyperparameter tuning (such as learning rate, regularization coefficient) using the validation set, and stop training when the performance of the validation set no longer improves to prevent overfitting;

[0164] Use the test set to evaluate the performance of the trained model to ensure its generalization ability. Evaluation metrics include accuracy, precision, recall, and F1 score, etc.

[0165] S45. Dynamically adjust and optimize the final evaluation model, including the following steps:

[0166] Adopt online learning algorithms (online SVM, incremental random forest) to dynamically update the model. At the same time, model the emergency avoidance behavior of miners as a Markov decision process (MDP), and use reinforcement learning algorithms (such as Q-learning, deep Q-network) to optimize the avoidance strategy, and dynamically adjust the model according to environmental feedback (such as whether the avoidance is successful).

[0167] S451. Dynamically update the online learning model:

[0168] The purpose of this step is to update the weight evaluation model in real time to ensure that it can adapt to the changing scenario data distribution. During the dynamic update process, the online SVM dynamically adjusts the support vectors and hyperplane by processing new samples one by one. When new data arrives, only the impact of the current sample on the model needs to be calculated, rather than retraining the entire model. The incremental random forest adapts to new data by gradually expanding the tree structure or updating the node statistics of the existing tree. The new data is used to adjust the splitting rules or weights of each tree.

[0169] Regularly evaluate the model performance on the validation set to ensure its accuracy and stability.

[0170] S452. Model the emergency avoidance behavior of miners as a Markov decision process:

[0171] Based on the Markov decision process (MDP) modeling, formalize the emergency avoidance behavior of miners as an MDP problem, providing a basis for reinforcement learning. The Markov decision process is a mathematical model used for decision-making in a stochastic environment, which can be used to describe how an agent reaches the optimal solution through a series of decisions in a certain environment. In the MDP, the agent takes an action at each time step, transfers from the current state to the next state, and obtains the corresponding reward. The key feature of the MDP is that the future state depends only on the current state and action, and is independent of the past state and action, which is very suitable for describing the decision-making and action process of miners in emergency avoidance.

[0172] The core components of the MDP problem include:

[0173] State(s): Describe the miner's current location, physiological state (such as heart rate, body temperature), environmental conditions (such as gas concentration, temperature), etc.;

[0174] Action(a): Describe the risk avoidance behaviors that miners can take, such as choosing an escape route, wearing a self-rescuer, using a fire extinguisher, etc.;

[0175] Reward(r): Define the reward function according to the result of the risk avoidance behavior. For example, a positive reward is obtained for successful escape, and a negative reward is obtained for incorrect operation or failure to avoid risks in time.

[0176] State transition probability (P): Describe the probability of transitioning from one state to another, which is affected by the miner's actions and environmental conditions.

[0177] S453. Optimize the risk avoidance strategy using reinforcement learning algorithms:

[0178] The goal of solving the MDP is to find a policy π that selects the optimal action a in each state s, so as to maximize the expected cumulative reward obtained in the long term. Common methods include value iteration, policy iteration, and Q-learning, etc.

[0179] In this embodiment, the optimal risk avoidance strategy is learned through reinforcement learning algorithms (Q-learning or deep Q-network DQN) to maximize the long-term cumulative reward. The steps are as follows:

[0180] 1) Initialize the Q-table or neural network:

[0181] If using Q-learning, initialize the Q-table to store the value of each state-action pair.

[0182] If using DQN, initialize the neural network as an approximator of the Q-value function.

[0183] The Q-value function represents the expected return that can be obtained after performing action a in state s:

[0184]

[0185] where r t is the reward at the t-th step;

[0186] γ is the discount factor, which is used to control the weight of future rewards.

[0187] 2) Exploration and exploitation:

[0188] In the initial stage, the miner tries different actions through exploration and gradually accumulates experience.

[0189] As the experience increases, more use is made of the learned knowledge to select the optimal action.

[0190] 3) Update the Q-value or network parameters:

[0191] For Q - learning: Update the Q - value according to the Bellman expectation equation. The formula is:

[0192]

[0193] where Q(s,a) is the expected return, R(s,a) is the immediate return obtained after executing action a in state s, γ is the discount factor, with a value in the range of 0 - 1, and max a′ Q(s′,a′) represents the best future return starting from state s′.

[0194] For DQN: Store experiences in a replay buffer and update network parameters using stochastic gradient descent.

[0195] 4) Generate the optimal policy:

[0196] Generate the optimal risk - avoidance policy based on the learned Q - value to guide the miner to take the best actions in different states.

[0197] S454. Dynamically adjust the classification threshold and apply it to the online SVM or incremental random forest model:

[0198] Obtain environmental parameters such as gas concentration and temperature in real - time, and dynamically adjust the classification threshold according to predefined threshold adjustment rules to adapt to different scenario requirements.

[0199] For example, the threshold adjustment rules stipulate that:

[0200] When the gas concentration is high, lower the classification threshold for the "high - risk" category to improve the early - warning sensitivity.

[0201] When the temperature is low, appropriately relax the classification criteria for the "low - risk" category.

[0202] Apply the adjusted threshold to the online SVM or incremental random forest model to ensure that the classification results match the actual environment.

[0203] S455. Further optimize the model parameters and policies according to the environmental feedback:

[0204] Use the environmental feedback to adjust the model parameters to better predict future rewards;

[0205] Adopt a greedy policy or ε - greedy policy to select actions, making it tend to choose actions with higher rewards;

[0206] Repeat the iterative optimization process of "executing actions - observing feedback - updating the model and policy", and stop the optimization when the policy tends to be stable.

[0207] After the execution of step S455 is completed, the final optimized weights are obtained.

[0208] S5. Update the initial weights of the emergency avoidance ability assessment system, complete the assessment scoring, generate a personalized assessment report, and feedback the assessment results to the trainees and management personnel, including:

[0209] Update the initial weights in the emergency avoidance ability assessment system to optimized weights according to S4.

[0210] Calculate the scores of each assessment index and the final assessment score according to the emergency avoidance ability assessment system, and generate a personalized assessment report. The content of the assessment report includes the miner's performance in different dimensions (such as physiological state, behavior standardization, environmental adaptability), the final assessment score, weak link analysis, and improvement suggestions. Visual icons are provided in the assessment report to intuitively display the assessment results of the miner's avoidance ability.

[0211] Define the assessment levels according to the score range of the final assessment score. Among them, 90 - 100 points is excellent; 80 - 89 points is good; 70 - 79 points is qualified; 60 - 69 points requires improvement; 0 - 59 points is unqualified. Trainees with unqualified assessment results need to retake training and drills.

[0212] Provide personalized improvement suggestions based on the miner's weak links. For example, for miners who are not proficient in using equipment, it is recommended to strengthen the study of equipment operation procedures; for miners who perform poorly in choosing escape routes, it is recommended to strengthen map memory training; for miners with weak psychological qualities, it is recommended to participate in psychological counseling courses.

[0213] Based on the assessment results, help management personnel optimize the training plan, focusing on solving common problems. For example, if most miners have deficiencies in equipment operation, the class hours of relevant training courses can be increased.

[0214] Regularly re - assess the miners to track the changes in their emergency avoidance abilities; provide continuous guidance according to the re - assessment results to help miners continuously improve their abilities.

[0215] At the same time, the assessment report can also provide data analysis tools for management personnel to assist them in better formulating emergency plans.

[0216] The above are only embodiments of the present invention. The invention is not limited to the fields involved in this embodiment case, and common knowledge such as specific structures and characteristics in the solution is not described in detail here. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application should be subject to the content of its claims, and the specific implementation manners in the specification can be used to interpret the content of the claims.

Claims

1. An intelligent assessment method for emergency risk avoidance of mine accidents, characterized by: The following steps are involved: S1. Collect multi-dimensional data during the students' emergency risk avoidance ability drill; S2, performing data preprocessing on the collected multidimensional data; S3. Construct an emergency risk avoidance capability evaluation system, wherein the emergency risk avoidance capability evaluation system is a tree structure, including evaluation items and initial weights of the evaluation items; S4. Establish a weight evaluation model based on machine learning algorithms and dynamically optimize the initial weights of evaluation items in combination with environmental data; S5. Update the initial weight of the emergency risk avoidance capability assessment system based on the results of S4, complete the assessment and scoring, generate a personalized assessment report and feed back the assessment results to trainees and managers.

2. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 1 is characterized by: The S1 includes: S11. Build an emergency risk avoidance drill environment based on VR technology; S12. Trainees enter the emergency evacuation drill environment to complete the drill, and collect multi-dimensional data during the trainees' drill.

3. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 2 is characterized by: The emergency evacuation drill environment is a three-dimensional mine model constructed based on VR technology. In the three-dimensional model, a number of three-dimensional tunnels and through nodes are constructed according to a variety of mine environment data collected on the spot. The three-dimensional tunnel contains a mine operation scene model and a coal mine equipment and facility model; the emergency evacuation drill environment also includes a mine accident simulation function, which can simulate the selected mine disaster scene in the selected three-dimensional tunnel or node.

4. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 2 is characterized by: In the step S12, the trainee wears VR equipment to enter the emergency evacuation drill environment and completes the emergency evacuation drill through the VR equipment. The VR equipment includes VR glasses, VR handles and a universal action platform. The VR glasses provide visual input of the three-dimensional scene. At the same time, the VR glasses communicate with the control background of the drill environment in real time and can receive the disaster parameters set by the control background in real time; the VR handle completes the trainee's hand positioning and movement simulation, and the universal action platform provides the trainee with real movement simulation; the trainee also wears smart wearable devices that can collect physiological data.

5. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 2 is characterized by: The multidimensional data includes action data, physiological data and environmental data, wherein: The action data includes movement data and behavior data, wherein the movement data includes the trainee's movement direction, movement speed, movement trajectory, and movement posture; the behavior data includes the trainee's action sequence of operating emergency equipment or wearing a self-rescuer; The physiological data include the trainee's heart rate, blood oxygen saturation, and body temperature; The environmental data is the environmental data set for the emergency risk avoidance drill environment; The trainees' physiological data, action data and environmental data of the exercise environment settings are stored in the control background according to a unified timestamp.

6. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 5 is characterized by: The action data of trainees during emergency avoidance are collected through VR equipment and cameras set up in the drill environment. Various physiological data of trainees are collected in real time through smart wearable devices. The physiological data, action data and environmental data set in the drill environment are stored in the control background of the drill environment according to a unified timestamp.

7. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 6 is characterized by: The smart wearable device is a smart watch or a smart bracelet.

8. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 1 is characterized by: The S2 comprises the following steps: S21, remove outliers in the data; S22, missing values ​​in supplementary data; S23, eliminating noise in the data; S24. Standardize the multidimensional data.

9. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 8 is characterized by: In S24, the physiological data such as heart rate and body temperature are Z-score standardized to convert the physiological data into a distribution with a mean of 0 and a standard deviation of 1. The specific process is as follows: Among them, μ is the mean and σ is the standard deviation; Min-Max normalization is performed on the environmental data, and the environmental data is linearly transformed to the range of [0,1] or [-1,1]. The specific process is as follows: Among them, X min is the minimum value, X max is the maximum value.

10. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 1 is characterized by: The emergency risk avoidance capability evaluation system is a tree-like hierarchical structure, including primary evaluation indicators and secondary evaluation indicators.

11. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 10 is characterized in that: The first-level evaluation indicators include miners' physiological status indicators, behavioral norms indicators and environmental adaptability indicators, among which: The physiological status index is used to evaluate the stability and tolerance of the miners' physical condition during emergency avoidance, including three secondary evaluation indicators, namely heart rate stability, respiratory rate and body temperature changes; The behavioral normative index is used to check whether the miners' operations in emergency avoidance comply with safety regulations and standard procedures, including three secondary evaluation indicators, namely, avoidance path selection, action completion time, and action normativeness; The environmental adaptability index is used to evaluate miners' ability to adapt to complex mining environments and cope with sudden changes, and includes three secondary evaluation indicators, namely environmental perception ability, use of emergency equipment and ability to adjust avoidance routes.

12. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 10 is characterized in that: The emergency risk avoidance capability assessment system sets weights for each assessment indicator, obtains the first-level assessment indicator score by weighted summation of the second-level assessment indicators, and then obtains the final assessment score by weighted summation of the first-level assessment indicators.

13. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 1 is characterized in that: The S4 comprises the following steps: S41, build evaluation data set; S42, performing feature engineering on the evaluation data set to extract feature values ​​from the original data; S43. Construct an intelligent weight evaluation model; S44, training and verifying weight evaluation; S45. Dynamically adjust and optimize the weight evaluation model to obtain optimized weights.

14. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 13, characterized in that: The S41 comprises the following steps: S411. Classify data sources; S412. Formulate data labeling rules according to data classification and complete data labeling; S413, integrating multidimensional data and annotation information to establish an evaluation data set; S414, dividing the evaluation data set into a training set, a validation set, and a test set.

15. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 13, characterized in that: The S42 includes: Extract time domain features and frequency domain features from physiological data; Extract statistical features of behavioral data, including frequency, duration, and trajectory features including path length and speed; Extract statistical features of environmental data, including mean and variance; Standardize or normalize the extracted feature values.

16. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 13, characterized in that: The weight evaluation model includes a data input layer, a feature extraction layer, a base model layer, an integration layer and an output layer, wherein: The data input layer obtains input data from the evaluation dataset; The feature extraction layer is used to extract hidden feature information of the input data, including the CNN model based on the convolutional neural network algorithm and the RNN model based on the recurrent neural network algorithm; The base model layer includes an SVM base model based on the support vector machine algorithm and an RF base model based on the random forest algorithm. The SVM base model is used to capture the nonlinear relationship in the input data, and the RF base model improves the robustness and generalization ability of the model by integrating multiple decision trees. The integration layer is used to integrate the output results of multiple base models to generate the final prediction results; The output layer outputs the weight evaluation results according to the calculated values ​​of the integration layer.

17. The intelligent assessment and evaluation method for emergency risk avoidance of mine accidents according to claim 13, characterized in that: The S45 comprises the following steps: S451, dynamically update online learning model; S452. Model the emergency risk-avoidance behavior of miners as a Markov decision process; S453. Use reinforcement learning algorithms to optimize risk hedging strategies; S454, dynamically adjust the classification threshold and apply it to the online SVM or incremental random forest model; S455. Further optimize model parameters and strategies based on environmental feedback.

Images