Electrocardio-based fatigue prediction method and device and storage medium

Abstract

The present application relates to a fatigue prediction method and device based on electrocardio and a storage medium, applied to the technical field of fatigue state detection, comprising: obtaining the subjective and objective data of firefighters by simulating the training movement of firefighters, obtaining the hierarchical threshold table under the optimal clustering number by clustering analysis on the subjective data, obtaining the weight of each subjective feature by weight analysis on the subjective data, establishing the fatigue index F based on the weight of the subjective feature, obtaining the time domain, frequency domain and nonlinear index by heart rate variability analysis on the objective data, establishing a model, taking the time domain, frequency domain and nonlinear index as the input of the model, taking the fatigue index F as the output of the model to train the model, obtaining the fatigue prediction model, and predicting the fatigue grade of firefighters through the fatigue prediction model.

Description

Technical Field

[0001] This invention relates to the field of fatigue state detection technology, specifically to a fatigue prediction method, device, and storage medium based on electrocardiogram. Background Technology

[0002] Firefighters are prone to fatigue during firefighting operations. Firefighting work is demanding, and firefighters face complex and ever-changing high-temperature fire environments, as well as multiple challenges such as carrying heavy loads and climbing to heights. This can easily lead to fatigue, affecting firefighting efficiency and judgment, and causing terrible consequences. Therefore, it is crucial to predict and assess the level of firefighter fatigue.

[0003] Current methods for fatigue monitoring mainly focus on the application of electroencephalography (EEG) and electromyography (EMG), and tend to be suitable for fatigue problems of personnel in static working states. They are not applicable to fatigue research of dynamic jobs such as firefighters who perform a large number of high-frequency and large-amplitude movements.

[0004] Further research on fatigue prediction for firefighters' physiological state is lacking, making it impossible to achieve targeted real-time monitoring and early warning of fatigue status for firefighters operating in fire situations. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a fatigue prediction method, device and storage medium based on electrocardiogram, so as to solve the problem that the existing fatigue monitoring methods mainly focus on the application of electroencephalogram, electromyography and other methods, and tend to be suitable for the fatigue problem of personnel in static working state, which is not applicable to the fatigue level prediction of dynamic jobs such as firefighters with a large number of high frequency and large amplitude movements.

[0006] According to a first aspect of the present invention, a fatigue prediction method based on electrocardiogram is provided, the method comprising:

[0007] By simulating firefighter training exercises with wearers of ECG vests, multiple sets of subjective and objective index data under simulated firefighter exercise were obtained. The objective index data are the firefighters' ECG index data, and the subjective index data are the firefighters' subjective fatigue value (RPE) and reaction time (SRT) after exercise.

[0008] K-means clustering analysis was performed on the subjective fatigue value (RPE) and reaction time (SRT) in each group of subjective index data. The distribution of clustering coefficients under different K values ​​was traversed to obtain the optimal number of clusters for the subjective index data, and a grading threshold table of subjective fatigue value (RPE) and reaction time (SRT) under the optimal number of clusters was output.

[0009] The subjective fatigue value (RPE) and reaction time (SRT) are objectively weighted using the entropy weight method to obtain their respective weights.

[0010] The fatigue index F is obtained based on the weights of the subjective fatigue value RPE and the reaction time SRT. The fatigue index F is then substituted into the classification threshold table of subjective fatigue value RPE and reaction time SRT under the optimal cluster number to obtain the classification threshold table of fatigue index F.

[0011] The electrocardiogram data of the firefighters were analyzed in the time domain, frequency domain, and nonlinear domain to obtain time domain index, frequency domain index, and nonlinear index data.

[0012] Using the time-domain index, frequency-domain index, and nonlinear index as inputs, and the fatigue index F as output, a neural network model is built using the newff function. The initial parameters of the neural network model are set, and the parameters of the neural network model are iteratively trained using the gradient descent method until the error of the output of the neural network model meets the preset error standard, thus obtaining the fatigue prediction model.

[0013] The real-time time-domain index, frequency-domain index, and nonlinear index of the firefighter are obtained. The real-time time-domain index, frequency-domain index, and nonlinear index of the firefighter are input into the fatigue prediction model, and the real-time fatigue index F of the firefighter is output. The fatigue level of the firefighter is obtained based on the grading threshold table of the fatigue index F.

[0014] Preferably,

[0015] The acquisition of multiple sets of subjective and objective indicator data under simulated firefighter movement includes:

[0016] The simulated firefighter training exercise is treadmill running with a gradient increase in speed. The initial speed is 4 km / h, each stage lasts for 4 minutes, and the speed increases by 1 km / h in each stage. After each stage of exercise, the ECG data of the wearer of the ECG vest is obtained, and the wearer is assessed for subjective fatigue (RPE) to obtain the subjective fatigue value (RPE value), and a reaction time (SRT) value is obtained by reaction time testing.

[0017] Preferably,

[0018] The step of performing K-means clustering analysis on the subjective fatigue value (RPE) and reaction time (SRT) values ​​in each group of subjective index data, and iterating through the clustering coefficient distribution under different K values ​​to obtain the optimal number of clusters for the subjective index data, includes:

[0019] Create a dataset containing N objects, each object including the subjective fatigue value (RPE) and reaction time (SRT) obtained after each stage of exercise;

[0020] K sample points are randomly selected as initial centroids, which are the center points of a cluster. The value of K ranges from 2 to 10.

[0021] Iterate through each remaining sample point, calculate its Euclidean distance to the K centroids, and assign it to the cluster containing the centroid with the smallest distance.

[0022] In each iteration, the mean of all samples in each cluster is calculated, and this mean is used as the new centroid of each cluster.

[0023] Repeat the above process until the centroids and clusters no longer change or the maximum number of iterations is reached to obtain the classification result.

[0024] Based on the new classification results, the silhouette coefficients of different clustering K values ​​are calculated and output, and the optimal number of clusters is obtained from the silhouette coefficients.

[0025] Preferably,

[0026] The objective weighting calculation of the subjective fatigue value (RPE) and reaction time (SRT) using the entropy weighting method yields the following weights for the subjective fatigue value (RPE) and reaction time (SRT):

[0027] Each set of subjective fatigue value (RPE) and reaction time (SRT) is treated as a sample object, and each feature in the sample object is normalized one by one.

[0028] Calculate the proportion of all values ​​of the normalized feature in each sample;

[0029] The entropy of each feature is calculated based on the proportion of all possible values ​​for each feature.

[0030] The difference coefficient for each feature is calculated based on the entropy of each feature;

[0031] The difference coefficients of each feature are normalized to obtain the weights of the subjective fatigue value (RPE) and reaction time (SRT).

[0032] Preferably,

[0033] The fatigue index F, obtained based on the weights of the subjective fatigue value (RPE) and the reaction time (SRT), includes:

[0034] The fatigue index F is obtained by multiplying the subjective fatigue value RPE by its weight and adding the reaction time SRT by its weight.

[0035] Preferably,

[0036] The acquisition of firefighters' real-time time-domain indicators, frequency-domain indicators, and nonlinear indicators includes:

[0037] Firefighters wear ECG vests, and ECG sensor in the vests is used to acquire the firefighters' electrocardiogram data.

[0038] The collected electrocardiogram data of the firefighters is sent to the terminal interface through the Bluetooth module in the wearable electrocardiogram vest.

[0039] The terminal interface performs time-domain, frequency-domain, and nonlinear analyses on the electrocardiogram index data to obtain the firefighter's real-time time-domain, frequency-domain, and nonlinear index data.

[0040] The terminal interface inputs the firefighter's real-time time-domain index, frequency-domain index, and nonlinear index data into the fatigue prediction model, and the fatigue prediction model outputs the firefighter's fatigue level.

[0041] The terminal interface sends the fatigue level to the display interface of the wearer wearing the ECG vest via the Bluetooth module.

[0042] According to a second aspect of the present invention, an electrocardiogram-based fatigue prediction device is provided, the device comprising:

[0043] Subjective and objective data acquisition module: used to acquire multiple sets of subjective and objective index data under simulated firefighter training exercise by wearing an ECG vest. The objective index data is the firefighter's ECG index data, and the subjective index data is the firefighter's subjective fatigue value (RPE) and reaction time (SRT) value after exercise.

[0044] The grading threshold acquisition module is used to perform K-means cluster analysis on the subjective fatigue value (RPE) and reaction time (SRT) in each group of subjective index data, traverse the distribution of clustering coefficients under different K values, obtain the optimal number of clusters for the subjective index data, and output a grading threshold table for the subjective fatigue value (RPE) and reaction time (SRT) under the optimal number of clusters.

[0045] Weighting Analysis Module: This module is used to objectively calculate the weights of two feature data, the Subjective Fatigue Value (RPE) and the Reaction Time (SRT), using the entropy weighting method, to obtain the weights of the RPE and SRT values, respectively.

[0046] Fatigue index acquisition module: used to obtain fatigue index F based on the weight of subjective fatigue value RPE value and reaction time SRT value, and substitute the fatigue index F into the classification threshold table of subjective fatigue value RPE value and reaction time SRT value under the optimal cluster number to obtain the classification threshold table of fatigue index F;

[0047] ECG data analysis module: used to perform time-domain, frequency-domain, and nonlinear analyses on the ECG data of the firefighters to obtain time-domain, frequency-domain, and nonlinear index data;

[0048] Model training module: This module is used to build a neural network model using the time-domain index, frequency-domain index, and nonlinear index as inputs, and the fatigue index F as output. It sets the initial parameters of the neural network model, and uses the gradient descent method to iteratively train the parameters of the neural network model until the error of the output of the neural network model meets the preset error standard, thereby obtaining the fatigue prediction model.

[0049] Fatigue level acquisition module: used to acquire the firefighter's real-time time domain index, frequency domain index and nonlinear index, input the firefighter's real-time time domain index, frequency domain index and nonlinear index into the fatigue prediction model, output the firefighter's real-time fatigue index F, and obtain the firefighter's fatigue level based on the grading threshold table of fatigue index F.

[0050] According to a third aspect of the present invention, a storage medium is provided, the storage medium storing a computer program, which, when executed by a host controller, implements the steps of the above-described method.

[0051] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0052] This application simulates firefighter training movements using an ECG vest, acquiring subjective and objective data from firefighters. Cluster analysis of the subjective data yields a grading threshold table with the optimal number of clusters. Weight analysis of the subjective data assigns weights to each subjective feature, establishing a fatigue index F based on these weights. Time-domain, frequency-domain, and nonlinear analyses of the objective data yield time-domain, frequency-domain, and nonlinear indices. A model is then built, initial parameters are set, and the time-domain, frequency-domain, and nonlinear indices are used as inputs. The fatigue index F is used as the model's output for training, resulting in a fatigue prediction model. This model predicts the fatigue level of firefighters. Compared to existing methods using EEG and EMG, this approach is more suitable for predicting fatigue levels in dynamic jobs like firefighters, which involve numerous, high-frequency, and large-amplitude movements.

[0053] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0054] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0055] Figure 1 This is a flowchart illustrating an electrocardiogram-based fatigue prediction method according to an exemplary embodiment;

[0056] Figure 2 This is a schematic diagram of a fatigue prediction device based on electrocardiogram, according to another exemplary embodiment.

[0057] In the attached diagram: 1-Subjective and objective data acquisition module, 2-Grading threshold acquisition module, 3-Weight analysis module, 4-Fatigue index acquisition module, 5-ECG data analysis module, 6-Model training module, 7-Fatigue level acquisition module. Detailed Implementation

[0058] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0059] Example 1

[0060] Figure 1 This is a flowchart illustrating an electrocardiogram-based fatigue prediction method according to an exemplary embodiment, the method comprising:

[0061] S1, by simulating firefighter training exercise through the wearer of the ECG vest, multiple sets of subjective and objective index data under simulated firefighter exercise are obtained. The objective index data is the firefighter's ECG index data, and the subjective index data is the firefighter's subjective fatigue value (RPE) and reaction time (SRT) value after exercise.

[0062] S2, perform K-means cluster analysis on the subjective fatigue value (RPE) and reaction time (SRT) in each group of subjective index data, traverse the distribution of clustering coefficients under different K values, obtain the optimal number of clusters for the subjective index data, and output the hierarchical threshold table of subjective fatigue value (RPE) and reaction time (SRT) under the optimal number of clusters.

[0063] S3, the subjective fatigue value RPE value and reaction time SRT value are objectively weighted using the entropy weight method to obtain the weights of the subjective fatigue value RPE value and reaction time SRT value respectively.

[0064] S4. Based on the weights of the subjective fatigue value (RPE) and the reaction time (SRT), the fatigue index F is obtained. The fatigue index F is then substituted into the classification threshold table of the subjective fatigue value (RPE) and the reaction time (SRT) under the optimal cluster number to obtain the classification threshold table of the fatigue index F.

[0065] S5, perform time-domain, frequency-domain and nonlinear analysis on the electrocardiogram data of the firefighters to obtain time-domain index, frequency-domain index and nonlinear index data;

[0066] S6. Using the time-domain index, frequency-domain index, and nonlinear index as inputs, and the fatigue index F as output, a neural network model is built using the newff function. The initial parameters of the neural network model are set, and the gradient descent method is used to iteratively train the parameters of the neural network model until the error of the output of the neural network model meets the preset error standard, thereby obtaining the fatigue prediction model.

[0067] S7. Obtain the firefighter's real-time time-domain index, frequency-domain index, and nonlinear index. Input the firefighter's real-time time-domain index, frequency-domain index, and nonlinear index into the fatigue prediction model and output the firefighter's real-time fatigue index F. Obtain the firefighter's fatigue level based on the grading threshold table of the fatigue index F.

[0068] Understandably, the ECG vest wearer simulates firefighter training exercises, and subjective and objective index data are sampled during the exercise. The ECG index is defined as X, the subjective fatigue value (RPE) as Y, and the reaction time (SRT) as Z. The two feature values ​​Y (RPE) and Z (SRT) from each data set are used for K-means cluster analysis on the MATLAB platform. The code specifies the number of clusters K as 2-10, and the optimal clustering criterion is determined based on the silhouette coefficient. After initializing the silhouette coefficient, the distribution of cluster coefficients under different K values ​​is iterated. Each clustering process is repeated ten times for iteration. The silhouette coefficient reflection plots for different clustering K values ​​are calculated and output, allowing for the calculation of the optimal number of clusters. A grading threshold table for Y (RPE value) and Z (SRT value) under the optimal clustering is also output. The entropy weight method is applied to objectively weight the two feature data, Y (RPE value) and Z (SRT value). Considering the diversity of the two feature data values, the greater the difference in the magnitude of the feature values, i.e., the higher the degree of dispersion, the greater the importance of the feature and the more information it contains. The collected ECG signals are then used... Heart rate variability analysis software was used to perform time-domain, frequency-domain, and nonlinear analyses, yielding time-domain indices HR, RRI, SDNN, and RMSSD; frequency-domain data LF, HF, and LF / HF; and nonlinear indices SD1, SD2, and SD2 / SD1. The time-domain analysis method directly calculates and analyzes the dispersion of the corresponding heartbeat intervals using continuously measured electrocardiogram waveforms to obtain time-domain indices, including RR (the time interval between R waves in a continuous electrocardiogram), SDNN (the standard deviation of normal heartbeat intervals in milliseconds), and RMSS. Physiological indicators such as D (root mean square of the difference between consecutive R-repetition intervals) are analyzed in the frequency domain. The discrete Fourier transform algorithm is used to convert the time series of heartbeat intervals into the frequency domain, which is represented by a spectral distribution or power spectral density map. The main data obtained are TP (the sum of the high-frequency, low-frequency, and very low-frequency variance of all normal heartbeat intervals), LF (the variance of normal heartbeat intervals in the low-frequency range, representing sympathetic and parasympathetic nerve activity, with a spectral range of 0.04-0.15Hz), and HF (the variance of normal heartbeat intervals in the high-frequency range, representing parasympathetic nerve activity, with a spectral range of 0.15-0.15Hz).The data were obtained from the following parameters: 4Hz), LF / HF (the ratio of low and high frequency power, representing the balance of autonomic nerve activity), and finally, time-domain indicators HR (heart rate), RRI (heart rate interval), SDNN, RMSSD, and frequency-domain data LF, HF, LF / HF, as well as nonlinear indicator data SD1, SD2, SD2 / SD1. These 10 sets of indicator data were used as the model input X, and the comprehensive evaluation index of work fatigue F was used as the output. A BP neural network was built for model training. 42 sample data were collected in the experiment. Each sample included 10 input feature values ​​and 1 output. The training and prediction data were set at a ratio of 19:2. The number of input layer nodes was 10, the number of hidden layer nodes was 8, and the number of output layer nodes was 1. The training set was processed using the mapminmax function [-1]. 1] Normalization processing was performed, and a neural network model was built using the newff function. The recursive function used was purelin, and the gradient descent method was employed for training. The maximum number of training iterations was set to 1000, the learning rate was 0.01, and the minimum training error was 0.00001. After training, the BP neural network model achieved the required error and the most accurate prediction, enabling it to predict the fatigue level of firefighters. This mobile terminal system for monitoring and warning firefighter fatigue collects the wearer's electrocardiogram (ECG) data and transmits it to the platform via Bluetooth. Heart rate variability analysis is performed to obtain time-domain and frequency-domain index data, which are then input into the fatigue state prediction model. The system outputs a comprehensive fatigue state evaluation index F for the wearer and automatically incorporates it into a three-level classification model for firefighter operational fatigue state to evaluate the fatigue index F. The fatigue level is then transmitted to the wearer via Bluetooth. This application simulates firefighter fatigue using an ECG vest. This method involves training firefighters, acquiring subjective and objective data, performing cluster analysis on the subjective data to obtain a grading threshold table with the optimal number of clusters, weighting the subjective data to obtain the weights of each subjective feature, establishing a fatigue index F based on these weights, and performing time-domain, frequency-domain, and nonlinear analysis on the objective data to obtain time-domain, frequency-domain, and nonlinear indices. A model is then established, initial parameters are set, and the time-domain, frequency-domain, and nonlinear indices are used as inputs. The fatigue index F is used as the output to train the model, resulting in a fatigue prediction model. This model predicts the fatigue level of firefighters. Compared to existing methods using EEG and EMG, this method is more suitable for predicting the fatigue level of firefighters, a dynamic occupation involving numerous, high-frequency, and large-amplitude movements.

[0069] Preferably,

[0070] The acquisition of multiple sets of subjective and objective indicator data under simulated firefighter movement includes:

[0071] The simulated firefighter training exercise is treadmill gradient running with an initial speed of 4 km / h and a duration of 4 minutes for each stage. The speed increases by 1 km / h for each stage. After each stage of exercise, the ECG data of the wearer of the ECG vest is obtained, and the wearer is given a subjective fatigue score (RPE) to obtain a subjective fatigue value (RPE value). A reaction time test is also performed to obtain a reaction time (SRT) value.

[0072] Understandably, the firefighter training exercise involves treadmill running with a gradient increase in speed. The initial speed is 4 km / h, each phase lasts 4 minutes, and the speed increases by 1 km / h per phase. Subjective fatigue assessment (RPE) and simple reaction time (SRT, ms) are performed immediately after each phase. Before conducting the experiment, the sensors need to be tested and calibrated to ensure the accuracy and reliability of the data. The scoring principle of the 6-20 Borg's RPE (Responsive Physical Exhaustion) scale is as follows: it combines the individual's subjective perception of fatigue changes during the task process, dividing the state into nine different levels: very easy, very easy, easy, somewhat easy, slightly tired, tired, very tired, extremely tired, and exhausted. Individuals score from 6 points onwards, with higher RPE scores for more severe fatigue, up to a maximum of 20 points. The simple reaction time (SRT, ms) test scoring principle is as follows: reaction time is a reliable indicator of fatigue. The H.Benchmark 2.0 reaction time training device is selected, requiring full concentration on the test screen. When the screen color changes from blue to green, the button is pressed, and the reaction time is recorded in millimeters (ms). Each group is tested 5 times, and the average value is taken as the final result. By measuring the reaction time of the subjects, the changes in the individual's concentration and reaction speed as the task progresses are analyzed.

[0073] Preferably,

[0074] The step of performing K-means clustering analysis on the subjective fatigue value (RPE) and reaction time (SRT) values ​​in each group of subjective index data, and iterating through the clustering coefficient distribution under different K values ​​to obtain the optimal number of clusters for the subjective index data, includes:

[0075] Create a dataset containing N objects, each object including the subjective fatigue value (RPE) and reaction time (SRT) obtained after each stage of exercise;

[0076] K sample points are randomly selected as initial centroids, which are the center points of a cluster. The value of K ranges from 2 to 10.

[0077] Iterate through each remaining sample point, calculate its Euclidean distance to the K centroids, and assign it to the cluster containing the centroid with the smallest distance.

[0078] In each iteration, the mean of all samples in each cluster is calculated, and this mean is used as the new centroid of each cluster.

[0079] Repeat the above process until the centroids and clusters no longer change or the maximum number of iterations is reached to obtain the classification result.

[0080] Based on the new classification results, calculate and output profile coefficient rendering maps of different cluster K values, and obtain the optimal number of clusters based on the profile coefficient rendering maps;

[0081] It is understandable that obtaining the optimal number of clusters includes:

[0082] (1) Create a dataset containing 42 objects, each object having two feature values ​​Ti = (Yi, Zi). Where, Yi: RPE value, Zi: SRT value, i = 1, 2, ..., 42;

[0083] (2) Randomly select K sample points as the initial centroids, which are also the center points (initial means) of a cluster (aj, bj), where j = 1, 2...k;

[0084] (3) Iterate through each remaining sample point, calculate its Euclidean distance d to the K centroids, and assign it to the cluster containing the centroid with the smallest distance. Where i = 1, 2, ..., 42;

[0085]

[0086]

[0087] ...

[0088] ...

[0089]

[0090] (4) Calculate the new centroids. In each iteration, calculate the mean of all samples in each cluster, which is the new centroid of each cluster;

[0091] Repeat steps (3) and (4) until the centroid and clusters no longer change or the maximum number of iterations is reached to obtain the classification result.

[0092] Preferably,

[0093] The objective weighting calculation of the subjective fatigue value (RPE) and reaction time (SRT) using the entropy weighting method yields the following weights for the subjective fatigue value (RPE) and reaction time (SRT):

[0094] Each set of subjective fatigue value (RPE) and reaction time (SRT) is treated as a sample object, and each feature in the sample object is normalized one by one.

[0095] Calculate the proportion of all values ​​of the normalized feature in each sample;

[0096] The entropy of each feature is calculated based on the proportion of all possible values ​​for each feature.

[0097] The difference coefficient for each feature is calculated based on the entropy of each feature;

[0098] The difference coefficient of each feature is normalized to obtain the weights of the subjective fatigue value (RPE) and reaction time (SRT).

[0099] It is understandable that the weights used to obtain the subjective fatigue value (RPE) and the reaction time (SRT) include:

[0100] (1) Data normalization. Normalization is performed on each feature of the sample object individually:

[0101]

[0102] Y" i =Y′ i +0.001

[0103]

[0104] Z" i =Z′ i +0.001

[0105] (2) For a given feature, calculate Y″ for each sample. i Z″ i The proportion of all values ​​under this feature:

[0106]

[0107]

[0108] (3) Calculate the entropy e of each feature. j :

[0109]

[0110]

[0111] (4) Calculate the difference coefficient g for each feature. j :

[0112] g1 = 1 - e1

[0113] g2=1-e2

[0114] (5) Regarding the difference coefficient g j Perform normalization and calculate the weight w for each feature. j :

[0115]

[0116]

[0117] Preferably,

[0118] The fatigue index F, obtained based on the weights of the subjective fatigue value (RPE) and the reaction time (SRT), includes:

[0119] The fatigue index F is obtained by multiplying the subjective fatigue value RPE by its weight and adding the reaction time SRT by its weight.

[0120] Understandably, for example, the K-means clustering algorithm is used to obtain the grading thresholds for subjective fatigue Y (RPE) and reaction time Z (SRT) when the optimal number of clusters K=5. After transformational analysis, the weight coefficients of the two features, subjective fatigue value Y (RPE value) and reaction time Y (SRT value), are found to be 0.61 and 0.39, respectively. These are then substituted into the firefighter's comprehensive evaluation index model for work fatigue: F = 0.61Y + 0.39Z, to obtain the grading threshold for the comprehensive evaluation index F of work fatigue. Finally, the fatigue level is divided into three levels according to the analysis, establishing a three-level classification model for firefighter's work fatigue state. The first level is from easy to mild fatigue, and the fatigue index F[0,0]. Level 2 (0.27) requires firefighters to quickly enter the fire scene and carry out firefighting and rescue work; Level 2 is moderate fatigue, which is a stage of rapid increase in fatigue development, with a fatigue index of F[0.27, 0.66). This stage is the critical time for firefighters to carry out firefighting and rescue work. Under the dual regulation of the sympathetic / parasympathetic nervous system and the vagus nerve, the body is in an active state, with highly concentrated attention and action; Level 3 is severe fatigue to exhaustion, with a fatigue index of F[0.66, 1.00]. In this stage, the body is affected by the vagus nerve of the nervous system. As the rescue work continues, the firefighters' fatigue accumulates to a peak, and their attention and reaction speed decline rapidly. Firefighters need to withdraw from the fire scene in time and take turns to rest.

[0121] Preferably,

[0122] The acquisition of firefighters' real-time time-domain indicators, frequency-domain indicators, and nonlinear indicators includes:

[0123] Firefighters wear ECG vests, and ECG sensor in the vests is used to acquire the firefighters' electrocardiogram data.

[0124] The collected electrocardiogram data of the firefighters is sent to the terminal interface through the Bluetooth module in the wearable electrocardiogram vest.

[0125] The terminal interface performs time-domain, frequency-domain, and nonlinear analyses on the electrocardiogram index data to obtain the firefighter's real-time time-domain, frequency-domain, and nonlinear index data.

[0126] The terminal interface inputs the firefighter's real-time time-domain index, frequency-domain index, and nonlinear index data into the fatigue prediction model, and the fatigue prediction model outputs the firefighter's fatigue level.

[0127] The terminal interface sends the fatigue level to the display interface of the wearable ECG vest via the Bluetooth module;

[0128] Understandably, this firefighter fatigue monitoring and early warning mobile terminal system collects the wearer's electrocardiogram data, transmits it to the platform via Bluetooth, performs heart rate variability analysis, obtains time-domain and frequency-domain index data, inputs it into the fatigue state prediction model, outputs the wearer's comprehensive fatigue state evaluation index F, and automatically inputs it into the firefighter's three-level classification model for operational fatigue state to evaluate the fatigue index F, and transmits the fatigue level to the wearer via Bluetooth.

[0129] Example 2:

[0130] Figure 2 This is a system schematic diagram of an electrocardiogram-based fatigue prediction device according to another exemplary embodiment, the device comprising:

[0131] Subjective and objective data acquisition module 1: used to simulate firefighter training exercises by wearing an ECG vest, and acquire multiple sets of subjective and objective index data under simulated firefighter exercise. The objective index data is the firefighter's ECG index data, and the subjective index data is the firefighter's subjective fatigue value (RPE) and reaction time (SRT) value after exercise.

[0132] The grading threshold acquisition module 2 is used to perform K-means cluster analysis on the subjective fatigue value (RPE) and reaction time (SRT) in each group of subjective index data, traverse the distribution of clustering coefficients under different K values, obtain the optimal number of clusters for the subjective index data, and output the grading threshold table for the subjective fatigue value (RPE) and reaction time (SRT) under the optimal number of clusters.

[0133] Weighting Analysis Module 3: This module is used to objectively calculate the weights of two feature data, the Subjective Fatigue Value (RPE) and the Reaction Time (SRT), using the entropy weighting method, to obtain the weights of the Subjective Fatigue Value (RPE) and the Reaction Time (SRT) respectively.

[0134] Fatigue index acquisition module 4: used to obtain fatigue index F based on the weight of subjective fatigue value RPE value and reaction time SRT value, and substitute the fatigue index F into the classification threshold table of subjective fatigue value RPE value and reaction time SRT value under the optimal cluster number to obtain the classification threshold table of fatigue index F;

[0135] ECG data analysis module 5: used to perform time-domain, frequency-domain, and nonlinear analyses on the ECG index data of the firefighters to obtain time-domain index, frequency-domain index, and nonlinear index data;

[0136] Model training module 6: It is used to build a neural network model with the time domain index, frequency domain index and nonlinear index as input and the fatigue index F as output, through the newff function, set the initial parameters of the neural network model, and use the gradient descent method to iteratively train the parameters of the neural network model until the error of the output of the neural network model meets the preset error standard, so as to obtain the fatigue prediction model.

[0137] Fatigue level acquisition module 7: used to acquire the firefighter's real-time time domain index, frequency domain index and nonlinear index, input the firefighter's real-time time domain index, frequency domain index and nonlinear index into the fatigue prediction model, output the firefighter's real-time fatigue index F, and obtain the firefighter's fatigue level based on the grading threshold table of fatigue index F.

[0138] Understandably, the subjective and objective data acquisition module 1 is used to simulate firefighter training exercises by wearing an ECG vest, acquiring multiple sets of subjective and objective indicator data under simulated firefighter exercise. The objective indicator data is the firefighter's ECG data, and the subjective indicator data is the firefighter's subjective fatigue value (RPE) and reaction time (SRT) after exercise. The grading threshold acquisition module 2 is used to perform K-means clustering analysis on the subjective fatigue value (RPE) and reaction time (SRT) in each set of subjective indicator data, traversing the clustering coefficient distribution under different K values, obtaining the optimal number of clusters for the subjective indicator data, and outputting a grading threshold table for the subjective fatigue value (RPE) and reaction time (SRT) under the optimal number of clusters. Through weighted... The reanalysis module 3 is used to objectively calculate the weights of two feature data, the subjective fatigue value (RPE) and the reaction time (SRT), using the entropy weight method, to obtain the weights of the subjective fatigue value (RPE) and the reaction time (SRT), respectively. The fatigue index acquisition module 4 is used to obtain the fatigue index F based on the weights of the subjective fatigue value (RPE) and the reaction time (SRT), and then substitutes the fatigue index F into the grading threshold table of the subjective fatigue value (RPE) and the reaction time (SRT) under the optimal cluster number to obtain the grading threshold table of the fatigue index F. The electrocardiogram (ECG) data analysis module 5 is used to perform time-domain, frequency-domain, and nonlinear analyses on the firefighters' ECG index data to obtain time-domain index, frequency-domain index, and nonlinear index data. Model training module 6 is used to build a neural network model using the time-domain index, frequency-domain index, and nonlinear index as inputs, and the fatigue index F as output. It sets the initial parameters of the neural network model and iteratively trains the parameters using gradient descent until the error of the neural network model's output meets a preset error standard, thus obtaining a fatigue prediction model. Fatigue level acquisition module 7 is used to acquire the firefighter's real-time time-domain index, frequency-domain index, and nonlinear index. These indicators are then input into the fatigue prediction model, outputting the firefighter's real-time fatigue index F. Based on the fatigue index F's grading threshold table, the firefighter's fatigue level is determined. The present application simulates the training exercise of firefighters by wearing an ECG vest, obtains subjective and objective data of firefighters, performs cluster analysis on the subjective data to obtain a classification threshold table with the optimal number of clusters, performs weight analysis on the subjective data to obtain the weight of each subjective feature, establishes a fatigue index F based on the weight of the subjective features, and performs time-domain, frequency-domain and nonlinear analysis on the objective data to obtain time-domain, frequency-domain and nonlinear indicators, establishes a model, sets the initial parameters of the model, uses the time-domain, frequency-domain and nonlinear indicators as the input of the model, and uses the obtained fatigue index F as the output of the model to train the model to obtain a fatigue prediction model, and predicts the fatigue level of firefighters through the fatigue prediction model.Compared to existing research methods using electroencephalography (EEG) and electromyography (EMG), the proposed method is more suitable for predicting fatigue levels in dynamic jobs such as firefighters, which involve a large number of high-frequency, large-amplitude movements.

[0139] Example 3:

[0140] This embodiment provides a storage medium storing a computer program. When the computer program is executed by a host controller, it implements the various steps in the above method.

[0141] It is understood that the storage medium mentioned above can be a read-only memory, a hard disk, or an optical disk, etc.

[0142] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0143] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.

[0144] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0145] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0146] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0147] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0148] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0149] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0150] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A fatigue prediction method based on electrocardiogram, characterized in that, The method includes: By simulating firefighter training exercises with wearers of ECG vests, multiple sets of subjective and objective index data under simulated firefighter exercise were obtained. The objective index data were the firefighters' ECG index data, and the subjective index data were the firefighters' subjective fatigue value (RPE) and reaction time (SRT) after exercise. K-means clustering analysis was performed on the subjective fatigue value (RPE) and reaction time (SRT) values ​​in each group of subjective index data. The distribution of clustering coefficients under different K values ​​was traversed to obtain the optimal number of clusters for the subjective index data, including: Step S201: Create a dataset, which includes N objects, each object including the subjective fatigue value (RPE) and reaction time (SRT) obtained after each stage of exercise; Step S202: Randomly select K sample points as initial centroids, which are the center points of a cluster. The value of K ranges from 2 to 10. Step S203: Traverse each remaining sample point, calculate its Euclidean distance to the K centroids, and assign it to the cluster containing the centroid with the smallest distance. Step S204: In each iteration, calculate the mean of all samples in each cluster and use the mean as the new centroid of each cluster; Step S205: Repeat steps S203 to S204 until the centroid and cluster do not change or the maximum number of iterations is reached to obtain the classification result; Based on the classification results, calculate and output profile coefficient line graphs for different cluster K values, obtain the optimal number of clusters based on the profile coefficient line graphs, and output a grading threshold table for subjective fatigue value (RPE) and reaction time (SRT) value under the optimal number of clusters. The subjective fatigue value (RPE) and reaction time (SRT) are objectively weighted using the entropy weight method. The weights for RPE and SRT are then calculated, including: Each set of subjective fatigue value (RPE) and reaction time (SRT) is treated as a sample object, and each feature in the sample object is normalized one by one. Calculate the proportion of all values ​​of the normalized feature in each sample; The entropy of each feature is calculated based on the proportion of all possible values ​​for each feature. The difference coefficient for each feature is calculated based on the entropy of each feature; The difference coefficient of each feature is normalized to obtain the weights of the subjective fatigue value (RPE) and reaction time (SRT). The fatigue index F is obtained based on the weights of the subjective fatigue value RPE and the reaction time SRT, including: multiplying the subjective fatigue value RPE by its weight and adding the reaction time SRT by its weight to obtain the fatigue index F; and substituting the fatigue index F into the classification threshold table of the subjective fatigue value RPE and the reaction time SRT under the optimal cluster number to obtain the classification threshold table of the fatigue index F. The electrocardiogram data of the firefighters were analyzed in the time domain, frequency domain, and nonlinear domain to obtain time domain index, frequency domain index, and nonlinear index data. Using the time-domain index, frequency-domain index, and nonlinear index as inputs, and the fatigue index F as output, a neural network model is built using the newff function. The initial parameters of the neural network model are set, and the parameters of the neural network model are iteratively trained using the gradient descent method until the error of the output of the neural network model meets the preset error standard, thus obtaining the fatigue prediction model. The real-time time-domain index, frequency-domain index, and nonlinear index of the firefighter are obtained. The real-time time-domain index, frequency-domain index, and nonlinear index of the firefighter are input into the fatigue prediction model, and the real-time fatigue index F of the firefighter is output. The fatigue level of the firefighter is obtained based on the grading threshold table of the fatigue index F.

2. The method according to claim 1, characterized in that, The acquisition of multiple sets of subjective and objective indicator data under simulated firefighter movement includes: The simulated firefighter training exercise is treadmill running with a gradient increase in speed. The initial speed is 4 km / h, each stage lasts for 4 minutes, and the speed increases by 1 km / h in each stage. After each stage of exercise, the ECG data of the wearer of the ECG vest is obtained, and the wearer is assessed for subjective fatigue (RPE) to obtain the subjective fatigue value (RPE value), and a reaction time (SRT) value is obtained by reaction time testing.

3. The method according to claim 1, characterized in that, The acquisition of firefighters' real-time time-domain indicators, frequency-domain indicators, and nonlinear indicators includes: Firefighters wear ECG vests, and ECG sensor in the vests is used to acquire the firefighters' electrocardiogram data. The collected electrocardiogram data of the firefighters is sent to the terminal interface through the Bluetooth module in the wearable electrocardiogram vest. The terminal interface performs time-domain, frequency-domain, and nonlinear analyses on the electrocardiogram index data to obtain the firefighter's real-time time-domain, frequency-domain, and nonlinear index data. The terminal interface inputs the firefighter's real-time time-domain index, frequency-domain index, and nonlinear index data into the fatigue prediction model, and the fatigue prediction model outputs the firefighter's fatigue level. The terminal interface sends the fatigue level to the display interface of the wearer wearing the ECG vest via the Bluetooth module.

4. A fatigue prediction device based on electrocardiogram, characterized in that, The device includes: Subjective and objective data acquisition module: used to acquire multiple sets of subjective and objective index data under simulated firefighter training exercise by wearing an ECG vest. The objective index data is the firefighter's ECG index data, and the subjective data is the firefighter's subjective fatigue value (RPE) and reaction time (SRT) value after exercise. The tiered threshold acquisition module performs K-means clustering analysis on the subjective fatigue value (RPE) and reaction time (SRT) values ​​in each group of subjective indicator data. It iterates through the clustering coefficient distribution under different K values ​​to obtain the optimal number of clusters for the subjective indicator data, including: Step S201: Create a dataset, which includes N objects, each object including the subjective fatigue value (RPE) and reaction time (SRT) obtained after each stage of exercise; Step S202: Randomly select K sample points as initial centroids, which are the center points of a cluster. The value of K ranges from 2 to 10. Step S203: Traverse each remaining sample point, calculate its Euclidean distance to the K centroids, and assign it to the cluster containing the centroid with the smallest distance. Step S204: In each iteration, calculate the mean of all samples in each cluster and use the mean as the new centroid of each cluster; Step S205: Repeat steps S203 to S204 until the centroid and cluster do not change or the maximum number of iterations is reached to obtain the classification result; Based on the classification results, calculate and output profile coefficient line graphs for different cluster K values, obtain the optimal number of clusters based on the profile coefficient line graphs, and output a grading threshold table for subjective fatigue value (RPE) and reaction time (SRT) value under the optimal number of clusters. Weighting Analysis Module: This module is used to objectively calculate the weights of two feature data points, the Subjective Fatigue Value (RPE) and the Reaction Time (SRT), using the entropy weighting method. It obtains the weights for both the RPE and SRT values, including: Each set of subjective fatigue value (RPE) and reaction time (SRT) is treated as a sample object, and each feature in the sample object is normalized one by one. Calculate the proportion of all values ​​of the normalized feature in each sample; The entropy of each feature is calculated based on the proportion of all possible values ​​for each feature. The difference coefficient for each feature is calculated based on the entropy of each feature; The difference coefficient of each feature is normalized to obtain the weights of the subjective fatigue value (RPE) and reaction time (SRT). Fatigue Index Acquisition Module: Used to obtain the fatigue index F based on the weighted subjective fatigue value (RPE) and reaction time (SRT), including: Multiply the subjective fatigue value RPE by its weight and add the reaction time SRT by its weight to obtain the fatigue index F. Substitute the fatigue index F into the classification threshold table of subjective fatigue value RPE and reaction time SRT under the optimal cluster number to obtain the classification threshold table of fatigue index F. ECG data analysis module: used to perform time-domain, frequency-domain, and nonlinear analyses on the ECG data of the firefighters to obtain time-domain, frequency-domain, and nonlinear index data; Model training module: This module is used to build a neural network model using the time-domain index, frequency-domain index, and nonlinear index as inputs, and the fatigue index F as output. It sets the initial parameters of the neural network model, and uses the gradient descent method to iteratively train the parameters of the neural network model until the error of the output of the neural network model meets the preset error standard, thereby obtaining the fatigue prediction model. Fatigue level acquisition module: used to acquire the firefighter's real-time time domain index, frequency domain index and nonlinear index, input the firefighter's real-time time domain index, frequency domain index and nonlinear index into the fatigue prediction model, output the firefighter's real-time fatigue index F, and obtain the firefighter's fatigue level based on the grading threshold table of fatigue index F.

5. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by the main controller, implements each step of the electrocardiogram-based fatigue prediction method as described in any one of claims 1-3.

Images