Electrified work human body fatigue detection method based on electrocardiosignals

Abstract

A physical fatigue prediction method based on electrocardiosignal analysis aims to solve the problems that a traditional fatigue monitoring method is high in subjectivity and lack of scientificity, and objective and quantitative evaluation of physical fatigue of power workers is achieved by recognizing key physiological indexes and establishing a mathematical model related to the fatigue degree. According to the specific scheme, firstly, electrocardiosignals of a testee are collected, and heart rate variability (HRV) analysis is carried out to extract feature values; secondly, evaluating the mental load and the processing capacity of the testee through an n-back cognitive experiment, and taking the mental load and the processing capacity as an auxiliary judgment basis of a fatigue state; and finally, constructing a fatigue prediction model by using a support vector machine (SVM), and realizing dichotomy prediction of the fatigue state by taking the HRV characteristic value as input. The fatigue state of the operator is predicted in advance, the scientificity and objectivity of fatigue evaluation are improved, meanwhile, the influence of psychological factors of the operator is considered, dynamic monitoring and real-time feedback are achieved, and a manager is helped to optimize work arrangement and reduce safety accident risks.

Description

Technical Field

[0001] The present invention relates to the technical field of health monitoring and safety management for power operation personnel, and particularly relates to a method for detecting human body fatigue during live working based on electrocardiogram signals. Background Art

[0002] In modern industrial production and daily life, fatigue problems are widespread, posing a serious threat to the safety, health, and work efficiency of operation personnel. Due to the high-intensity and high-risk characteristics of power operations, extremely high requirements are placed on the physical and mental strength of operation personnel, and long-term work loads are extremely likely to cause physical fatigue of operation personnel. At the same time, with the popularization of stereoscopic display technology, long-term viewing of stereoscopic display content has also become an important factor leading to visual fatigue. Therefore, how to scientifically and objectively evaluate the fatigue state of operation personnel has become a research hotspot.

[0003] Currently, for the fatigue monitoring of power operation personnel, traditional methods mainly rely on subjective evaluation means, such as self-reported fatigue levels or simple questionnaires. However, these methods are easily affected by individual differences and subjective emotions and lack scientificity and objectivity. In recent years, with the development of bio-signal processing technology, fatigue assessment methods based on physiological indicators have gradually received attention. For example, the document "Fatigue Detection Method" proposes a method for predicting physical fatigue based on electrocardiogram signal analysis. This method comprehensively evaluates the fatigue state from both physiological and psychological dimensions by collecting and analyzing the electrocardiogram signals of operation personnel and combining with the n-back cognitive experiment. At the same time, a fatigue prediction model is constructed using the support vector machine (SVM) algorithm to achieve real-time monitoring and early warning of the fatigue state, providing an effective fatigue assessment tool for the power industry.

[0004] For the problem of stereoscopic display visual fatigue, the existing technologies mainly adopt two methods: subjective evaluation and objective measurement. The subjective evaluation method assesses the degree of visual fatigue by describing the subjective feelings of users after viewing or completing specially designed questionnaires, but it is easily affected by individual differences and lacks a unified quantification standard. The objective measurement method evaluates the degree of visual fatigue by measuring physiological indicators related to visual fatigue, such as heart rate variability (HRV) and electrooculogram (EOG). However, the measurement of a single physiological indicator often fails to comprehensively reflect the visual fatigue state of users. For this reason, CN106534844A proposes a three-dimensional display visual fatigue assessment system and method based on electrocardiogram and electrooculogram signals. This system constructs a three-dimensional visual fatigue prediction model by collecting electrocardiogram signals and electrooculogram signals of users when viewing stereoscopic display content and combining with subjective visual fatigue score values, realizing objective and quantitative evaluation of stereoscopic display visual fatigue.

[0005] The technical solution of CN106534844A provides a new idea and method for stereoscopic display visual fatigue assessment by comprehensively collecting and analyzing electrocardiogram signals, electrooculogram signals, and subjective visual fatigue scoring values. This technical solution realizes an objective and quantitative assessment of visual fatigue degree through multi-dimensional evaluation and model prediction, which helps to guide the design and optimization of stereoscopic display devices. However, this technical solution still has the following defects and deficiencies: Limitations in physiological signal collection: Although electrocardiogram signals and electrooculogram signals are collected, other physiological signals related to visual fatigue, such as electroencephalogram signals, may not be fully considered. In addition, the collection of physiological signals is vulnerable to external interferences, such as environmental noise and equipment accuracy, which may affect the accuracy of the assessment results.

[0006] Subjectivity of subjective evaluation: Although the subjective visual fatigue scoring value can reflect the user's visual fatigue feeling to a certain extent, there are still subjectivity and individual differences. Different users may have different feelings and evaluation criteria for visual fatigue, resulting in inconsistencies in the assessment results.

[0007] Insufficient model generalization ability: The prediction model constructed by this technical solution may only be applicable to specific stereoscopic display devices and viewing environments, and there may be problems with insufficient generalization ability for visual fatigue assessment under different devices and environmental conditions.

[0008] Lack of personalized assessment: Factors such as different users' physiological characteristics and visual habits may affect the degree of visual fatigue. However, this technical solution may not fully consider these individual differences, resulting in a lack of personalization in the assessment results.

[0009] To solve the above problems, the present invention proposes a more comprehensive and accurate fatigue detection method. On the basis of inheriting the advantages of the existing technology, this method further optimizes the collection and analysis methods of physiological signals, introduces more physiological indicators related to fatigue, and constructs a more accurate fatigue prediction model in combination with machine learning algorithms. At the same time, the present invention also considers the individual differences and changing factors of different users and different environments, realizes personalized and real-time monitoring and early warning of the fatigue state, and provides a more scientific and effective solution for the fatigue management of electric power operation personnel and the visual fatigue assessment of stereoscopic displays. Summary of the Invention

[0010] The technical problem to be solved by the present invention is to provide a method for detecting human body fatigue during live work based on electrocardiogram signals, to solve the problems of insufficient accuracy, strong subjectivity, poor real-time performance, and lack of personalized assessment existing in traditional fatigue detection methods in the field of electric power operation and stereoscopic display technology, and the defects such as incomplete collection of physiological signals and insufficient model generalization ability existing in the existing technology although it has tried to combine physiological signals and subjective feelings for comprehensive assessment.

[0011] To solve the above technical problems, the technical solution adopted by the present invention is: a physical fatigue prediction method based on electrocardiogram signal analysis, comprising the following steps: Step1: HRV signal acquisition, acquiring the electrocardiogram signal of the subject and performing heart rate variability (HRV) analysis; Step2: HRV analysis method, performing HRV time-domain analysis and frequency-domain analysis on the acquired electrocardiogram signal, and extracting HRV characteristic values; Step3: n-back cognitive experiment, performing an n-back cognitive experiment on the subject to evaluate the mental workload and processing ability of the subject, as an auxiliary basis for judging the fatigue state; Step4: SVM fatigue detection: using the support vector machine (SVM) to construct a fatigue prediction model, taking the HRV characteristic values extracted in step B as the input, and performing binary classification prediction on the fatigue state of the subject.

[0012] In a preferred embodiment, the HRV signal acquisition in Step1 is performed once before and after fatigue induction, specifically at 10 minutes before the start of physical activity and 10 minutes after the end of physical activity. The measurement duration for each time is 30s, and each measurement is repeated 3 times to reduce errors.

[0013] In a preferred embodiment, during the HRV signal acquisition process, the subject maintains a sitting posture, relaxes the hands and arms and gently places them on the table, holds the recorder with the index finger and thumb lightly, so that the upper side of the index finger fully contacts and adheres to the electrode, and the subject remains calm and stable during the measurement process.

[0014] In a preferred embodiment, the time-domain analysis in Step2 includes the mean of normal R-R intervals, the standard deviation of normal heart beat intervals, the root mean square of the differences between adjacent R-R intervals, the standard deviation of the differences between adjacent R-R intervals, the percentage of R-R intervals greater than 50ms, and the heart rate.

[0015] In a preferred embodiment, the frequency-domain analysis in Step2 includes the very low frequency band (VLF), the low frequency band (LF), the high frequency band (HF), and the low frequency to high frequency power balance ratio LF / HF.

[0016] In a preferred embodiment, the n-back cognitive experiment in Step3 includes the letter n-back task, the stimulus is presented for 400ms, the stimulus interval is 1500ms, the participant monitors a series of random letters in the middle of the screen, and indicates whether the current letter stimulus is the same as the nth letter stimulus before by pressing a button on the keyboard. The task includes two working memory load levels (n = 1, 2), and 20 letter stimuli appear in each n-back trial.

[0017] In a preferred embodiment, in the SVM fatigue detection in Step 4, HRV eigenvalue with obvious change trend and reflecting multiple physiological characteristics is selected as the input feature of the subsequent classifier. Specifically, time domain feature SDNN and frequency domain feature LF / HF are selected as the feature inputs of the fatigue recognition model classifier.

[0018] In a preferred embodiment, the SVM fatigue detection further includes the step of optimizing the parameters of the SVM model. The optimal kernel function and penalty parameter are selected through the cross-validation method to improve the accuracy of fatigue prediction.

[0019] In a preferred embodiment, the prediction method further includes the preprocessing step of the collected electrocardiogram signal, including signal denoising, filtering and baseline correction, to improve the accuracy of HRV analysis.

[0020] In a preferred embodiment, the method further includes the step of setting a personalized fatigue threshold according to the individual differences of the subjects. According to the historical fatigue data and physiological characteristics of the subjects, the threshold of fatigue warning is dynamically adjusted to improve the pertinence and accuracy of fatigue assessment.

[0021] In a preferred embodiment, the method further includes a real-time feedback step. When it is detected that the subject is in a fatigue state, a warning signal is sent to the subject or the manager in time through a warning device, so as to take measures in time to prevent accidents caused by fatigue.

[0022] A method for detecting human body fatigue during live working based on electrocardiogram signal provided by the present invention has the following beneficial effects: 1. The present invention solves the problems of insufficient accuracy, strong subjectivity, poor real-time performance and lack of personalized evaluation existing in the traditional fatigue detection methods in the field of electric power operation and stereoscopic display technology, and overcomes the technical defects such as incomplete collection of physiological signals and insufficient model generalization ability although the prior art attempts to combine physiological signals and subjective feelings for comprehensive evaluation.

[0023] 2. Compared with the traditional subjective evaluation methods (such as self-report or questionnaire survey), the present invention realizes the scientific and objective evaluation of the fatigue state through the analysis of physiological indexes (electrocardiogram signal).

[0024] 3. The multi-dimensional evaluation method adopted by the present invention combines electrocardiogram signal and cognitive task performance to comprehensively evaluate the fatigue state from two dimensions of physiology and psychology, improving the comprehensiveness and accuracy of the evaluation.

[0025] 4. The real-time monitoring system developed by the present invention can dynamically collect and analyze the electrocardiogram signal of the operators, and timely warn of the fatigue state, which helps to reduce the risk of safety accidents caused by fatigue.

[0026] 5. The setting of the personalized fatigue threshold of the present invention takes into account individual differences, enabling the system to more accurately identify the fatigue status of each operator, avoiding misjudgment caused by a unified standard, and improving the pertinence and accuracy of fatigue assessment.

[0027] 6. Aiming at the physical fatigue problem that power operators are prone to in a high-intensity and high-risk working environment, the present invention provides an effective fatigue assessment tool, which helps managers optimize work arrangements and rest plans, and improve work safety and efficiency.

[0028] 6. The method of the present invention provides new ideas and methods for safety management and health management in the power industry, and helps to promote the sustainable development of the industry.

[0029] 7. By comprehensively considering multiple physiological signals and cognitive task performance, the present invention can more comprehensively evaluate the fatigue status of an individual, further improving the accuracy of assessment.

[0030] 8. The real-time monitoring system of the present invention can dynamically collect and analyze physiological signals, realize real-time monitoring and early warning of fatigue status, and further reduce the risk of safety accidents caused by fatigue.

[0031] 9. The present invention not only has important application value in the power industry, but also has broad application prospects in the field of stereoscopic display technology and other fields that require fatigue status assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The following further describes the present invention in conjunction with the drawings and embodiments: Figure 1 Schematic diagram of the physiological signal recognition model structure in Embodiment 1 of the present invention; Figure 2 Typical electrocardiogram signal waveform diagram of the present invention; Figure 3 Partial schematic diagram of the typical electrocardiogram signal waveform of the present invention; Figure 4 Schematic diagram of the SVM classification theory of the present invention; Figure 5 Fatigue classification process of the SVM of the present invention; Figure 6 30s electrocardiogram of the same subject in the fatigue state in Embodiment 2 of the present invention; Figure 7 30s electrocardiogram of the same subject in the non-fatigue state in Embodiment 2 of the present invention; Figure 8 Task flow of the letter n-back in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0033] The technical solutions in the present invention will be further described below in conjunction with the accompanying drawings and embodiments: Embodiment 1 As Figures 1 to 6 shown, a fatigue detection method based on multi - physiological signal fusion includes the following steps: 1. HRV signal acquisition In this embodiment, a high - precision electrocardiograph is used to collect the electrocardiogram signals of the subjects, as Figure 1 shown. Before the start of the experiment, the subjects are introduced in detail to the measurement method and precautions of the test instrument to ensure that they understand and follow the correct operation steps. During the measurement, the subjects maintain a sitting posture, gently hold the recorder with both hands, make the upper side of the index finger fully contact and closely adhere to the electrode, and keep both hands and arms relaxed and placed on the table. During the measurement process, the subjects remain calm and stable, avoiding any strenuous exercise or emotional fluctuations. Each measurement lasts for 30 seconds. To reduce errors, each subject repeats the measurement 3 times, and the data measured in the experiment, as Figure 2 and Figure 3 shown, is wirelessly transmitted to the mobile APP by the recorder for subsequent analysis. The HRV signal acquisition is performed once before and once after fatigue induction, specifically 10 minutes before the start of physical activity and 10 minutes after the end of physical activity.

[0034] 2. HRV analysis method The collected electrocardiogram signals are subjected to HRV analysis, including two methods: time - domain analysis and frequency - domain analysis.

[0035] Time - domain analysis: A series of statistical indicators are extracted from the RR sequence, including the mean of normal R - R intervals, the standard deviation of normal heartbeat intervals, the root mean square of the differences between adjacent R - R intervals, the standard deviation of the differences between adjacent R - R intervals, the percentage of R - R intervals greater than 50 ms, and the heart rate. The specific formulas are as follows: (1) Mean of normal R - R intervals : (1) In the formula, refers to the th normal R - R interval, represents the total number of normal heartbeats; (2) Standard deviation of normal heartbeat intervals : (2); (3) Root mean square of the differences between adjacent R - R intervals : (3) In the formula, refers to the th normal R - R interval; (4) Standard deviation of the difference between adjacent R-R intervals : (4); (5) Percentage of R-R intervals greater than 50 ms : (5) In the formula, represents the number of differences between adjacent R-R intervals greater than 50 ms during the entire sampling period; (6) Heart rate : (6); During the recording process, all ectopic beats (such as atrial, junctional, and ventricular premature beats, etc.) were excluded to ensure the accuracy of the data.

[0036] Frequency domain analysis: After passing the stationary electrocardiogram signal through the fast Fourier transform (FFT, Fast Fourier Transform), a power spectrum diagram was plotted with frequency (Hz) as the abscissa and power spectral density as the ordinate for analysis. The heart rate power spectrum curve of normal people under the basal state is between 0 and 0.4 Hz, where 0.003 - 0.04 Hz is the very low frequency band (VLF), 0.04 - 0.15 Hz is the low frequency band (LF), and 0.15 - 0.4 Hz is the high frequency band (HF). The balance state of the sympathetic and parasympathetic nervous systems was reflected by analyzing the balance ratios of LF, HF, and LF / HF. The reference values of each characteristic index for HRV frequency domain analysis are shown in Table 1.

[0037] Table 1 Reference values of HRV frequency domain analysis characteristic indices

[0038] 3. n-back cognitive experiment To objectively reflect the fatigue state and degree of the subjects before and after fatigue-induced activities, the n-back cognitive test was conducted using E-Prime 3.0 software. The test was carried out on a personal desktop computer, and the same version of E-Prime software was used each time to display the stimuli of the letter n-back. For each letter n-back task, the stimulus was presented for 400 ms, and the stimulus interval was 1500 ms. During the task, the participants monitored a series of random letters in the middle of the screen and indicated whether the current letter stimulus was the same as the nth letter stimulus before by pressing a button on the keyboard. The task included two working memory load levels (n = 1, 2), and 20 letter stimuli appeared in each n-back test. The subjects first performed the 1-back task and then the 2-back task, and were required to complete the key-pressing task quickly and accurately.

[0039] 4. SVM Fatigue Detection The basic steps to build a human fatigue prediction model are as follows: (1) Data collection: Use an electrocardiograph to collect the electrocardiogram signals of the subjects; (2) Feature extraction: Extract HRV features from the collected electrocardiogram signals, including time-domain features (such as SDNN) and frequency-domain features (such as LF / HF), and form a training set and a test set; (3) Model training: Use the support vector machine (SVM) algorithm, as Figure 4 and Figure 5 shown, establish a classification model through the training set. During the training process, select an appropriate kernel function and optimize the model parameters to ensure the accuracy and generalization ability of the model; (4) Model testing: Use the test set to test the established fatigue prediction model to obtain the fatigue prediction results of unknown human states.

[0040] Example 2 In another preferred embodiment, based on the above Embodiment 1, this embodiment further elaborates and refines a fatigue detection method based on multi-physiological signal fusion: (1) Real-time collect the electrocardiogram signal parameters of the operators through HRV, timely detect the abnormal physiological state of the personnel, give early warnings in advance, and avoid it from expanding into unsafe behaviors. The specific process is as follows: The operators wear a heart rate recorder with wireless data transmission function to perform live work. In this embodiment, a total of 15 subjects are selected to collect HRV signals. Each subject participating in fatigue induction collects electrocardiogram data once before and after induction, and measures 3 groups each time. Therefore, each subject is collected 2 times a day for a total of 6 groups of data. So, the electrocardiogram data in the non-fatigue state and the electrocardiogram data in the fatigue state each account for 50%. In this embodiment, a total of 240 electrocardiogram data samples in the normal state and 240 electrocardiogram data samples in the fatigue state are collected. Each sample data is a continuous electrocardiogram signal with a duration of 30s.

[0041] To facilitate the comparison and analysis of the HRV feature differences between the fatigue group and the non-fatigue group samples, the statistical analysis of the HRV time-frequency domain features of the electrocardiogram data collected in this experiment is shown in Table 2 below: Table 2 HRV Time-Frequency Domain Features of the Electrocardiogram Data Collected in the Experiment

[0042] Figure 6 and Figure 7It shows a comparison of a set of typical data measured from the same subject before and after fatigue induction. After 1 hour of physical fatigue induction, the HRV characteristic values of the electrocardiogram signals in the fatigued state and the non-fatigued state are significantly different (continuous measurement for 30 s, and the two figures have the same proportion): In the normal state, the average heart rate of the subject is 77 / min, SDNN the value of is 51.04 ms, RMSSD the value of is 31.50 ms, SDSD the value of is 18.52 ms, LF the value of is 1217.85 ms 2 , HF the value of is 578.54 ms 2 , LF / HF and is 2.11; in the fatigued state, the average heart rate of the subject is 97 / min, SDNN the value of is 16.62 ms, RMSSD the value of is 14.42 ms, the SDSD value is 7.73 ms, and the LF value is 66.54 ms 2 and the HF value is 58.25 ms 2 and the LF / HF is 1.14.

[0043] (2) Conduct an n-back cognitive experiment on the subject. As Figure 8 shown, the task process is as follows. For each letter n-back task, the stimulus is presented for 400 ms, and the stimulus interval is 1500 ms. During the task, the participant monitors a series of random letters in the middle of the screen and indicates whether the current letter stimulus is the same as the nth previous letter stimulus by pressing a button on the keyboard. If the two numbers are the same, press the 1 key; if the two letters are different, press the 2 key. The task stimuli include two working memory load levels (n = 1, 2), and 20 letter stimuli appear in each n-back trial. The subject first performs the 1-back task and then the 2-back task. The subject is required to complete the key-pressing task quickly and accurately.

[0044] After the data collection of the n-back cognitive experiment, in this embodiment, the average response time of the fatigued group and the non-fatigued group of subjects to the stimulus is statistically analyzed, and the results are shown in Table 3 below. In the table, RT-1 and RT-2 respectively represent the average response time of the subject during the 1-back and 2-back tasks when performing the cognitive experiment. It can be seen from Table 3 that after fatigue induction, when performing the 1-back and 2-back cognitive tasks, the cognitive response time of the fatigued subjects is longer. This objectively shows that the fatigue induction activity causes fatigue in the subjects, thus affecting the response time of the subjects to the cognitive tasks.

[0045] Table 3 Average response time of subjects in the fatigue group and non-fatigue group to stimuli

[0046] (3)A method for detecting human body fatigue in live working based on electrocardiogram signals, fatigue detection based on SVM. The basic steps for constructing a human body fatigue prediction model are as follows: First, use an electrocardiograph to collect the electrocardiogram signals of the subjects; then, perform electrocardiogram feature extraction, form a training set and a test set, and establish a classification model with the classifier obtained from training; finally, test the samples in the test set to obtain the fatigue prediction results of unknown human body states.

[0047] In this embodiment, it is necessary to perform binary classification on the human body fatigue and non-fatigue states. Therefore, it is necessary to find out the change trends of each HRV feature before and after the fatigue state, and select the features with obvious change trends and rich physiological characteristics as the input features of the subsequent classifier. Through the time-domain and frequency-domain analysis of the HRV signal, the available features in the time domain are obtained: RR, SDNN, SDSD, RMSSD, pNN50 and HR , as well as the available features in the frequency domain: VLF, LFHF and LF / HF . Analyze all the above HRV feature values measured from all subjects before and after fatigue induction one by one.

[0048] First, by observing the change trends, it can be found that RMSSD, SDSD, pNN50 there is no statistical difference in these three features before and after human body fatigue induction. Therefore, these three features are excluded first, while SDNN, LF, HF, LFIHF these six features all showed an obvious downward trend during exercise with the appearance of fatigue, and the HR feature showed an obvious upward trend with the appearance of fatigue; secondly, after analyzing the above 6 features respectively, it is found that in terms of time-domain features, because SDNN has obvious differences in different states, so SDNN is selected as the electrocardiogram time-domain index for fatigue recognition.

[0049] In terms of frequency-domain features, LF, HF can respectively reflect the activity strengths of the sympathetic nerve and the vagus nerve, but LF / HF can comprehensively reflect the relative strengths of the activities of the sympathetic and vagus nerves, has stronger expression ability, and can better reflect the differences in fatigue states. Therefore, LF / HF is selected as the electrocardiogram frequency-domain index for fatigue recognition; due to the large individual differences in HRV feature values during each test, in order to reduce errors, single features cannot be used as the input of the classifier alone. The HRV feature values with multiple inputs should be used to avoid the overall output error of the classifier being too large due to the excessive error of a single feature value caused by individual differences, so as to improve the accuracy of fatigue state prediction.

[0050] In summary, the time domain features of heart rate variability (HRV) are used in this embodiment. SDNN and the frequency domain features LF / HF are used as the feature inputs for the fatigue recognition model classifier.

[0051] Embodiment 3 In another preferred embodiment, based on the above Embodiments 1 and 2, to reflect the diversity and practicality of the present invention, this embodiment provides another fatigue detection method based on multi - physiological signal fusion, which is as follows: 1. Multi - physiological signal acquisition In addition to the electrocardiogram (ECG) signal, this embodiment also acquires the electro - oculogram (EOG) signal and electroencephalogram (EEG) signal of the subject. The method for acquiring the ECG signal is the same as that in Embodiments 1 and 2. The EOG signal is acquired using an electro - oculogram recorder. The subject wears electro - oculogram electrodes and the recording is carried out while keeping the eyes naturally open. The EEG signal is acquired using an electroencephalogram recorder. The subject wears an electroencephalogram cap and the recording is carried out in a quiet environment. All the acquired data are sent to the data processing center through wireless transmission.

[0052] 2. Multi - physiological signal analysis Pre - processing and feature extraction are respectively carried out on the acquired ECG signal, EOG signal and EEG signal. The method for analyzing the ECG signal is the same as that in Embodiment 1. For the EOG signal, characteristic indexes such as blink frequency and blink amplitude are mainly extracted. For the EEG signal, characteristic indexes such as alpha wave and beta wave are mainly extracted, and these indexes can reflect the activity state and fatigue degree of the brain.

[0053] 3. Comprehensive fatigue assessment The characteristic indexes of the acquired ECG signal, EOG signal and EEG signal are fused to construct a comprehensive fatigue assessment model. This model is trained using machine learning algorithms (such as random forest, neural network, etc.) and verified through a large amount of experimental data to ensure the accuracy and reliability of the model.

[0054] 4. Real - time monitoring and early warning A real - time monitoring system is developed to monitor and analyze the multi - physiological signals of the subject in real - time. When the system detects that the subject is in a fatigue state, an early warning signal is sent in time to remind the subject to pay attention to rest or adjust the working state. The early warning signal can be prompted in various ways such as sound, light, and push on a mobile device.

[0055] Compared with the closest prior art, this embodiment has the following significant differences and technological improvements: Multi - physiological signal fusion: This embodiment not only uses the ECG signal for fatigue detection, but also combines other physiological signals such as the EOG signal and EEG signal for comprehensive analysis, improving the accuracy and comprehensiveness of fatigue detection.

[0056] Advanced signal processing technology: Advanced signal processing technology is used to preprocess and extract features from the collected physiological signals, improving the signal quality and the accuracy of feature extraction.

[0057] Application of machine learning algorithms: Machine learning algorithms such as support vector machine (SVM) are used to build a fatigue prediction model, which is trained and verified through a large amount of experimental data to ensure the accuracy and generalization ability of the model.

[0058] Real-time monitoring and early warning: A real-time monitoring system is developed to achieve real-time monitoring and early warning of the fatigue state, reducing the risk of safety accidents caused by fatigue.

[0059] Example 4 In another preferred embodiment, based on the above Embodiments 1, 2, and 3, this embodiment provides a fatigue detection system for implementing a fatigue detection method based on multi-physiological signal fusion, specifically as follows: 1. System architecture The fatigue detection system of this embodiment mainly includes a data acquisition module, a signal processing module, a feature extraction module, a model prediction module, and a warning feedback module. The data acquisition module is responsible for real-time acquisition of the electrocardiogram signal and other physiological signals of the subject; the signal processing module preprocesses the acquired signals to improve the signal quality; the feature extraction module extracts physiological feature indicators related to fatigue; the model prediction module uses the trained SVM model to predict the extracted features to determine whether the subject is in a fatigue state; the warning feedback module issues a warning signal in a timely manner when a fatigue state is detected.

[0060] 2. Data acquisition and transmission A high-precision physiological signal acquisition device (such as an electrocardiogram recorder) is used to real-time acquire the electrocardiogram signal of the subject. The acquired data is sent to the data processing center for analysis and processing through a wireless transmission method (such as Bluetooth, Wi-Fi, etc.).

[0061] 3. Signal processing and feature extraction The acquired physiological signals are preprocessed, including steps such as denoising, filtering, and baseline correction. Then, physiological feature indicators related to fatigue are extracted through time-domain analysis and frequency-domain analysis methods, such as SDNN, LF, HF, and LF / HF, etc.

[0062] 4. Model prediction and warning feedback The extracted physiological feature indicators are input into the trained SVM model for prediction. When the model determines that the subject is in a fatigue state, the warning feedback module issues a warning signal in a timely manner to remind the subject to pay attention to rest or adjust the working state. The warning signal can be prompted through methods such as sound, light, or push on a mobile device.

[0063] In the preferred embodiment, in the preferred embodiment, the HRV signal is collected once before and once after fatigue induction in Step 1. Specifically, the measurement is carried out 10 minutes before the start of physical activity and 10 minutes after the end of physical activity. The duration of each measurement is 30 s, and each measurement is repeated 3 times to reduce errors. The above settings ensure the accurate capture of HRV signal data before and after the change in fatigue state. At the same time, to improve the reliability of the data, a 1-minute rest period is set between each measurement to avoid the mutual influence between consecutive measurements. All data are analyzed by taking the average value of the three measurements.

[0064] In the preferred embodiment, during the HRV signal collection process, the subject maintains a sitting position, with both hands and arms relaxed and gently placed on the table. The index finger and thumb gently hold the recorder, so that the upper side of the index finger fully contacts and adheres tightly to the electrode. During the measurement, the subject remains calm and stable. The above settings aim to minimize the influence of external interference and the subject's own physiological fluctuations on the heart rate variability (HRV) data, ensure the acquisition of accurate and reliable HRV signals, and provide a solid foundation for subsequent cardiovascular health assessment.

[0065] In the preferred embodiment, the time domain analysis in Step 2 includes the mean of normal R-R intervals, the standard deviation of normal heartbeat intervals, the root mean square of the differences between adjacent R-R intervals, the standard deviation of the differences between adjacent R-R intervals, the percentage of R-R intervals greater than 50 ms, and the heart rate. The above settings can comprehensively evaluate the stability and variability of the cardiac rhythm, provide key indicators regarding the heart health status, assist in diagnosing problems such as arrhythmia, and ensure the accuracy and clinical practicality of the analysis results.

[0066] In the preferred embodiment, the frequency domain analysis in Step 2 includes the very low frequency band (VLF), the low frequency band (LF), the high frequency band (HF), and the low frequency to high frequency power balance ratio LF / HF. The above settings can comprehensively cover different frequency components of the cardiac electrical signals. The very low frequency band reflects the cardiac autonomic regulation, the low frequency band is related to sympathetic nerve activity, the high frequency band is related to parasympathetic nerve activity, and LF / HF evaluates the balance state between the two, providing key indicators for cardiac health assessment.

[0067] In the preferred solution, the n-back cognitive experiment in Step 3 includes the letter n-back task. The stimulus is presented for 400 ms, and the stimulus interval is 1500 ms. The participant monitors a series of random letters in the middle of the screen and indicates whether the current letter stimulus is the same as the nth letter stimulus before by pressing a button on the keyboard. The task includes two working memory load levels (n = 1, 2), and 20 letter stimuli appear in each n-back trial. The above settings ensure the standardization and difficulty gradient of the experiment; n = 1 is the low load to test basic attention; n = 2 increases the difficulty to evaluate the working memory capacity; a short training is carried out before the experiment to ensure that the participant understands the rules; the reaction time and accuracy are recorded throughout the process to comprehensively evaluate the cognitive function.

[0068] In the preferred solution, in the SVM fatigue detection in Step 4, the HRV eigenvalue with an obvious change trend and reflecting multiple physiological characteristics is selected as the input feature of the subsequent classifier. Specifically, the time-domain feature SDNN and the frequency-domain feature LF / HF are selected as the feature inputs of the fatigue recognition model classifier. The above settings can significantly improve the accuracy of fatigue detection; at the same time, through the effective extraction and analysis of the SDNN and LF / HF features, the physiological signal differences in different fatigue states can be finely distinguished, providing a scientific basis for the fatigue warning system.

[0069] In the preferred solution, the SVM fatigue detection also includes the step of optimizing the parameters of the SVM model. The optimal kernel function and penalty parameter are selected through the cross-validation method to improve the accuracy of fatigue prediction. The above settings can significantly improve the generalization ability of the model and reduce the risk of overfitting; at the same time, the grid search strategy is introduced to further refine the parameter selection to ensure stable fatigue detection performance in the complex and changeable driving environment.

[0070] In the preferred solution, the prediction method also includes the preprocessing step of the collected electrocardiogram signal, including signal denoising, filtering and baseline correction, to improve the accuracy of HRV analysis. The above settings can reduce the influence of interference factors on the electrocardiogram signal, ensure the data quality, and further enhance the reliability and clinical value of the heart rate variability (HRV) analysis results, laying a solid foundation for subsequent health assessment and disease prediction.

[0071] In the preferred solution, the method also includes the step of setting a personalized fatigue threshold according to the individual differences of the subjects. According to the historical fatigue data and physiological characteristics of the subjects, the threshold of fatigue warning is dynamically adjusted to improve the pertinence and accuracy of fatigue assessment. The above settings can ensure that the fatigue monitoring system is more in line with the actual situation of each subject, effectively avoid false alarms and missed alarms, further improve the practicability and user experience of the system, and provide a strong guarantee for safety in scenarios such as long-term operation or driving.

[0072] In a preferred embodiment, the method further includes a real-time feedback step. When it is detected that the subject is in a fatigued state, a warning signal is promptly sent to the subject or the manager through a warning device, so as to take timely measures to prevent accidents caused by fatigue. With the above settings, the safety of the test process and the experience of the subject are effectively improved. At the same time, the warning signal can also trigger an automatic adjustment mechanism, such as adjusting the test rhythm or providing a short break, to ensure that the test is carried out efficiently and humanely.

[0073] In summary, the present invention proposes a method for detecting human fatigue during live working based on electrocardiogram signals. This method not only solves the problems of insufficient accuracy, strong subjectivity, poor real-time performance, and lack of personalized evaluation existing in traditional fatigue detection methods in the field of power operation and stereoscopic display technology, but also overcomes the technical defects such as incomplete collection of physiological signals and insufficient model generalization ability in the prior art although attempts have been made to combine physiological signals and subjective feelings for comprehensive evaluation. By focusing on the time-domain and frequency-domain characteristics of electrocardiogram signals and extracting multi-dimensional features, the present invention more comprehensively reflects the fatigue state of the subject, exceeding the limitations of traditional single-dimensional analysis. At the same time, this method also combines the results of the n-back cognitive experiment to evaluate the fatigue state of the subject from the psychological dimension, making the fatigue detection more comprehensive and accurate. By developing a real-time monitoring system, this method can dynamically collect and analyze the electrocardiogram signals of operators, promptly warning of the fatigue state, which is relatively advanced among existing fatigue detection methods. In addition, this method takes into account individual differences, more accurately identifies the fatigue state of each operator by setting personalized fatigue thresholds, and proposes personalized fatigue management strategies for each operator, which is innovative in the field of fatigue management. This method combines electrocardiogram signal analysis with cognitive task performance, innovatively proposes a method for comprehensively evaluating the fatigue state from both physiological and psychological dimensions, and uses the support vector machine (SVM) algorithm to construct a fatigue prediction model, improving the accuracy and efficiency of fatigue prediction, and providing new ideas and methods for fatigue management in the power industry.

Claims

1. A physical fatigue prediction method based on electrocardiogram signal analysis, characterized in that: The following steps are involved: Step 1: HRV signal acquisition: collect the subject's ECG signal and perform heart rate variability HRV analysis; Step 2: HRV analysis method, perform HRV time domain analysis and frequency domain analysis on the collected ECG signals to extract HRV feature values; Step 3: n-back cognitive experiment: The subjects were subjected to n-back cognitive experiment to evaluate their mental workload and processing ability as an auxiliary basis for judging fatigue status; Step 4: SVM fatigue detection: Use support vector machine (SVM) to build a fatigue prediction model, take the HRV feature value extracted in step B as input, and perform binary classification prediction on the fatigue state of the subject.

2. The method for predicting physical fatigue based on electrocardiogram signal analysis according to claim 1, characterized in that: In the Step 1, HRV signal collection is performed once before and after fatigue induction, specifically 10 minutes before the start of physical activity and 10 minutes after the end of physical activity. Each measurement lasts 30 seconds, and each measurement is repeated 3 times to reduce errors.

3. The method for predicting physical fatigue based on electrocardiogram signal analysis according to claim 2, characterized in that: During the HRV signal collection process, the subject remains in a sitting position, with his hands and arms relaxed and placed lightly on the table, and his index finger and thumb lightly hold the recorder so that the upper side of the index finger fully contacts and is close to the electrode. During the measurement process, the subject remains calm and stable.

4. The method for predicting physical fatigue based on electrocardiogram signal analysis according to claim 3, characterized in that: The Step 2 time domain analysis includes the mean of normal RR intervals, the standard deviation of normal heartbeat intervals, the root mean square of the difference between adjacent RR intervals, the standard deviation of the difference between adjacent RR intervals, the percentage of RR intervals greater than 50ms and the heart rate.

5. The method for predicting physical fatigue based on electrocardiogram signal analysis according to claim 4, characterized in that: The frequency domain analysis in Step 2 includes a very low frequency band VLF, a low frequency band LF, a high frequency band HF, and a low frequency and high frequency power balance ratio LF / HF.

6. The method for predicting physical fatigue based on electrocardiogram signal analysis according to claim 5, characterized in that: The n-back recognition experiment in Step 3 includes a letter n-back task, in which the stimulus is presented for 400 ms and the stimulus interval is 1500 ms. The participants monitor a series of random letters in the middle of the screen and indicate whether the current letter stimulus is the same as the previous nth letter stimulus by pressing a button on the keyboard. The task includes two working memory load levels, and 20 letter stimuli appear in each n-back trial.

7. The method for predicting physical fatigue based on electrocardiogram signal analysis according to claim 6, characterized in that: In the SVM fatigue detection in Step 4, HRV feature values ​​with obvious change trends and more physiological characteristics are selected as input features of the subsequent classifier. Specifically, time domain features SDNN and frequency domain features LF / HF are selected as feature inputs of the fatigue recognition model classifier.

8. The method for predicting physical fatigue based on electrocardiogram signal analysis according to claim 7, characterized in that: The SVM fatigue detection also includes a step of optimizing the parameters of the SVM model, and selecting the optimal kernel function and penalty parameters through a cross-validation method to improve the accuracy of fatigue prediction.

9. The method for predicting physical fatigue based on electrocardiogram signal analysis according to claim 8, characterized in that: The prediction method also includes a preprocessing step for the collected ECG signal, including signal denoising, filtering and baseline correction, so as to improve the accuracy of HRV analysis.

10. The method for predicting physical fatigue based on electrocardiogram signal analysis according to claim 9, characterized in that: The method also includes a real-time feedback step. When it is detected that the subject is in a fatigue state, a warning signal is promptly sent to the subject or the manager through the warning device so that measures can be taken in time to prevent accidents caused by fatigue.

Images