A cluster operator fatigue prediction method and system

Abstract

This invention relates to a method and system for predicting fatigue in swarm operators. The method includes the following steps: extracting physiological characteristic data of multiple subjects in non-fatigue and fatigued states when performing swarm operation tasks of different scales; using the fatigue and non-fatigue states corresponding to the physiological characteristic data as labels; training a classification model based on the extracted physiological characteristic data and corresponding labels to obtain a fatigue detection model; calculating the fatigue arrival time for each subject in each scale of swarm operation task based on the fatigue detection model; determining the optimal scale of the swarm operation task based on the fatigue arrival time for each subject in each scale of swarm operation task; the operator performing the optimal scale swarm operation task; collecting physiological characteristic data of the operator at each moment during the swarm operation task; and predicting the operator's current and future state based on the collected physiological characteristic data and the fatigue detection model.

Description

Technical Field

[0001] This invention relates to the field of fatigue prediction technology, and in particular to a method and system for predicting fatigue among cluster operators. Background Technology

[0002] Current methods for detecting fatigue in group control systems can be divided into two types: subjective definition and objective assessment. Subjective methods primarily include the Eysenck Personality Questionnaire (EPQ), the Simplified Coping Style Questionnaire, the Symptom Checklist-90 (SCL-90), the Self-Rating Psychological Scale (SAS), and the Human Depression Scale (HAMD). Objective assessment mainly involves methods based on facial expressions, physiological characteristics, and behavioral features. Through signal acquisition, analysis, feature extraction, and classification, fatigue monitoring based on a single feature is ultimately achieved. Subjective definition can reflect the causes of changes in condition to some extent, but it is easily influenced by subjective factors.

[0003] Current research on objective assessment mainly focuses on detecting fatigue levels. Furthermore, there is a lack of quantitative research on the number of clusters that operators can control. The evaluation and empowerment research of cluster control systems are carried out under their respective proposed task backgrounds, resulting in a variety of cluster sizes controlled by operators. Human-machine ergonomics evaluation lacks a unified task background, and the evaluation results also lack guiding significance. Summary of the Invention

[0004] Based on the above analysis, the present invention aims to provide a method and system for predicting fatigue of cluster operators, in order to solve the problems of insufficient quantitative research on the number of clusters that operators can control and inaccurate fatigue detection.

[0005] On one hand, embodiments of the present invention provide a method for predicting fatigue among cluster operators, comprising the following steps:

[0006] Physiological feature data of non-fatigue and fatigue states of multiple subjects performing cluster manipulation tasks of different scales were extracted. The fatigue and non-fatigue states corresponding to the physiological feature data were used as labels. A fatigue detection model was trained based on the extracted physiological feature data and the corresponding labels.

[0007] The fatigue detection model was used to calculate the arrival fatigue time for each subject under each scale of cluster manipulation task; the optimal scale of the cluster manipulation task was determined based on the arrival fatigue time for each subject under each scale of cluster manipulation task.

[0008] Operators execute optimal-scale cluster control tasks, and physiological characteristic data of operators are collected at each moment during the execution of cluster control tasks; based on the collected physiological characteristic data, fatigue detection models are used to predict the current and future states of operators.

[0009] Based on further improvements to the above technical solution, the fatigue detection model is used to calculate the fatigue time for each subject under each scale of cluster manipulation task, including:

[0010] Physiological characteristic data of each subject were extracted for each unit time between the non-fatigue state and the fatigue state under each scale of cluster control task. Based on the fatigue detection model, the state of each subject was determined for each unit time between the non-fatigue state and the fatigue state under each scale of cluster control task.

[0011] For each subject, the time between the non-fatigue state and the fatigue state determined by the fatigue detection model from the first fatigue state to the initial time of the cluster control task is the fatigue time for that subject under that scale of cluster control task.

[0012] Furthermore, the optimal size of the cluster manipulation task was determined based on the arrival fatigue time for each subject under each scale of the cluster manipulation task, including:

[0013] The optimal task size for each subject is calculated based on the time to fatigue under each size of cluster manipulation task.

[0014] The optimal task size for cluster manipulation tasks is the one with the most optimal task sizes among all subjects.

[0015] Furthermore, based on the arrival fatigue time for each subject under each scale of cluster manipulation task, the optimal task scale for each subject was calculated, including:

[0016] The minimum time to fatigue for the current subject under different scale cluster control tasks is the shortest evaluation time; each moment when the difference between the subject and the start time of each cluster control task is less than the shortest evaluation time is the evaluation time of each cluster control task.

[0017] For each size of cluster control task, calculate the fatigue rate of controlling one machine at each evaluation time. The size of the cluster control task corresponding to the minimum fatigue rate is the optimal size for each evaluation time.

[0018] The task size with the most optimal sizes across all evaluation times is the optimal task size for the current subject.

[0019] Furthermore, for cluster control tasks of each size, the fatigue rate of controlling one machine at each evaluation time is calculated, including:

[0020] According to the formula Among them, P iN represents the fatigue probability value determined by the fatigue detection model at the i-th evaluation time; j T represents the number of machines performing cluster control tasks of size j; i This represents the fatigue rate of operating a machine at the i-th evaluation time.

[0021] Furthermore, based on the collected physiological characteristic data and a fatigue detection model, the current and future states of the operator are predicted, including:

[0022] Obtain the physiological characteristic data at the current moment, and predict the fatigue probability value of the operator at the current moment based on the fatigue detection model to obtain the operator's current state;

[0023] The fatigue probability value at the current moment and the fatigue probability values ​​at multiple moments prior to the current moment are used to calculate the fatigue probability value at the future moment, thus determining the operator's state at the future moment.

[0024] Furthermore, the fatigue probability value at future moments is calculated using the following formula:

[0025]

[0026]

[0027] Where t is the current time, p t+k This represents the predicted fatigue probability value at time t+k. α represents the fatigue probability value at time t+k predicted based on time ti. t-i The weight at time ti represents the weight of the s time steps before time t. t-i This represents the fatigue probability value that the fatigue detection model determines for the operator at time ti.

[0028] Furthermore, the physiological characteristic data includes electroencephalogram (EEG) characteristic data, electrocardiogram (ECG) characteristic data, and electromyogram (EMG) characteristic data;

[0029] The EEG feature data includes the mean, standard deviation, kurtosis, skewness, entropy of each EEG channel, and mutual information between EEG channels per unit time.

[0030] The electrocardiogram characteristic data includes heart rate per unit time, standard deviation of NN interval, root mean square of the difference between adjacent NN intervals, low frequency band area, high frequency band area, and low frequency to high frequency power ratio.

[0031] The electromyographic characteristic data includes the mean and variance of electromyography per unit time.

[0032] Furthermore, after extracting physiological feature data of multiple subjects in non-fatigue and fatigued states when performing cluster manipulation tasks of different scales, before training a classification model based on the extracted physiological feature data and corresponding labels to obtain a fatigue detection model, the following steps are also included:

[0033] Calculate the Pearson correlation coefficient between each physiological characteristic and its corresponding label;

[0034] A statistic was constructed based on the Pearson correlation coefficient. The p-value corresponding to each feature is retrieved based on the t-value. Based on the p-value, physiological feature data with high correlation to the label are selected for classification model training to obtain the fatigue detection model. Here, r represents the Pearson correlation coefficient and n represents the number of samples for the current feature.

[0035] On the other hand, embodiments of the present invention provide a fatigue prediction system for cluster operators, comprising the following modules:

[0036] The model training module is used to extract physiological feature data of multiple subjects in non-fatigue and fatigue states when performing cluster manipulation tasks of different scales. The fatigue and non-fatigue states corresponding to the physiological feature data are used as labels. Based on the extracted physiological feature data and the corresponding labels, a classification model is trained to obtain a fatigue detection model.

[0037] The optimal scale determination module is used to calculate the fatigue time for each subject under each scale of cluster manipulation task based on the fatigue detection model; and to determine the optimal scale of the cluster manipulation task based on the fatigue time for each subject under each scale of cluster manipulation task.

[0038] The state prediction module is used to control personnel to perform cluster control tasks of optimal scale. It collects physiological characteristic data of the personnel at each moment when performing cluster control tasks. Based on the collected physiological characteristic data, it predicts the state of the personnel at the current moment and the future moment using a fatigue detection model.

[0039] Compared with existing technologies, this invention calculates the fatigue time for each subject under each scale of cluster manipulation task based on a fatigue detection model; and determines the optimal scale of the cluster manipulation task based on the fatigue time for each subject under each scale of cluster manipulation task. This allows for quantitative research on the number of clusters that operators can control, maximizing the operator's capabilities and providing a theoretical scientific basis for subsequent empowerment technologies. By predicting the operator's state at future moments based on collected physiological characteristic data using a fatigue detection model, and by comprehensively considering current and past fatigue states, the degree of fatigue at future moments can be predicted more accurately.

[0040] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0041] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0042] Figure 1 This is a flowchart of a method for predicting fatigue among cluster operators according to an embodiment of the present invention;

[0043] Figure 2 This is a block diagram of the fatigue prediction system for cluster operators according to an embodiment of the present invention. Detailed Implementation

[0044] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0045] A specific embodiment of the present invention discloses a method for predicting fatigue among cluster operators, such as... Figure 1 As shown, it includes the following steps:

[0046] S1. Extract physiological feature data of multiple subjects in non-fatigue and fatigue states when performing cluster manipulation tasks of different scales. Use the fatigue and non-fatigue states corresponding to the physiological feature data as labels. Train a classification model based on the extracted physiological feature data and the corresponding labels to obtain a fatigue detection model.

[0047] S2. Calculate the fatigue time for each subject under each scale of cluster control task based on the fatigue detection model; determine the optimal scale of cluster control task based on the fatigue time for each subject under each scale of cluster control task.

[0048] S3. The operator performs a cluster control task of optimal scale, and collects physiological characteristic data of the operator at each moment when performing the cluster control task; based on the collected physiological characteristic data, the operator's current and future state is predicted based on a fatigue detection model.

[0049] This invention calculates the time to fatigue for each subject under each scale of cluster manipulation task based on a fatigue detection model; and determines the optimal scale of the cluster manipulation task based on the time to fatigue for each subject under each scale of cluster manipulation task. This allows for a quantitative study of the number of clusters that operators can control, maximizing the operator's capabilities and providing a theoretical scientific basis for subsequent empowerment technologies. By predicting the operator's state at future moments based on collected physiological characteristic data using a fatigue detection model, and by comprehensively considering current and past fatigue states, the degree of fatigue at future moments can be predicted more accurately.

[0050] During implementation, a well-designed experimental paradigm is crucial. Subjects should begin completing swarm manipulation tasks of varying sizes while in a mentally (i.e., non-fatigued) state. The experimental duration must ensure that subjects remain fatigued for at least a period before the experiment ends. For example, in an experiment lasting 3-4 hours, subjects should be fatigued for the last half hour. All physiological data should be collected from the start to the end of the experiment. Since subjects begin the experiment in a non-fatigued state, they will certainly be in a non-fatigued state for a period at the beginning. Therefore, physiological data from the period after the start of the experiment are considered as data from the non-fatigued state, while physiological data from the period before the end of the experiment are considered as data from the fatigued state. This ensures the accuracy of the subsequent fatigue detection model.

[0051] The corresponding physiological characteristic data is obtained based on the physiological information data. Specifically, the physiological information data includes electroencephalogram (EEG), electrocardiogram (ECG), and electromyogram (EMG) data, and the physiological characteristic data includes EEG characteristic data, ECG characteristic data, and EMG characteristic data.

[0052] Specifically, EEG characteristic data includes the mean, standard deviation, kurtosis, skewness, entropy of each EEG channel, and mutual information between EEG channels within a unit of time. For example, if the unit of time is one minute, then the mean, standard deviation, kurtosis, skewness, entropy of each EEG channel, and mutual information between EEG channels are calculated for each minute.

[0053] Among them, according to Calculate the kurtosis K of the data per unit time, where n represents the number of EEG data points per unit time, x i σ represents the i-th EEG data point within a unit of time, μ represents the mean of the data within a unit of time, and σ represents the variance of the data within a unit of time. Kurtosis is used to measure the steepness of the probability distribution of a random variable; the larger the kurtosis, the steeper and more angular the distribution, and vice versa.

[0054] according to Calculate the data skewness S per unit time, where n represents the number of EEG data points per unit time, and x... iσ represents the i-th data point within a unit of time, μ represents the mean of the data within a unit of time, σ represents the variance of the data within a unit of time, and skewness is used to measure the asymmetry of the probability distribution of a random variable. Skewness > 0 indicates that the probability distribution graph is right-skewed, and vice versa.

[0055] Entropy represents the amount of information contained in a thing / event. The greater the uncertainty of an event / event, the greater the amount of information, and the greater the entropy; conversely, the smaller the uncertainty of an event / event, the smaller the amount of information, and the smaller the entropy. For continuous time series such as EEG signals, the amplitude of the signal is first segmented. For example, if an EEG signal has 10,000 data points with an amplitude range of -0.2V to 0.2V, it can be divided into 100 segments according to amplitude. The amplitude range of each segment is [-0.2, -0.196], [-0.196, -0.192], ..., [0.196, 0.2]. The number of data points whose EEG signal amplitude falls within each of these segments is counted and divided by the total number of EEG signal data points (here, 10,000) to obtain the probability of each segment.

[0056] The formula for calculating entropy is as follows:

[0057]

[0058] Where H represents the entropy of a continuous time-series signal of a certain EEG channel; in the above formula, m represents the number of segments, and z i Let p(z) represent the i-th segment. i ) represents the probability of the i-th segment.

[0059] Mutual information between two random variables is a measure of the interdependence between the variables.

[0060] The mutual information I(X1; X2) between any two EEG channels can be calculated using the following formula:

[0061]

[0062] Where X1 and X2 represent the EEG data of the two channels, p(x1) and p(x2) represent the marginal probabilities of x1 and x2 respectively, and p(x1,x2) represents the joint probability of x1 and x2.

[0063] Specifically, ECG characteristic data include heart rate per unit time, standard deviation of NN interval, root mean square of the difference between adjacent NN intervals, low-frequency band area, high-frequency band area, and low-frequency to high-frequency power ratio.

[0064] Here, the NN interval represents the interval between normal R peaks. The standard deviation of the NN interval is the standard deviation of all NN intervals per unit time.

[0065] Calculate the difference between adjacent NN intervals within a unit time, and then calculate the root mean square of the difference between adjacent NN intervals within a unit time.

[0066] Specifically, low frequency refers to frequencies in the range of 0.04 to 0.15 Hz; high frequency refers to frequencies in the range of 0.15 to 0.4 Hz.

[0067] It should be noted that the bandwidth area refers to the bandwidth area in the power spectral density curve.

[0068] The electromyographic characteristic data includes the mean and variance of electromyography per unit time.

[0069] After extracting physiological feature data, the feature data and the corresponding labels (i.e., fatigue or non-fatigue, with a value of 1 for fatigue and 0 for non-fatigue) are used to form training data for training the classification model, thus obtaining the fatigue detection model.

[0070] To reduce computational load and improve model accuracy, correlation analysis is performed on features, and highly correlated features are extracted for classification model training. Therefore, after extracting physiological feature data from multiple subjects performing cluster manipulation tasks of varying scales in both non-fatigue and fatigue states, before training a classification model based on the extracted physiological feature data and corresponding labels to obtain the fatigue detection model, the following steps are also included:

[0071] Calculate the Pearson correlation coefficient between each physiological characteristic and its corresponding label;

[0072] A statistic was constructed based on the Pearson correlation coefficient. The p-value corresponding to each feature is retrieved based on the t-value. Based on the p-value, physiological feature data with high correlation to the label are selected for classification model training to obtain the fatigue detection model. Here, r represents the Pearson correlation coefficient and n represents the number of samples for the current feature.

[0073] Specifically, through the formula Calculate the Pearson correlation coefficient r for each physiological characteristic data point, where n represents the number of current physiological characteristic data points; x i This represents the i-th data point representing the current physiological characteristic. y represents the mean of the current physiological characteristic data. i This represents the label corresponding to the i-th data point of the current physiological characteristic. This represents the mean of the labels corresponding to the current physiological features. A t-statistic is constructed for t-testing, and the corresponding p-value is obtained by looking up the t-value in a table. The smaller the p-value, the higher the correlation. Physiological feature data with high correlation to the labels are selected for training the classification model to obtain the fatigue detection model. For example, physiological feature data with p < 0.01 are selected for model training.

[0074] In implementation, EEG, ECG, and EMG feature data are used to train different classifier models. An ensemble classification system is then used to fuse the decisions of these three classifier models to obtain the final fatigue detection model. During implementation, the following methods can be employed: Calculate the final output of the fatigue detection model, where, and These represent the predicted values ​​of the three classifier models, with α, β, and γ representing the weights of each model. The weights can be obtained based on the training accuracy of each classification model.

[0075] After obtaining the fatigue detection model, the fatigue time for each subject under each scale of cluster manipulation task is calculated based on the fatigue detection model. Specifically, step S2 includes:

[0076] S211. Extract the physiological characteristic data of each subject from the non-fatigue state to the fatigue state at each unit time under each scale of cluster control task, and determine the state of each subject from the non-fatigue state to the fatigue state at each unit time under each scale of cluster control task based on the fatigue detection model.

[0077] S212. The time between the non-fatigue state and the fatigue state determined by the fatigue detection model for each subject under each scale of cluster control task, from the time corresponding to the first fatigue state to the initial time of the cluster control task, is the fatigue time for that subject under that scale of cluster control task.

[0078] The fatigue detection model is trained using data from the beginning and end of the experiment. For data in the middle of the experiment, fatigue detection is performed based on the model to obtain the state at each unit time. Following the same method as in step S1, physiological characteristic data for each subject is extracted for each time step between the non-fatigue state and the fatigue state under each scale of the cluster manipulation task. This data is then input into the fatigue detection model to obtain the predicted fatigue value for each subject at each unit time step between the non-fatigue state and the fatigue state under each scale of the cluster manipulation task, which is a number between 0 and 1. The corresponding state can be obtained based on the predicted fatigue value. For example, if the fatigue threshold is set to 0.6, a predicted value greater than or equal to 0.6 indicates a fatigue state, and a value less than 0.6 indicates a non-fatigue state.

[0079] At this point, the fatigue status of each subject at every moment of the entire experimental process under each scale of cluster manipulation task has been obtained.

[0080] The time from the moment corresponding to the first fatigue state between the non-fatigue state and the fatigue state determined by the fatigue detection model to the initial moment of the cluster control task is the fatigue time for the subject under this scale of cluster control task.

[0081] The optimal size of the cluster manipulation task can be determined based on the time to reach fatigue for each subject under each scale of the cluster manipulation task. Specifically, step S2 includes:

[0082] S221. Calculate the optimal task size for each subject based on the arrival fatigue time of each subject under each scale of cluster manipulation task.

[0083] S222. The optimal size for a cluster manipulation task is the size with the largest number of optimal task sizes among all subjects.

[0084] In specific step S221, the optimal task size for each subject is calculated based on the arrival fatigue time for each subject under each scale of cluster manipulation task, including:

[0085] S2211. The minimum time to fatigue for the current subject under different scale cluster control tasks is the shortest evaluation time; each moment when the difference between the subject and the start time of each cluster control task is less than the shortest evaluation time is the evaluation time of each cluster control task.

[0086] For example, the shortest evaluation time is t. min Then 0~t min Each moment in time is an evaluation moment. If the unit of time is one minute, then each minute starting from 0 is a moment.

[0087] S2212. For each size of cluster control task, calculate the fatigue rate of controlling one machine at each evaluation time. The size of the cluster control task corresponding to the minimum fatigue rate is the optimal size corresponding to each evaluation time.

[0088] Specifically, according to the formula Among them, P i N represents the fatigue probability value determined by the fatigue detection model at the i-th evaluation time; j T represents the number of machines performing cluster control tasks of size j; i This represents the fatigue rate of operating a machine at the i-th evaluation time.

[0089] The lower the fatigue probability value, the less likely fatigue will occur under cluster control tasks of this scale.

[0090] S2213. The task size with the most optimal sizes among all evaluation times is the optimal task size for the current subject.

[0091] For example, there are three task sizes, with the number of machines controlled being N1, N2, and N3, respectively. There are 100 evaluation times, of which the optimal size is N2 for 60 evaluation times, N1 for 25 evaluation times, and N3 for 15 evaluation times. Therefore, the optimal task size for the current subject is N2.

[0092] If there are a total of 50 subjects, and the optimal size is N1 for 36 subjects, then the optimal task size for the cluster control task is N1.

[0093] By determining the optimal task size for cluster control tasks, operators can control the cluster at the optimal size when executing tasks.

[0094] The operator performs a cluster control task of optimal task size, and collects physiological characteristic data of the operator at each moment when performing the cluster control task. For details on obtaining the physiological characteristic data, please refer to the method in step S1, which will not be repeated here.

[0095] Specifically, step S3 involves predicting the operator's current and future state based on the collected physiological characteristic data and a fatigue detection model, including:

[0096] S31. Obtain the physiological characteristic data at the current moment, and predict the fatigue probability value of the operator at the current moment based on the fatigue detection model to obtain the operator's current state.

[0097] Physiological characteristic data are input into the fatigue detection model to obtain a fatigue probability value, which is a number between 0 and 1. The higher the fatigue probability value, the more likely the person is to be in a state of fatigue.

[0098] S32. Calculate the fatigue probability value of the future moment based on the fatigue probability value of the current moment and the fatigue probability values ​​of multiple moments before the current moment to obtain the operator's state at the future moment.

[0099] For example, let t represent the current time. After obtaining the fatigue probability value at the current time, calculate the fatigue probability value at future times based on the fatigue probability value at the current time and the fatigue probability values ​​at times before the current time.

[0100] Specifically, the fatigue probability value at future moments is calculated using the following formula:

[0101]

[0102]

[0103] Where p is the current time. t+k This represents the predicted fatigue probability value at time t+k. α represents the fatigue probability value at time t+k predicted based on time ti.t-i The weight at time ti represents the weight of the s time steps before time t. t-i This represents the fatigue probability value that the fatigue detection model determines for the operator at time ti.

[0104] That is, the fatigue probability value at the next time step (t+k) is predicted by using the difference between the current time and the adjacent times before the current time as the slope. In practice, s can be 10, that is, the fatigue probability value of the operator at the next time step is calculated based on the fatigue probability values ​​of the current time and the 10 times before the current time.

[0105] The corresponding state can be obtained based on the fatigue prediction value. For example, if the fatigue threshold is set to 0.6, then a predicted value greater than or equal to 0.6 is a fatigue state, and a predicted value less than 0.6 is a non-fatigue state.

[0106] Empower the operator based on their current and future state.

[0107] In implementation, the fatigue probability value at different future times can be predicted, thus more accurately determining the operator's state. For example, predicting the state at time t+k1 and t+k2, where k1 < k2, the operator's empowerment intensity is divided into four types based on the state at time t, t+k1, and t+k2: no empowerment required, empowerment intensity 1, empowerment intensity 2, and empowerment intensity 3. No empowerment required means the operator will not reach a fatigue state within the next k2 time period. Empowerment intensity 1 means the operator will not be fatigued within the next k1 time period, but will become fatigued within the next k2 time period. Empowerment intensity 2 means the operator will become fatigued within the next k1 time period. Empowerment intensity 3 means the operator is currently fatigued.

[0108] Different empowerment methods are adopted for different empowerment intensities, as shown below: (1) The empowerment method corresponding to empowerment intensity 1 is to use only cold air blowing and olfactory stimulation, with an empowerment time of k1. (2) The empowerment method corresponding to empowerment intensity 2 is to use only cold air blowing and olfactory stimulation, with an empowerment time of k2. (3) The empowerment method corresponding to empowerment intensity 3 is to use cold air blowing, olfactory stimulation and electrical stimulation simultaneously, with an empowerment time of k2. This allows the controller to remain awake for a long time and better complete the cluster control task. It further increases the controller's control time and work performance.

[0109] A specific embodiment of the present invention discloses a fatigue prediction system for cluster operators, such as... Figure 2 As shown, it includes the following modules:

[0110] The model training module is used to extract physiological feature data of multiple subjects in non-fatigue and fatigue states when performing cluster manipulation tasks of different scales. The fatigue and non-fatigue states corresponding to the physiological feature data are used as labels. Based on the extracted physiological feature data and the corresponding labels, a classification model is trained to obtain a fatigue detection model.

[0111] The optimal scale determination module is used to calculate the fatigue time for each subject under each scale of cluster manipulation task based on the fatigue detection model; and to determine the optimal scale of the cluster manipulation task based on the fatigue time for each subject under each scale of cluster manipulation task.

[0112] The state prediction module is used to control personnel to perform cluster control tasks of optimal scale. It collects physiological characteristic data of the personnel at each moment when performing cluster control tasks. Based on the collected physiological characteristic data, it predicts the state of the personnel at the current moment and the future moment using a fatigue detection model.

[0113] The above-described method and system embodiments are based on the same principles, and their related aspects can be referenced from each other to achieve the same technical effects. For specific implementation processes, please refer to the foregoing embodiments, which will not be repeated here.

[0114] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0115] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting fatigue among cluster operators, characterized in that, Includes the following steps: Physiological feature data of non-fatigue and fatigue states of multiple subjects performing cluster manipulation tasks of different scales were extracted. The fatigue and non-fatigue states corresponding to the physiological feature data were used as labels. A fatigue detection model was trained based on the extracted physiological feature data and the corresponding labels. The fatigue detection model was used to calculate the fatigue time for each subject under each size of cluster manipulation task. The optimal size of the cluster control task was determined based on the arrival fatigue time of each subject under each size of the cluster control task. Operators execute cluster control tasks of optimal scale, and collect physiological characteristic data of operators at every moment when executing cluster control tasks; Based on the collected physiological characteristic data, a fatigue detection model is used to predict the operator's current and future state. The physiological characteristic data includes electroencephalogram (EEG) characteristic data, electrocardiogram (ECG) characteristic data, and electromyogram (EMG) characteristic data; Electroencephalogram (EEG), electrocardiogram (ECG), and electromyogram (EMG) feature data were used to train different classifier models. An ensemble classification method was then used to fuse the three classifier models to obtain the final fatigue detection model. The optimal size of the cluster manipulation task was determined based on the arrival fatigue time for each subject under each size of the cluster manipulation task, including: The optimal task size for each subject is calculated based on the time to fatigue under each size of cluster manipulation task. The optimal task size for cluster manipulation tasks is the one with the most optimal task sizes among all subjects. The optimal task size for each subject was calculated based on the arrival fatigue time for each subject under each size of cluster manipulation task, including: The minimum time to fatigue for the current subject under different scale cluster control tasks is the shortest evaluation time; each moment when the difference between the subject and the start time of each cluster control task is less than the shortest evaluation time is the evaluation time of each cluster control task. For each size of cluster control task, calculate the fatigue rate of controlling one machine at each evaluation time. The size of the cluster control task corresponding to the minimum fatigue rate is the optimal size for each evaluation time. The task size with the most optimal sizes across all evaluation times is the optimal task size for the current subject.

2. The fatigue prediction method for cluster operators according to claim 1, characterized in that, The fatigue detection model was used to calculate the time to fatigue for each subject under each size of cluster manipulation task, including: Physiological characteristic data of each subject were extracted for each unit time between the non-fatigue state and the fatigue state under each scale of cluster control task. Based on the fatigue detection model, the state of each subject was determined for each unit time between the non-fatigue state and the fatigue state under each scale of cluster control task. For each subject, the time between the non-fatigue state and the fatigue state determined by the fatigue detection model from the first fatigue state to the initial time of the cluster control task is the fatigue time for that subject under that scale of cluster control task.

3. The fatigue prediction method for cluster operators according to claim 1, characterized in that, For cluster control tasks of each size, calculate the fatigue rate of controlling one machine at each evaluation time, including: According to the formula ,in, This represents the fatigue probability value determined by the fatigue detection model at the i-th evaluation time. This represents the number of machines performing cluster control tasks of the j-th size. This represents the fatigue rate of operating a machine at the i-th evaluation time.

4. The fatigue prediction method for cluster operators according to claim 1, characterized in that, Based on the collected physiological characteristic data, a fatigue detection model is used to predict the operator's current and future state, including: Obtain the physiological characteristic data at the current moment, and predict the fatigue probability value of the operator at the current moment based on the fatigue detection model to obtain the operator's current state; The fatigue probability value at the current moment and the fatigue probability values ​​at multiple moments prior to the current moment are used to calculate the fatigue probability value at the future moment, thus determining the operator's state at the future moment.

5. The fatigue prediction method for cluster operators according to claim 4, characterized in that, The fatigue probability value at future moments is calculated using the following formula: in, For the current moment, Indicates prediction The probability of fatigue at any given moment. Indicates according to Time prediction The probability of fatigue at any given moment. express Weight of time, Indicates taking the time before time t At that moment, express The constant fatigue detection model determines the probability of operator fatigue.

6. The fatigue prediction method for cluster operators according to claim 1, characterized in that, The EEG feature data includes the mean, standard deviation, kurtosis, skewness, entropy of each EEG channel, and mutual information between EEG channels per unit time. The electrocardiogram characteristic data includes heart rate per unit time, standard deviation of NN interval, root mean square of the difference between adjacent NN intervals, low frequency band area, high frequency band area, and low frequency to high frequency power ratio. The electromyographic characteristic data includes the mean and variance of electromyography per unit time.

7. The fatigue prediction method for cluster operators according to claim 6, characterized in that, After extracting physiological feature data of multiple subjects in non-fatigue and fatigued states when performing cluster manipulation tasks of different scales, and before training a classification model based on the extracted physiological feature data and corresponding labels to obtain a fatigue detection model, the following steps are also included: Calculate the Pearson correlation coefficient between each physiological characteristic and its corresponding label; A statistic was constructed based on the Pearson correlation coefficient. The fatigue detection model is obtained by querying the p-value corresponding to each feature based on the t-value, selecting physiological feature data with high correlation to the label based on the p-value, and training the classification model. represents the Pearson correlation coefficient, and n represents the number of samples for the current feature.

8. A fatigue prediction system for cluster operators, characterized in that, Includes the following modules: The model training module is used to extract physiological feature data of multiple subjects in non-fatigue and fatigue states when performing cluster manipulation tasks of different scales. The fatigue and non-fatigue states corresponding to the physiological feature data are used as labels. Based on the extracted physiological feature data and the corresponding labels, a classification model is trained to obtain a fatigue detection model. The optimal scale determination module is used to calculate the fatigue time for each subject under each scale of cluster manipulation task based on the fatigue detection model. The optimal size of the cluster control task was determined based on the arrival fatigue time of each subject under each size of the cluster control task. The state prediction module is used to collect physiological characteristic data of operators at each moment when they perform cluster control tasks of optimal scale. Based on the collected physiological characteristic data, a fatigue detection model is used to predict the operator's current and future state. The physiological characteristic data includes electroencephalogram (EEG) characteristic data, electrocardiogram (ECG) characteristic data, and electromyogram (EMG) characteristic data; Electroencephalogram (EEG), electrocardiogram (ECG), and electromyogram (EMG) feature data were used to train different classifier models. An ensemble classification method was then used to fuse the three classifier models to obtain the final fatigue detection model. The optimal size of the cluster manipulation task was determined based on the arrival fatigue time for each subject under each size of the cluster manipulation task, including: The optimal task size for each subject is calculated based on the time to fatigue under each size of cluster manipulation task. The optimal task size for cluster manipulation tasks is the one with the most optimal task sizes among all subjects. The optimal task size for each subject was calculated based on the arrival fatigue time for each subject under each size of cluster manipulation task, including: The minimum time to fatigue for the current subject under different scale cluster control tasks is the shortest evaluation time; each moment when the difference between the subject and the start time of each cluster control task is less than the shortest evaluation time is the evaluation time of each cluster control task. For each size of cluster control task, calculate the fatigue rate of controlling one machine at each evaluation time. The size of the cluster control task corresponding to the minimum fatigue rate is the optimal size for each evaluation time. The task size with the most optimal sizes across all evaluation times is the optimal task size for the current subject.

Images