Prediction Of Fatigue Of High-Speed Railway Dispatchers Based On EEG Time Series Data

The fatigue operation of high-speed railway dispatchers seriously threatens the safety of railway traffic.The monitoring and early warning of the fatigue of dispatchers is an important topic of high-speed railway driving safety.Due to the low task density and long working hours of high-speed rail dispatching work,it is necessary to conduct fatigue prediction research on the work characteristics of dispatchers.To improve the timeliness of fatigue warning of highspeed railway dispatchers and predict the fatigue state of high-speed railway dispatchers,this thesis builds a high-speed railway dispatcher fatigue prediction method based on EEG time series signals.By analyzing the characteristics of high-speed train dispatching operation,a dispatching experiment and an EEG acquisition and analysis scheme are designed,a fatigue prediction method for high-speed train dispatchers is constructed based on PAM-LSTM.The main research contents are as follows:(1)A high-speed train dispatching experiment was designed.The subjective KSS scores,supervisory KSS scores and EEG time series signals of fatigue behaviors of 20 dispatchers during the scheduling operation were collected,and the EEG characteristic signals that could reflect the fatigue state were screened.First,the frequency band of the collected EEG signals was reconstructed,and then the power spectral density PSD and its combined equation were calculated as EEG indicators.The number of principal components with a contribution rate exceeding 90% is obtained as 10 by KPCA principal component analysis.The smallest 10 EEG indicators with the fatigue correlation state greater than 0.7,the significance level of 0.01,the area under receiver operating characteristic curve greater than 0.75 and the smallest correlation between the indicators was calculated and screened as the EEG characteristic indicators.(2)PAM cluster analysis was performed on the fatigue behavior indicators data collected during the dispatching experiment.The fatigue degree was determined as 3 by calculating the silhouette coefficient and pseudo-F statistic.Combined with the dispatcher’s subjective KSS score,the fatigue degree of dispatchers during operation period was verified and calibrated.(3)Based on the calibration of the period fatigue degree of high-speed train dispatchers,the LSTM was used to predict the related fatigue degree.The results show that the model has high accuracy,which can predict the occurrence of fatigue in time,and characterize the development of fatigue reasonably.Through the analysis of the confusion matrix and the ROC curve,there is no situation where the fatigue degree is predicted to be 1 for the period of 2 or3,indicating that the model has high sensitivity and specificity.Using HMM for comparative analysis,showing that the PAM-LSTM prediction model constructed in this thesis is better than the PAM-HMM model in terms of accuracy and prediction timeliness.This work focuses on screening the EEG characteristic indicators that can reflect the dispatching fatigue,a PAM-LSTM fatigue prediction model is constructed and the time window and number of clusters in the model should be reasonably defined.The fatigue prediction results are finally calculated and verified.