Study On The Prediction Method Of Driver’s Simple Reaction Time

As the manipulator of traffic system,the driving state of drivers is the key factor that influences the transportation system security.Nowadays,accident rate is rising because of human factors,thus the condition monitoring of drivers becomes the important issue of traffic safety research.Previous studies said the reaction speed responding to emergencies of drivers is the most important psychological character that influences the driving safety,and the physiological state fluctuation of drivers which could influence reaction time has obvious characterization on physiological signals,especially on EEG,moreover,current studies have shown that there was explicit qualitative relationship between the reaction time and the EEG frequency signals.EEG signals were widely used to the recognition of cognitive ability of drivers because of its high precision.At present study,the recognition of the decrease of cognitive ability that caused by fatigue and the decreased vigilance mainly relies on EEG signals,but there is little study researching on the reaction time prediction of drivers by EEG signals.In the view of above problems,this paper analyses the relationship of simple reaction time and EEG signals,construct the EEG signals frequency feature as inputs,and construct model to predict the simple reaction time.In the current study,ICA analytical approach was adopted to remove the artifact,the Fourier transform was adopted to extract spectral characteristics of EEG and to smooth it,GRA were used to select features,furthermore,LSSVM and SVR which were optimized by PSO were built to predict the driver’s simple reaction time.Based on analysis of experimental data,the selection results of GRA indicated that the frequency domain characteristics ofαandβpower band in Pz electrode has the highest correlation with response time.Correlation analysis showed that there is a positive correlation betweenαpower band and response time(r(28)0.474,p(27)0.05),there is a negative correlation betweenβpower band and response time(r(28)-0.331,p(27)0.05).By comparing these two prediction models,the LSSVM outperformed the SVR,and the result showed that using the curve as a context could enhance the performance of the prediction model.In general,GRA-LSSVM is with the mixed features as inputs was the best combination.Modelling for each participants,the mean RMSE of model prediction results is 162.21ms,the mean R~2 is 0.841.Based on the data analysis of validation experiment,the validity and feasibility of the proposed method was verified.