Research On Driver Load Prediction And Risk Level Considering Time Series Matching And Multi-Feature Fusion

In recent years,the number of motor vehicles in our country has increased year by year,and with it comes a high number of deaths caused by traffic accidents.In the "people-vehicle-road" complex traffic system,the driver,as the core factor,plays the most critical role in the safety of the entire system.It has become an important measure to improve driving safety by monitoring and evaluating the load state of drivers based on their physiological signals and constructing a risk prediction model.In previous studies,most of the evaluation of the driver’s state is based on a single feature,or the fusion method between multiple features is not considered,and the time sequence matching between features is not considered during multi-feature fusion,which makes the driving Staff status assessments are biased.Therefore,this paper considers the fusion of multiple features in the driver risk assessment,and introduces time series matching to the fused features,predicts the driving load state,and evaluates the driver’s risk level through the prediction results.The following is the main research work of this paper.Firstly,various factors that affect the driving state are analyzed,the driving simulation experiment is designed by the control variable method,and the driving simulation experiment platform is built,and a series of driving simulation experiments are carried out by recruiting volunteers,the electrocardiograph(ECG),electroencephalogram(EEG)and vehicle manipulation data of drivers in different traffic flows and driving states were collected.Secondly,this paper proposes a driver load classification method based on multi-feature fusion with time series matching.The method is mainly divided into three parts.The first part first classifies drivers based on their reaction time and evaluates timing matching values;The second part constructs the basic probability distribution of parameters of D-S evidence theory through the training set results;The third part used the complete D-S evidence theory rules to verify the multi-features through the test set,and output the final decision.The decision results show that the driving load recognition accuracy after considering the matching time difference and using D-S fusion are significantly higher than those without matching time difference or only using ordinary fusion.Finally,according to the results of the driver load classification method,a driver risk level assessment model is proposed.This model is based on long short-term memory network,and uses the matching time difference with the highest accuracy in the driver load classification method to perform multi-feature fusion input to predict the driver’s load state,and evaluate the driver’s driving risk level according to the predicted driving state and the current driving state.The prediction evaluation results of this model show that compared with the actual driver’s risk level,the accuracy of the predicted driver’s risk level is as high as 85%,and the prediction results of this model in different categories of traffic flow are not very different.The research content of this paper provides a new idea and direction for driver load assessment,and considers the relationship between driver and vehicle handling more comprehensively and specifically,which has positive significance for the future research and development of road traffic safety.