Risk Prediction Model of Vehicle Driving in Small Radius Curves Based on Simulation Experiment

Abstract:

The driving data of car driving in low-grade highway was collected from the simulation experiment. Focus on the horizontal risk versus vertical risk， 8 major statistics were calculated from the driving data， from which， 3 core statistics （standard deviation of velocity， tangential acceleration and steering wheel speed） were screened by principal component analysis（PCA）. Based on the 3 core statistics， 4 kinds of risks （weak risk， sharp brake， sharp turn， strong risk） were separated by k-mean clustering. Then， a risk prediction model of vehicle driving in small curves was founded based on Fisher， whose test set was the 3 core statistics of the initial 1/4 period that vehicle driving in curves， and training set was that of the whole curve. The accuracy of the risk prediction model was 76.5%.

Key words:simulation experiment;low-grade highway;k-mean clustering;Fisher Model