Research For Car-following Model Based On Genetic Algorithm Optimized BP Neural Network

Car-following behavior is one of the common driving behaviors,and different drivers keep different safety distance and relative speed and so on because of their specific psychological response. If we can simulate characteristics of the driver and establish a specific car-following model,we can compare similar driving behavior and decide whether the driver take the presence of abnormalities or not. Then we can take appropriate measures to send timely warning of dangerous behavior to effectively reduce the incidence of accidents.In this paper, we use video surveillance systems, millimeter-wave radar and other equipment to collect car-following data.Through the analysis of relevant data that impact driver’s acceleration and deceleration during stable car-following, we can fund the prediction model based on effective test data. The main contents and conclusions:(1) By retrospective studies of car-following models and extract vehicle operation parameters and road environment data related to driver acceleration and deceleration. Preliminary analysis finds that relative vehicle speed, relative distance and the vehicle speed are the possible input parameters.(2) By processing the extracted preliminary characteristic parameter and we establish a BP neural network car-following model. Prediction analysis shows that BP model is easy to fall into extreme point, and this problem cannot be resolved by optimizing the structure of its own, so we consider using genetic algorithm to optimize it.(3) After using genetic algorithm to optimize the structure of single hidden layer BP neural network model, it shows that relative speed, relative distance and of the vehicle acting as a combination input provides the highest predictive accuracy, but it is only 90.29%. Through trial and error we see that genetic algorithm optimized double hidden layer BP neural network can increase forecasting rate to 94.17%.The results show that genetic algorithm optimized double hidden layer BP neural network can be a great predictor of car-following state.This research was sponsored by the Changjiang Scholars and Innovative Resear ch Team Support Program of the Ministry of Education( IRT1286) and the Application Basic Research Project of the Ministry of Transport(2013319812150).