Least square-support vector regression based car-following model with sparse sample selection

Abstract

Car-following model is the basis of driving behavior modeling in microscopic traffic simulation. This paper proposes a car-following model based on Least Square-Support Vector Regression (LS-SVR). In order to reduce the computational complexity of LS-SVR, the maximum entropy theory is introduced to select typical samples from training data. Experimental results indicate that this selection method can ensure the accuracy of car-following model with the least samples. This car-following model is evaluated and validated by USTC Microscopic Traffic Simulation System (UMTSS). Simulation results of trajectory, speed and acceleration are accordance with those of field data. In addition, the proposed model is robust and reliable in the cases of both mild and severe disturbances. © 2010 IEEE.