Vehicle Lane Change Model Based on Improved Bayesian Network Structure Learning

10.6148/IJITAS.201812_11(4).0002

Tong Dang；Fang Dai

Traffic safety ； Vehicle lane change model ； Bayesian network ； K2 algorithm ； I-CH scoring function

International Journal of Intelligent Technologies and Applied Statistics

11卷4期（2018 / 12 / 01）

255 - 270

英文

In order to reduce traffic safety issues caused by vehicle lane change, a lane change model based on improved Bayesian network structure learning is established. The assumption of the traditional K2 algorithm that the prior probability of the network structure follows the uniform distribution is too arbitrary. Based on the Cooper-Herskovits (CH) scoring function, a new I-CH scoring function, which is constructed by using the connection probability information, is proposed to describe the network structure prior. After selecting experiments samples from the dataset Next Generation Simulation (NGSIM), the samples are discretized by the ChiMerge algorithm and then, the connection probability matrix with initial Bayesian network is obtained by computing the mutual information (MI) between variables. For the problem of node sequence in K2 algorithm, the principal component analysis is applied to optimize the node order and the combining I-CH score was proposed to train and verify this model. The experimental results show that the recognition rate of this model for the lane change behavior is up to 94.9% and the lane change model can be used in real-time situation.