Real-time Driving Pattern Prediction Based on KPCA and Neural Network

The approach of real-time driving pattern recognition based on Kernel Principle Component Analysis (KPCA) and Neural Network is put forward on uncertain driving conditions. Firstly, the dimension of vehicle driving patterns is reduced by KPCA. Then, the acquired low dimensional features are selected as the input of Neural Network for driving pattern classification. The simulation results show that KPCA + Neural Network has an advantage of nonlinear characteristics handling, which makes up for the limit of dealing with linear data by Principle Component Analysis (PCA). Moreover, Simulation results show that the dimension reduction of input patterns has simplified the structure of Neural Network by decreasing the number of hidden neurons greatly, and enhanced its real time recognition capability for driving patterns. The recognition accuracy reaches 95.24 percent.