Abstract: In order to utilize the statistic characteristics among the information that associated with categories in vehicle driving cycle effectively, the speed and the pedal signal data information are selected to construct feature set to improve the accuracy and rapidity of distinguishing the vehicle driving cycle firstly. Then, to realize the second reduction of the input variables, second feature which sensitively reflects characteristic data information of driving cycle category is extracted through the correlation analysis and the kernel principal component analysis. According to the standard of accumulation contribution rate that greater than 90%, the main feature is chosen to achieve secondary reduction of input variables. The nonlinear mapping ability of the wavelet kernel function is adopted to establish semi-supervised kernel fuzzy C means clustering algorithm for the working condition judgments. The experimental results show that this algorithm can reflect the driving cycle characteristics more comprehensively and accurately. In addition, the dimension of the input characteristic parameters is reduced effectively. And the data characteristics of vehicle driving condition are extracted under different working conditions more accurately and availably. Simultaneously, the clustering accuracy can reach 98.75% by adding a small amount of training samples in semi-supervised kernel fuzzy C means clustering algorithm to guide the clustering process.

Key words: driving cycle identification, kernel fuzzy C means clustering algorithm, kernel principal component, semi-supervised