Application of PCA-K-means++ combination model to construction of light vehicle driving conditions in intelligent traffic

Abstract

The construction of typical driving condition of vehicles in line with the actual road traffic conditions in China requires the selection of the same vehicle for two months to collect driving data and the obtention of 496000 driving condition data of light vehicles. The sample data are preprocessed by using multivariate statistical theory and MATLAB. After the elimination of abnormal data, the effective data are extracted before being divided into 3020 kinematic segments. Then, it takes a principal component analysis to reduce the dimension of the characteristic parameter matrix. Through K-means++ clustering algorithm, the six principal components obtained by principal component analysis are clustered into two categories. Then the typical kinematic segments are selected from various fragment libraries by using correlation coefficient method, so as to construct a typical driving condition of the vehicles in a certain city. With the application of PCA-K-means and PCA-K-means++ clustering algorithm, a driving condition curve with a duration of 1200s is constructed before its effectiveness and accuracy being compared and analyzed. The results show that the error rate of driving condition between sample data and driving condition constructed by PCA-K-mean++ clustering algorithm is less than 6 % and the error rate of average speed and acceleration standard deviation is less than 1 %. The correlation degree between working condition curve constructed by PCA-K-means ++ clustering algorithm and sample data is increased by 4.08 %. The proportion of deceleration time and idle time in vehicle driving state is obviously different, which indicates that PCA-K-means++ is a better way to solve the problem and the clustering algorithm can effectively construct the driving condition curve of light vehicles suitable for local cities.