抄録
This study aims to explore the trip fuel consumption from a large-scale dataset. To better understand how the multiple variables (e.g., average travel speed, trip distance) influence the trip fuel consumption, we propose the support vector machine (SVM) to learn the relationship between the trip fuel consumption and the corresponding factors. A large-scale GPS and CAN (Controller Area Network) bus data provided by 153 probe vehicles during one month are used. Elasticity analysis indicates that trip distance and coefficient of variance of link speed have relatively great importance on the SVM model. To demonstrate the performance of the proposed method, three other regression methods, i.e., the multiple linear regression model, artificial neural network (ANN), and the link fuel summation SVM model (LSSVM) are also adopted for performance comparisons. The results show that SVM model is much closer to the target than the other three models.
