Energy oriented driving behavior analysis and personalized prediction of vehicle states with joint time series modeling

Abstract

Analyzing the energy consumption for road entities and the corresponding driving behaviors are critical tasks for the realization of public traffic with a low energy cost and high efficiency. In this study, a personalized energy consumption analysis and prediction framework are proposed to estimate future energy consumption and the speed of a vehicle. An accumulation energy consumption index is predicted based on the features of the driving behavior. This approach is independent of the vehicle style, and it can play a critical role in the estimation of energy consumption as well as energy management for both petrol and electric vehicles. Three different energy-oriented driving behaviors are first identified and compared. It is shown that the vehicles with heavy energy usage have the characteristics of a higher speed, larger acceleration, larger headway space, and smaller headway time. The relationship between the energy consumptions and acceleration-deceleration characteristics are analyzed, and it is noted that the heavy energy users tend to perform acceleration maneuvers more frequently and with a longer period. Finally, a personalized joint time series modeling system based on the long short-term memory and a recurrent neural network is designed to jointly estimate the future energy consumption index considering different driving styles. It is found that the proposed personalized sequence prediction framework can generate more accurate results than the models that do not consider the energy cost levels and driving behaviors. The next-generation simulation data for free highway driving behaviors are used for the analysis and model evaluation.