Contextual on-board learning and prediction of vehicle destinations

Abstract

This paper deals with the problem of on-board learning of typical stop locations and the prediction of the vehicle destination. Such a learning and prediction procedure is used to summarize the stop locations, estimate the frequent destinations, and learn the driver's decision model of selecting the next destinations under different conditions. The prediction of the driver's usage pattern is useful in generating optimal control policies for energy management control in electrified vehicles. The proposed approach is based on the real-time clustering and learning of a decision model that combines fuzzy and Markov models. The former is applied to represent possibilistically the context of the destination selection while the latter covers the probabilistic process of choosing a destination for given conditions.