Level-wise driving knowledge induction for embedded automatic train driving system

Abstract

Automatic driving of trains can significantly reduce the energy cost and enhance the operating efficiency and safety. The automatic train driving system has to be an embedded system that can run onboard with low power, which necessitates an efficient inference model. In this article, a level-wise driving knowledge induction approach is proposed for embedded automatic train driving systems. The coincident driving patterns in the records of drivers with different experience levels suggest the suitability of a driving experience knowledge rule induction approach. We design a two-level learning approach to obtain both the driving experience pattern in fuzzy rule-based knowledge form and the detailed parameters of velocity and gear by regression learning methods. With 8.93% energy consumption reduction compared with average human drivers, the experiments indicate the effectiveness of our approach.