Toward Intelligent Train Driving through Learning Human Experience

As the most sustainable means of modern transportation, modern railway is approaching a new era of intelligence. A major part of this momentum is the autonomous driving of railway vehicles to enable energy efficiency and safety. The derivation of an energy-optimal solution to the intelligent train driving problem in an efficient fashion, however, proves to be a significant challenge due to the high dimension, nonlinearity, complex constraints, time-varying characteristics as well as the limited computing power of onboard computers. This paper proposes a human experience driven fuzzy strategy tree framework for online generation of the energy optimized driving solution for railway trains. Starting from the frequent driving patterns of human drivers, we propose a fuzzy strategy tree structure to effectively encode the driving strategy for different working conditions. The strategy tree is built with an error driven Ripple Down Rules (RDR) method through trip simulations. The genetic algorithm is exploited to optimize the strategy parameters. We also incorporate fuzzy inference methods to handle the uncertainty of a trip. An onboard automatic controller searches the fuzzy strategy tree according to the route information to identify an optimized driving strategy for detailed control operations. Such hybrid techniques make it feasible to identify and apply the best practices of human drivers under a computation-efficient framework. The proposed techniques are validated by field test on commercial trains and enable an energy saving of over 10%.