Optimal control of automatic train operation based on multi-scale dynamic programming

Subway system plays an increasingly important role in urban transportation system, thus the optimization of subway control under multi-criteria is of great significance. Study on subway optimal control is to find a control strategy to regulate train´s velocity under constraints and given performance requirements, so as to obtain the optimal train speed curve. Many methods have been used to solve this problem such as Predictive Fuzzy Control, Genetic Algorithm and Simulated Annealing Algorithm. But they are not adaptive in general cases for they are just optimal under only one or two criteria. In this paper, we have model the train´s performance with five criteria. Moreover, an Multi-scale Dynamic Programming (MDP) is proposed to optimize the train´s regulation. Inspired from DP´s idea that a bigger problem could be solved based on solving its subproblems, MDP repeatedly divide distance section into smaller ones to be optimized, which makes our algorithm applicable on the computer. Simulation results demonstrate that MDP could obtain better train speed curve, 7.16% reduction of unpunctuality rate, when comparing to a Predictive Fuzzy Control method. It also demonstrates Multi-criteria model could make it flexible in different application scenarios.