Learning to optimize timetables for efficient transfers in public transportation systems

Abstract

This article presents the application of a learning-based optimization method to solve the Bus Synchronization Problem, a relevant problem in public transportation systems. The problem consists in synchronizing the timetable of buses to optimize the transfer of passengers between bus lines. A new problem model is proposed, extending previous formulations in the literature, and solved using Virtual Savant. Virtual Savant is a novel soft computing method inspired by the Savant Syndrome that combines machine learning and optimization to solve complex real-world problems in a massively parallel way. The proposed methodology is validated and evaluated over a set of synthetic and realistic instances based on the public transportation system of Montevideo, Uruguay, and compared against a reference evolutionary algorithm and the current solution defined by the city authorities. The main results indicate that Virtual Savant is able to compute accurate solutions and outperform baseline results in eleven out of fifteen realistic instances. This is the first reported research applying Virtual Savant to a problem with a high synergy between its decision variables. The obtained results suggest that it is a suitable tool for solving this kind of optimization problems.