Near real-time timetabling for metro system energy optimization considering passenger flow and random delays

Researchers have focused on optimizing rail transit speed trajectory and timetable in the past decades, and a highly optimized timetable can be obtained based on their achievements. However, train drivers frequently face various disturbances during the travel procedure, such as boarding time changes and weight changes caused by passenger flow, thus rendering the optimized timetable ineffective. This paper proposes an innovative method of real-time timetable optimization based on dynamic passenger flow and random disturbances. Stations along a metro line are grouped into several clusters. The timetable and corresponding speed trajectory of each train within a cluster are individually optimized based on a trained neural network and a mixed-integer linear programming (MILP) model. Such a system provides an optimized travel plan while ensuring no significant difference exists between it and the predetermined one. The tested trains in the case study section show average energy savings of 5.11%. Furthermore, 8619 scenarios with only delay disturbance and 26,537 scenarios with only weight change disturbance were simulated in the data analysis section to discover the energy-saving efficiency for these two different types of disturbances; the analysis results show an average of 10.13% and 0.21% energy savings, respectively.