Dynamic Strategies for Decentralised Urban Rail Transit Network Operations under Demand Uncertainty
DescriptionUrban rail transit systems play a vital role in sustaining cities’ mobility and development. Hong Kong has a population of more than 7.5 million and a population density of more than 6,600 people per square kilometre, and is thus one of the most densely populated cities in the world. According to the Hong Kong Government’s Public Transport Strategy Study published in 2017, more than 12 million passenger trips are taken daily on the public transport system, with 40% of these trips served by the urban metro and light rail transit networks operated by the MTR Corporation. With its continuous development and connection with the Greater Bay Area in China, the urban rail transit system is expected to play an increasingly important role in Hong Kong as recognised by the Government’s study. Following the Smart City Blueprint issued by the Government in 2020, Hong Kong has launched several major initiatives to improve the capacity and effectiveness of transport systems by introducing innovation and technology. Hong Kong’s rail transit systems are now equipped with advanced technologies such as automated fare collection, train positioning, and communication systems which can record and convey a variety of demands and operational data in real time. However, an effective computational framework is needed to incorporate these technologies to derive real-time operational strategies according to prevailing service and demand conditions to benefit both passengers and operators. This proposed project aims to develop a dynamic optimisation framework for urban rail transit networks with sharing of train units and passengers’ transfers over multiple service lines in real-time. The optimisation objectives are focused on minimising passengers’ waiting times, occupancy of trains, and operators’ costs by simultaneous determination of routes, schedules, and fleet sizes for train runs according to prevailing service statuses and demand uncertainties. Considering the large state and decision spaces encountered in the optimisation problem, the solution procedure will be developed by using reinforcement learning technique which reduces the computational complexity by making appropriate approximations to the underlying state and decision spaces. We will further develop a decentralised solution framework that can derive and accommodate asynchronous decisions made by connected local control agents. The proposed algorithms will be implemented and tested with real-world scenarios of Hong Kong rail transit network. The vulnerability of the optimisation algorithms with respect to potential disruptions will also be investigated. This project contributes to dynamic rail transit network operations with advanced computing techniques.
|Effective start/end date||1/01/23 → …|