Use this URL to cite or link to this record in EThOS:
Title: Adaptive railway traffic control using approximate dynamic programming
Author: Ghasempournejad Seifdokht, Taha
ISNI:       0000 0004 8508 1945
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Railway networks around the world have become challenging to operate in recent decades, with a mixture of track layouts running several different classes of trains with varying operational speeds. This complexity has come about as a result of the sustained increase in passenger numbers where in many countries railways are now more popular than ever before as means of commuting to cities. To address operational challenges, governments and railway undertakings are encouraging development of intelligent and digital transport systems to regulate and optimise train operations in real-time to increase capacity and customer satisfaction by improved usage of existing railway infrastructure. Accordingly, this thesis presents an adaptive railway traffic control system for realtime operations based on a data-based approximate dynamic programming (ADP) approach with integrated reinforcement learning (RL). By assessing requirements and opportunities, the controller aims to reduce delays resulting from trains that entered a control area behind schedule by re-scheduling control plans in real-time at critical locations in a timely manner. The present data-based approach depends on an approximation to the value function of dynamic programming after optimisation from a specified state, which is estimated dynamically from operational experience using RL techniques. By using this approximation, ADP avoids extensive explicit evaluation of performance and so reduces the computational burden substantially. In this thesis, formulations of the approximation function and variants of the RL learning techniques used to estimate it are explored. Evaluation of this controller shows considerable improvements in delays by comparison with current industry practices.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available