Guo, Yida
(2021)
Multi-agent Near Real-Time Simulation of Light Train Network Energy Sustainability Analysis.
PhD thesis, University of Liverpool.
|
Text
PHD THESIS Revised by Yida 2021-09-01.pdf - Unspecified Download (5MB) | Preview |
Abstract
As an attractive transportation mode, rail transit consumes a lot of energy while transporting a large number of passengers annually. Most energy-aimed research in rail transit focuses on optimizing the train timetable and speed trajectory offline. However, some disturbances during travel will cause the train to fail to follow the offline optimized control strategy, thus invalids the offline optimization. In the typical rail transit control framework, the moving authority of trains is calculated by the zone controller based on the moving/fixed block system in the zone. The zone controller is used to ensure safety when the travel plan of trains changes due to disturbance. Safety is guaranteed during the process, but the change of travel plan leads to extra energy costs. The energy-aimed optimization problem in rail transit requires ensuring safety, pursuing punctuality with considering track slope, travel comfort, energy transferring efficiency, and speed limit, etc. The complex constraints lead to high computational pressure. Therefore, it is difficult for the regional controller to re-optimize the travel plan for all affected trains in near real-time. Multi-agent systems are widely used in many other fields, which show decent performance in solving complex problems by coordinating multiple agents. This study proposes a multi-agent system with multiple optimization algorithms to realize energy-aimed re-optimization in rail transit under different disturbances. The system includes three types of agents, train agents, station agents and central agents. Each agent exchanges information by following the time trigger mechanism (periodically) and the event trigger mechanism (occasionally). Trigger mechanism ensures that affected agents receive necessary information when interference occurs, and their embedded algorithms can achieve necessary optimization. Four types of cases 5 / 128 are tested, and each case has plenty of scenarios. The tested results show that the proposed system provides encouraging performance on energy savings and computational speed.
| Item Type: | Thesis (PhD) |
|---|---|
| Divisions: | Faculty of Science and Engineering > School of Engineering |
| Depositing User: | Symplectic Admin |
| Date Deposited: | 30 Sep 2021 15:44 |
| Last Modified: | 18 Jan 2023 21:27 |
| DOI: | 10.17638/03138666 |
| Supervisors: |
|
| URI: | https://livrepository.liverpool.ac.uk/id/eprint/3138666 |
Dimensions
Dimensions
