Evaluation and optimisation of traction system for hybrid railway vehicles

Din, Tajud (2023). Evaluation and optimisation of traction system for hybrid railway vehicles. University of Birmingham. Ph.D.

Preview

Din2023PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (3MB) | Preview

Abstract

Over the past decade, energy and environmental sustainability in urban rail transport have become increasingly important. Hybrid transportation systems present a multifaceted challenge, encompassing aspects such as hydrogen production, refuelling station infrastructure, propulsion system topology, power source sizing, and control. The evaluation and optimisation of these aspects are critical for the adaptation and commercialisation of hybrid railway vehicles. While there has been significant progress in the development of hybrid railway vehicles, further improvements in propulsion system design are necessary.

This thesis explores strategies to achieve this ambitious goal by substituting diesel trains with hybrid trains. However, limited research has assessed the operational performance of replacing diesel trains with hybrid trains on the same tracks. This thesis develops various optimisation techniques for evaluating and refining the hybrid traction system to address this gap.

In this research's first phase, the author developed a novel Hybrid Train Simulator designed to analyse driving performance and energy flow among multiple power sources, such as internal combustion engines, electrification, fuel cells, and batteries. The simulator incorporates a novel Automatic Smart Switching Control technique, which scales power among multiple power sources based on the route gradient for hybrid trains. This smart switching approach enhances battery and fuel cell life and reduces maintenance costs by employing it as needed, thereby eliminating the forced charging and discharging of excessively high currents. Simulation results demonstrate a 6% reduction in energy consumption for hybrid trains equipped with smart switching compared to those without it.

In the second phase of this research, the author presents a novel technique to solve the optimisation problem of hybrid railway vehicle traction systems by utilising evolutionary and numerical optimisation techniques. The optimisation method employs a nonlinear programming solver, interpreting the problem via a non-convex function combined with an efficient "Mayfly algorithm." The developed hybrid optimisation algorithm minimises traction energy while using limited power to prevent unnecessary load on power sources, ensuring their prolonged life. The algorithm takes into account linear and non-linear variables, such as velocity, acceleration, traction forces, distance, time, power, and energy, to address the hybrid railway vehicle optimisation problem, focusing on the energy-time trade-off. The optimised trajectories exhibit an average reduction of 16.85% in total energy consumption, illustrating the algorithm's effectiveness across diverse routes and conditions, with an average increase in journey times of only 0.40% and a 15.18% reduction in traction power. The algorithm achieves a well-balanced energy-time trade-off, prioritising energy efficiency without significantly impacting journey duration, a critical aspect of sustainable transportation systems.

In the third phase of this thesis, the author introduced artificial neural network models to solve the optimisation problem for hybrid railway vehicles. Based on time and power-based architecture, two ANN models are presented, capable of predicting optimal hybrid train trajectories. These models tackle the challenge of analysing large datasets of hybrid railway vehicles. Both models demonstrate the potential for efficiently predicting hybrid train target parameters. The results indicate that both ANN models effectively predict a hybrid train's critical parameters and trajectory, with mean errors ranging from 0.19% to 0.21%. However, the cascade-forward neural network topology in the time-based architecture outperforms the feed-forward neural network topology in terms of mean squared error and maximum error in the power-based architecture. Specifically, the cascade-forward neural network topology within the time-based structure exhibits a slightly lower MSE and maximum error than its power-based counterpart. Moreover, the study reveals the average percentage difference between the benchmark and FFNN/CNFN trajectories, highlighting that the time-based architecture exhibits lower differences (0.18% and 0.85%) compared to the power-based architecture (0.46% and 0.92%).

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):