Summary
Urban railway systems produce regenerative braking energy, which, if properly managed, can significantly reduce carbon emissions. This paper evaluates the ability of deep reinforcement learning (DRL) to optimise the use of regenerative energy from urban railway systems stored in an energy storage system, aiming to maximise CO2 emission savings. Unlike conventional approaches focused on economic optimisation, environmental impact is prioritised, addressing a harder control setting shaped by the uncertainty of train braking events and fluctuating grid emission intensities.
Read more Read lessThe proposed method formulates the problem as a partially observable sequential decisionmaking task and trains an agent to decide optimal energy discharge strategies without requiring future emission predictions. A custom dataset based on real-world energy generation and railway operation data at an electrical substation of Metro de Madrid, Spain, was used to train and evaluate the model, mimicking a real-time decision-making environment. The dataset also considers peak energy recovery events and carbon emission data.
Results show that the DRL agent captures a substantial fraction of an offline upper bound, within a few training hours, demonstrating adaptability even in challenging scenarios with atypical energy demands. The study underscores the robustness and efficiency of DRL-based policies in reducing carbon emissions, making it a viable solution for sustainable railway energy management.
Additional informations
| Publication type | Session Materials |
|---|---|
| Reference | C6_11129_2026 |
| Publication year | |
| Publisher | CIGRE |
| Country | United Kingdom |
| Study committees | |
| File size | 676 KB |
| Price for non member | 30 € |
| Price for member | 30 € |
Authors
UGALDE-LOO Carlos E - Cardiff University United Kingdom; SAIKIA Pranaynil - Cardiff University United Kingdom; MUÑOZ-CRIOLLO Jose J - Cardiff University United Kingdom; SÁNCHO ACEVEDO Ruben - UNED Spain; CÁRPIO IBÁÑEZ José - UNED Spain
Keywords
