AI-Enabled Rail Electrification and Sustainability: Optimizing Energy Usage with Deep Learning Models

Abstract

This work focuses on the optimization of metric parameters for rail electrification systems to minimize energy usage and CO2 emissions. A novel approach based on deep learning models is proposed for accurately predicting the need for new plants or for aligning existing systems. The whole workflow for dataset collection, model generation, and model usage is developed, explaining the key points of the process by testing the models on real sections. In this application, the use of deep learning techniques raised unexpected links between variables and allowed for a reduction in the number of inputs to the minimum requirements in real product usage. The data input and their technical meaning are listed to allow the right use for the specific plant geometry and for the specific cost function to be addressed (e.g., minimizing CO2 emissions instead of power usage). A detailed sensitivity analysis on the impact of input parameters is also provided to further investigate the key internal variables on the power requirement. According to the results, the models endorse a manageable approximation of the step impedance at different operating conditions, leading to a simplified representation of the pantograph interaction with the rail traction line. By reducing the input parameters—which are entirely real-time achievable—the proposed model can be utilized for feasibility studies of non-electrified railway lines or as a tool for prediction and comfort applications for other trains passing under energized catenaries, provided the proper training is addressed.

Keywords

• Rail Electrification, Energy Optimization, CO2 Emissions, Deep Learning, Sensitivity Analysis, Model Prediction, Pantograph Interaction, Input Parameters, Feasibility Studies, Real-time Data.