Machine Learning for Predicting the Energy Consumption of Catenary Passenger Trains

Keywords

Loading...
Thumbnail Image

Issue Date

2022-07-11

Language

en

Document type

Journal Title

Journal ISSN

Volume Title

Publisher

Title

ISSN

Volume

Issue

Startpage

Endpage

DOI

Abstract

NS (Dutch Railways) wants to reduce the traction energy used by the catenary passenger trains. One step towards this goal is providing feedback to the train drivers on the level of eco-driving of their journeys. Creating a baseline for the amount of consumed energy that adapts to the circumstances of a train trip would make comparisons realistic. Five machine learning models (linear regression, support vector regression, decision tree, random forest regression (RFR), extreme gradient boosting) are trained to predict the average energy consumption of train trips (kWh/km) from external factors which the train drivers do not have control over, creating a dynamic baseline for feedback on eco-driving. The best model is RFR (R2 = 0.91). Furthermore, four feature importance analysismethods (ablation, permutation, individual conditional expectation, Gini importance) are conducted to rank the input features by importance. Features regarding details about the rolling stock, trajectory, temperature and date are important for predicting the energy consumption.

Description

Citation

Faculty

Faculteit der Sociale Wetenschappen