Comparative Analysis of Predictive Models for Subway Rolling Stock Energy Forecasting

Devising robust energy management strategies for sustainable future urban transportation systems requires in depth understanding of the underlying generative data patterns. Once quality datasets are collected and pre-processed, various computational intelligence technniques can be applied to extract actionable forecasts in order to improve the overall system efficiency. We present a data science methodology for subway rolling stock energy forecasting based on real collected data from traction power transformers of two subway stations in the city of Bucharest, Romania. The effect of variable passenger loads is also investigated in a two-stage forecasting process involving exponential smoothing filtering and regressive forecasting. The main forecasting results yield a MAPE indicator under 0.2% for both evaluated scenarios, with an improvement achieved by integrating the passenger traffic as additional contextual explanatory variable for the model.