Rapid transit systems are critical components of urban public transportation networks in their impact, not only on personal mobility but also on the energy and environmental costs associated with network operations. To facilitate effective planning for current and future needs, a framework is required that provides important consumption metrics and also explains the various contributors to energy consumption, along with their interactions. To address this gap, we estimated models that utilized operational and ridership data for the Massachusetts Bay Transportation Authority’s rapid transit system, as well as ambient temperature, to accurately predict system-wide electricity consumption. The models were trained with data from 2019 and tested with data from 2020. The estimated multiple linear regression (MLR) and random forest (RF) models explained 93% and 95% of the variance in the data set, respectively. The MLR model provided predictions with a root mean squared error (RMSE) of 2.7 MWh and mean absolute percentage error (MAPE) of 4.68%, while the RF model resulted in an RMSE of 2.94 MWh and MAPE of 5.01%. We also investigated the impacts of COVID-19 on the transit system by exploring the effects on ridership, energy consumption, cost, and train movement metrics before and during the pandemic. We find that the models are robust and perform well, even with the significant disruptions associated with the COVID-19 pandemic.

Urban rail transit systems play an essential role in improving mobility and efficiency. A complex rail transit network serves the Boston metropolitan area, U.S., which costs $38 million for the 422 GWh of system electricity consumed annually. With the aim of developing a tool for energy and cost reduction decision support, we propose a comprehensive machine learning framework to investigate line-specific contributions to energy. This effort builds on prior work in estimating a system-wide energy model for the Boston network. By introducing line-specific train movement and operation variables, we obtain a higher-performing model [Formula: see text]. Furthermore, the model better explains the relationship between energy and train movement, ridership, and weather variables. Most importantly, the model facilitates analyses of how each line contributes to system consumption at the hour level. We found that the non-line-specific variables made a contribution of −2.7% to the average hourly energy of consumption of −5.4 MWh with a baseline energy consumption of 39 MWh. The Red Line dominates the energy consumption among line-specific variables, contributing 2.3% to the hourly average. Our model could be further enhanced to evaluate the energy and cost impacts of line-specific strategies that may be required for future planning and disaster response, as well as for real-time energy monitoring by line.