Digital Twin Implementation for Urban Tram Systems: A Data-Driven Model for Estimating Traction Current

The urgent need to decarbonize our cities places the transport sector—responsible for a significant share of greenhouse gas emissions—at the center of global sustainability efforts. Achieving this goal requires not only the electrification of public transport systems but also ensuring their efficiency and effectiveness to encourage widespread adoption over private vehicles. A promising strategy to modernize such systems is the implementation of a Digital Twin (DT), a virtual counterpart of the infrastructure that integrates real-time field data from strategically deployed sensors. This work focuses on trambased urban transport systems, where one of the most critical parameters for DT applications is the traction current absorbed by each vehicle. However, the deployment of dedicated onboard current sensors is often constrained by economic and timerelated limitations. To address this, we propose a transitional solution: a data-driven model capable of estimating tram current consumption using only GPS-based vehicle position data. Specifically, this paper presents a hybrid neural network architecture combining Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. The network processes speed and acceleration profiles derived from open-access GPS data, eliminating the need for onboard instrumentation. The model was trained and validated with data from a targeted field measurement campaign, achieving a Mean Absolute Error (MAE) of 61.3 A and a Root Mean Square Error (RMSE) of 95.6 A. Given that the maximum absorbed current exceeds 900 A, these error values indicate that the model's predictive performance is within an acceptable range. This methodology provides a robust, low-cost solution for enabling early-stage DT development in Urban Traction Electrification Systems (UTES), supporting simulation and planning in the ongoing digital transformation of public transport.