A cost-efficient data-driven framework for rail temperature prediction

Abstract

High rail temperatures pose significant risks to track integrity, increasing the likelihood of buckling. This study proposes a cost-effective, data-driven framework for rail temperature prediction using only ambient temperature and relative humidity as inputs, variables routinely collected by standard meteorological stations, thereby eliminating dependence on complex sensor arrays. Five machine learning algorithms were benchmarked: Artificial Neural Networks (ANN), Support Vector Regression (SVR), Random Forest (RF), Long Short-Term Memory networks (LSTM), and XGBoost. Two complementary prediction strategies were evaluated: a long-term global model trained on a two-year historical dataset, and a short-term sliding-window model predicting up to three hours ahead. Models were trained on Brazilian railway data from 2022–2023 and validated on an independent 2024 dataset. Hyperparameters were optimized via Optuna using a Tree-structured Parzen Estimator (TPE). Results demonstrate that ANN delivered the lowest long-term error (mean absolute error = 2.0978 ◦C; R2 = 0.8571), closely matched by XGBoost, which obtained the highest R2 (0.8652), while RF achieved the best short-term precision at the one-hour horizon (R2 = 0.9395; mean absolute error = 1.1944 ◦C). Despite the deliberately limited input set, the proposed framework achieves competitive accuracy, offering a practical solution for railway networks operating under limited monitoring infrastructure.