Machine Learning-Based Real-Time Axle Torque Prediction Model for Electric Tractors Using Field-Measured Data

Accurate estimation of axle torque is essential for performance evaluation and energy management of electric tractors. However, direct torque measurement and access to motor controller data are often limited in commercial platforms. This study proposes a machine learning-based framework for predicting axle torque in a commercial electric tractor using field-measured sensor signals. The framework incorporates a horizon-aware architecture to capture the temporal dependencies of dynamic load fluctuations. Field experiments were conducted during plow tillage operation under multiple gear–speed combinations. Several machine learning models (multiple linear regression, multilayer perceptron, and CatBoost) were evaluated for axle torque prediction. The results showed that rear axle torque exhibited a stronger relationship with traction demand under two-wheel-drive operation, resulting in higher prediction accuracy than front axle torque. Among the evaluated models, CatBoost achieved the best overall performance, with an R2 of 0.83 and an RMSE of 189.35 Nm for the rear axle prediction. The proposed framework enables real-time axle torque estimation using commonly available sensor signals and provides a practical alternative to direct torque measurement for onboard load monitoring and energy management in electric tractor systems.