Predictive power demand for energy management of hybrid electric tractors

Abstract

Stricter emission regulations are driving the electrification of agricultural and equipment machinery, including tractors. In order to enable predictive energy management strategies aimed at reducing fuel consumption and carbon emissions, this study presents a framework for accurately forecasting the power demand of hybrid electric tractors in agriculture. As agricultural duty cycles exhibit recurrent patterns, precise load forecasting can be achieved through non-linear models. This work focuses on two main contributions: the statistical analysis of tractor real-duty scenarios in agriculture and the development of a power demand predictor model for tractors. Based on the real-data framework, the characteristics of the farming cycles and relevant prediction features were constructed and selected. In particular, long short-term memory and convolutional neural networks were trained using extensive agricultural duty cycle data. Based on a comprehensive review of existing literature, the application of such neural network models to predict the power demand of tractors in agriculture represents a novel approach. The results demonstrate that the convolutional neural network model provides highly accurate power demand forecasts, which can be effectively utilized in predictive energy management strategies to enhance powertrain performance. Notably, the investigated neural network models achieve excellent accuracy for prediction horizons of up to one minute and can be used effectively with predictive control strategies to lower fuel consumption and extend powertrain lifetime. Overall, these findings have significant implications for advancing sustainable and efficient farming practices through the integration of hybrid electric technology and predictive control in agricultural machinery.