Fuel consumption prediction for the tractors augmented with GNSS recordings

With the rapid advancement of agricultural mechanization, fuel consumption in tractors has increased, leading to higher production costs. Selecting tractors with lower fuel consumption requires a method to evaluate fuel consumption in practical scenarios. In this study, we constructed a fuel consumption dataset based on a number of tractors operating in various regions in China, and we proposed a novel approach that can automatically predict the instant fuel flow rate of tractors. Firstly, we collected a large-scale fuel consumption dataset recorded by GNSS (Global Navigation Satellite System) and ECU (Electronic Control Unit) devices installed on 90 tractors with three tractor models. Then, we analyzed the interactions among a number of factors involved in their fuel consumption, and characterized each data point with the four parameters: two engine-based parameters (torque and engine speed) and two motion-based parameters (driving speed and acceleration). Based on the four parameters and a powerful machine learning method (Random Forest), we developed a fuel consumption prediction model, which predicts the fuel flow rate at each point. Finally, we made intensive experiments, which demonstrate that the proposed method achieves state-of-the-art performances in predicting fuel flow rate, yielding at least an average R² value of 0.88 on the three tractor models. Moreover, an in-depth analysis was made to examine the accuracy and transfer capability of the developed prediction models.