Fuel consumption estimation in heavy-duty trucks: Integrating vehicle weight into deep-learning frameworks

Abstract

Insufficient consideration of vehicle weight dynamics during real-world driving could lead to inaccurate fuel consumption estimates. This study examined the impact of vehicle weight on fuel consumption rate (FCR) by analyzing extensive, high-resolution operating data obtained from 162 heavy-duty trucks (HDTs). An engine output power-based (EOP) model, an artificial neural network (ANN) model, and a long short-term memory-convolutional (LSTM-Conv) model were developed and contrasted with conventional vehicle specific power (VSP) and Virginia Tech Microscopic (VT-Micro) models. The results indicated a significant, non-linear relationship between weight and FCR. Compared to 5-ton trucks, FCR for trucks weighing 15–25 tons and 45–55 tons increased by 290% and 755%, respectively, under low-speed and positive acceleration conditions. The LSTM-Conv model outperformed the VSP, VT-Micro, and EOP models, achieving MAPEs of 9.81% for FCR and 1.49% for trip fuel economy estimation. The deep-learning models exhibited enhanced stability across varying speeds, accelerations, and vehicle weights.