A Deep Learning Engine Power Model for Estimating the Fuel Consumption of Heavy-Duty Trucks

Abstract

An accurate heavy-duty truck (HDT) fuel consumption model is essential for estimating the truck energy consumption and evaluation of the effectiveness of energy saving strategies. Most of the existing models calculate the fuel consumption based on vehicle kinematics. However, based on recent truck field test data, we found the estimation discrepancies of several models published were considerable since they cannot accurately estimate the vehicle engine operation states and their base energy consumption rates. The engine power estimation is an essential part of the energy consumption model, which is the most likely cause of the model discrepancy. This inspired us to develop a generic modeling approach for vehicle engine-power estimation with a deep learning based on field experimental data. The data collected from a wide range of truck field tests were used for model development and fuel consumption model evaluation. The results show that the deep learning approach enables a much more accurate estimation of HDT engine power, and when applied as the input to the fuel consumption models (e.g., VT-CPFM, MOVES), average fuel estimation error is reduced to 13.71% from 28.9% which is the error resulted from the tractive power method. Besides, once calibrated with a small data set, it could be applied to various traffic scenarios without re-calibration. In addition, the Long Short-Term Memory (LSTM), a neural network structure component of the model, can accurately depict the fuel consumption during engine braking, which is largely missing in conventional HDT fuel models. The proposed model can benefit energy consumption related transportation planning and traffic operation studies that require more accurate vehicle fuel consumption estimation without detailed and complicated engine dynamics information.