Data-Driven Prediction Method for Truck Fuel Consumption Based on Car Networking

In order to search the appropriate truck fuel consumption prediction method based on the dynamic fuel consumption–speed data from the vehicle network, we first select the speed-related factors which are driver-controllable as fuel consumption influencing factors. After correlation analysis, we establish and implement a prediction model of truck instantaneous fuel consumption, which called the data-driven-based general regression neural network (GRNN) model. And beetle antennae search (BAS) algorithm is applied to find the proper training parameter of GRNN. Besides, three other models are established for contrast: back-propagation neural network based on kernel principal component analysis (KPCA-BPNN), representative of another kind of neural network model; the VT-Micro model, representative of traditional data-driven models; and VSP model, representative of traditional physical models widely used in practice. The results indicate that both two neural network models give out reasonable results better than the traditional VT-Micro model and VSP model. The fuel consumption predicted by VT-Micro model is obviously higher than the actual measurements when the idle ratio is abnormally high. KPCA-BPNN model performs best in fuel consumption prediction, but the KPCA-BPNN model requires excessive parameter adjustment which slows down computational effectiveness. Thus, the BAS-GRNN model with propel calibration and shorter training time is more suitable in practical application.