Model building and evaluation for 4-wheel drive hybrid vehicle based on the random forest

In order to improve the accuracy of 4WD hv-pm (hybrid vehicle powertrain model), a virtual controller (VC) is established based on machine learning methods and benchmark data. Considering that developed VC can reflect the real performance of vehicle, the accuracy of 4WD hv-pm is measured by VC output. The machine learning method adopted by VC is random forest (RF), and the preferred benchmark data for building the training set are speed, acceleration pedal, brake pedal, state of charge (SOC), battery current, battery voltage, and traction demand. Moreover, the workable VC can provide engine torque, engine speed, motor torque and motor speed. Therefore, the VC output has been compared with the simulation results of constructed 4WD hv-pm, and the performance difference was evaluated to estimate the accuracy of constructed 4WD hv-pm, so as to further achieve the improvement of the model. Ultimately, it is proved that proposed VCexerts good effect on the effectiveness in model improvement and vehicle development by the means of comparison.