Innovative modeling strategy of wind resistance for platoon vehicles based on real-time disturbance observation and parameter identification

Variable inter-vehicle distances influence significantly the wind resistance of platoon vehicles due to the sheltering of airflow. Accurate air drag estimation is extremely important for platoons in scenarios like energy-oriented driving and high-precision tracking. Most aerodynamic researchers have performed qualitative analysis of wind resistance for equally inter-spaced platoons, while the quantitative description of wind resistance variation with coupled inter-vehicle distances is rare. In addition, data measured through offline wind tunnel experiments, Computational Fluid Dynamics (CFD) simulations, or road tests via fuel consumption calibration for a period of time is unsynchronized, which may be unconvincing for real air drag estimation on road. Aiming at the quick and accurate approximation of platoon wind resistance, this paper proposes a novel and universal modeling strategy combining offline CFD simulation, online air drag observation, and real-time parameter identification. The variation characteristics of air drag with distance of a longitudinal platoon consisting of three homogeneous C-class Notchback cars are analyzed by CFD simulation. With appropriate data processing, a well-designed basis function is summarized. Then a novel wind resistance separation method combining Back-Propagation Neural Network (BPNN) and Extended State Observer (ESO) is proposed. Using the observed data stored in experience memory, the hybrid optimization method via particle swarm optimization (PSO) and gradient descent with momentum (GDM) is employed to identify the model parameters toward high accuracy and global optimality. Results of Hardware-In-the-Loop (HIL) experiment show that the proposed modeling strategy realizes effective real-time observation and accuracy description; the developed approximation model can describe the platoon wind resistance with continuous and coupled inter-vehicle distances, with the RMSE less than 11.4%.