Learning-based decoupled CVT-MPC for plug-in hybrid electric vehicles: An ANFIS-enabled approach

Model predictive control (MPC) enables real-time energy management for plug-in hybrid electric vehicles (PHEVs) by utilizing future driving information. However, conventional MPC mainly optimizes power split and overlooks engine operating point optimization. Although incorporating continuously variable transmission (CVT) ratio optimization can improve engine efficiency, it increases the dimension of the online optimization problem. To overcome this limitation, a learning-based decoupled CVT–MPC energy management strategy is proposed. First, two-dimensional dynamic programming (2D-DP) is employed to jointly optimize the SOC trajectory, power split, and CVT ratio. An adaptive neuro-fuzzy inference system (ANFIS) is then trained to learn the optimal CVT control policy and establish the mapping between vehicle operating conditions and the optimal transmission ratio. During online operation, the ANFIS policy determines the CVT ratio, after which PMP-based receding-horizon optimization within the MPC framework allocates engine and motor power using a 10-s demand preview and the corresponding SOC reference. This hierarchical implementation transfers the coupled 2D-DP control results to the online controller while avoiding joint online optimization of the CVT ratio and power split. The proposed strategy is evaluated under a representative driving cycle generated in VISSIM and compared with rule-based (RB), dynamic programming (DP), and 2D-DP strategies. Results show that the proposed method effectively concentrates engine operating points near the Optimal Efficiency Line (OEL). The ANFIS model achieves CVT-ratio RMSEs of 0.057 and 0.067 on the testing and checking datasets, respectively. After correcting the terminal-SOC difference, the equivalent fuel consumption is 2.83 L/100 km, representing a reduction of 17.7% compared with the RB strategy. The proposed strategy provides an effective solution for improving engine efficiency and fuel economy in CVT-based PHEVs.