Neural Network-based Control Strategy for Energy-efficient Front-rear Torque Vectoring in Electric Vehicles with Multiple Motors and Disconnect Clutches

The paper deals with neural network (NN)-based energy-efficient front/rear torque vectoring (TV) for an electric vehicle (EV) with multiple motors and disconnect clutches. The NN model reflects a binary classification problem, formulated to distinguish between four-wheel (4WD) and two-wheel drive (2WD) powertrain configurations depending on current vehicle velocity and total torque demand inputs. The model provides a probability of EV being set in 2WD configuration with disconnected clutches on the opposite axle. By introducing a probability threshold on the NN model output, a proper trade-off between energy efficiency and suppression of clutch state switching frequency can be posed. The NN model is trained based on the globally optimal dataset obtained by off-line executed dynamic programming (DP) optimizations over multiple certification driving cycles. The proposed NN-based TV control strategy is tested against the DP benchmark and a rule-based (RB) TV baseline.