Stochastic energy model predictive control for cooperative adaptive cruising

Efficient energy management represents a critical objective across diverse applications, notably within the automotive sector driving the development of Eco-Cooperative adaptive cruise control (Eco-CACC). While established control strategies frequently utilize model predictive control (MPC), these methods can exhibit limitations in stochastic energy optimization contexts. This study develops and evaluates a stochastic energy model predictive control (SE-MPC) framework designed for broader energy optimization problems, subsequently tailored and applied specifically to the CACC challenge. Key contributions include the integration of a Kolmogorov-Arnold network (KAN) based preceding vehicle speed estimation algorithm to bolster system robustness against uncertainties. Furthermore, a trajectory band loss function is integrated within the SE-MPC formulation. This function acts as an alternative to conventional hard state constraints, effectively mitigating issues related to solver infeasibility and convergence to local optima often encountered when imposing strict constraints under uncertainty. To enhance computational efficiency, a suboptimal SE-MPC variant is also proposed. Simulation results indicate that this suboptimal SE-MPC algorithm yields a 5.14 % reduction in energy consumption for the CACC application. Moreover, within multi-vehicle cooperative platooning scenarios, the SE-MPC demonstrates its capability to mitigate phantom traffic jams, achieving a 48.33 % reduction in energy consumption and enhancing vehicle string stability when responding to sudden lead vehicle braking. This research presents a versatile SE-MPC methodology applicable to energy optimization tasks, demonstrates its specific advantages within the demanding CACC context.