Information-Driven Model Predictive Control With Adaptive Partitioning for Energy Optimization in Automated Electric Vehicles

This paper presents a methodology to optimize energy consumption in electric vehicles (EVs) using a Model Predictive Control (MPC) framework integrated with detailed power loss models. Minimizing energy usage during drive cycles is a complex problem due to the nonlinear and non-convex characteristics of energy consumption models. We derive a power loss model quantifying mechanical and electrical losses, dependent on vehicle speed and acceleration. To handle the non-convex power loss function, we apply adaptive partitioning to fit convex quadratic models within smaller operational regions. These convex models are integrated into the MPC to compute optimal control inputs that minimize energy consumption while satisfying vehicle dynamics and constraints. We model packet loss and sensor noise to enhance robustness against communication losses and data uncertainties, simulating real-world scenarios. Our strategy reduces battery energy demand by selecting energy-efficient trajectories, balancing energy savings with ride comfort by minimizing abrupt speed and acceleration changes. This methodology is suitable for integration with Advanced Driver-Assistance Systems (ADAS), contributing to improved vehicle performance, safety, and sustainability. Simulations using National Renewable Energy Laboratory (NREL) datasets demonstrate significant energy savings across diverse driving scenarios. Our method achieves up to 34.28% energy savings, outperforming the policy-based energy optimization algorithm integrated with Adaptive Cruise Control (ACC). Results confirm that our algorithm effectively maintains desired speed following and distance coverage under varying packet error rates (PERs), ensuring safe and efficient operation while achieving a balanced trade-off between significant energy savings and acceptable passenger comfort.