RLE-Based Model Predictive Control vs. Rule-Based Control for Energy Management in Multi-Architecture Electric Vehicles

Abstract

Energy management is a key factor in the range, safety and battery life of electrified vehicles. In this study, a Recursive Least Squares (RLS)-augmented Model Predictive Control (MPC) framework is designed and validated for real-time energy management of four different EV powertrain architectures: Battery Electric Vehicle (BEV), Battery-Supercapacitor Hybrid, Battery-Flywheel Hybrid, and Battery-ICE Hybrid. The architecture is built in MATLAB/Simulink R2022b; an RLS subsystem is used to adapt in real time the battery impedance parameters. Representative multi-regime drive cycles were used to evaluate all eight quantitative metrics of performance. RLS-MPC constantly outperformed the rule-based control for all the architectures. The velocity tracking RMSE was reduced by 10.4% (BEV), 14.9% (Supercapacitor), and 13.9% (ICE Hybrid). Energy use was lowered for BEV (−4.5%) and Supercapacitor (−3.7%) architectures, and auxiliary energy use was shown to be greater for hybrid architectures: +61.5% (Supercapacitor), +430.6% (Flywheel), and +73.7% (ICE Hybrid), with auxiliary storage pre-positioned in advance of an expected deceleration. The Supercapacitor-MPC configuration was able to operate with a near-zero net battery draw (+0.006 kWh) compared to the baseline. Flywheel-MPC had a final state-of-charge of 84.7% (when including conversion losses, this was not yet energy neutral, but close). The vectorized MATLAB implementation required less than 30 μs to compute one step; the projected embedded execution time on automotive microcontroller hardware is estimated to be between 20–40 ms, consistent with the 100 ms sampling budget. RLS-MPC has great promise for future use with next-generation electrified powertrains, and future work will focus on embeddability.