Output-Constrained Linear Decision Trees for Optimal Energy Management Approximation of Hybrid Batteries

Rule-based control (RBC) strategies are easy to implement but often suffer from limited performance. In contrast, optimal control strategies can enhance performance by incorporating system dynamics and constraints; however, their high computational complexity makes them unsuitable for onboard implementation on low-cost hardware. In this paper, we introduce a novel decision tree regressor designed as a rule-based controller that closely approximates the performance of an optimal control-based strategy while maintaining a simple implementation. The proposed near-optimal RBC is applied to the energy management system (EMS) of a Hybrid Battery Energy Storage System (HBESS). A HBESS increases the performance of battery packs in electric vehicles (EVs). These energy storage systems combine high-energy (HE) and high-power (HP) cells to balance energy capacity and peak power output while reducing costs compared to using a single battery type. An EMS distributes the power between the HE and HP cells. The proposed decision tree regressor incorporates output-constrained linear models within its leaves, replacing the constant values used in traditional decision trees. This approach facilitates the integration of battery power constraints, resulting in a more compact rule-based controller that enhances interpretability while improving overall performance.