Digital-twin-enabled reinforcement learning for energy-efficient and reliability-aware software-defined electric vehicles

Abstract

Software-defined electric vehicles (SDEVs) require intelligent control frameworks that simultaneously optimize energy efficiency and powertrain reliability. However, existing studies typically treat reinforcement learning, digital twins, and predictive health monitoring as separate problems, limiting real-time reliability-aware decision-making. This paper proposes a digital twin-driven reinforcement learning framework that integrates adaptive energy management, predictive health monitoring, and physics-informed virtual modeling within a unified closed-loop architecture. The framework combines vehicle dynamics, battery modeling, remaining useful life estimation, and reinforcement learning-based torque control with an adaptive digital twin that continuously synchronizes virtual and physical system states through online parameter updating. The proposed approach is evaluated under urban, highway, and aggressive driving conditions using a MATLAB/Simulink co-simulation environment and compared with conventional PI and rule-based control strategies. The digital twin achieves root-mean-square errors of 0.0825 m/s for vehicle speed and 0.0603% for battery state-of-charge estimation. Furthermore, the proposed framework reduces energy consumption by 19.6% compared with PI control and by approximately 12% compared with rule-based energy management while improving tracking accuracy and dynamic response. The results demonstrate an effective framework for energy-efficient, health-aware, and intelligent software-defined electric vehicles.