Risk-Aware Digital Twin-Enabled Edge Intelligence for Battery Degradation, Thermal Stress, and Energy Management in Electric Vehicles

Energy-management strategies are needed to coordinate efficient operation, thermal-stress limitation, and battery-health considerations in electric-vehicle (EV) battery systems under varying operating conditions. This paper develops a risk-aware EV battery energy-management framework based on a Digital Twin, combining battery-state prediction, a heuristic uncertainty indicator, thermal-risk assessment, and battery-health information within a unified current-control scheme. The Digital Twin supplies battery-state information to the controller, while the edge-intelligence layer supports local processing without continuous cloud interaction. Five controllers—Rule-Based, Model Predictive Control (MPC), AI-only, Digital Twin-based, and the proposed risk-aware Digital Twin-enabled controller—are evaluated under common simulated conditions. In the nominal simulation, the proposed controller consumes 24.494 kWh, compared with 22.534 kWh for the Rule-Based controller and approximately 25.35–25.54 kWh for the MPC, AI-only, and Digital Twin-based controllers. Its simulated peak temperature is approximately 22.5 °C, compared with approximately 22.7–22.8 °C for the other controllers; however, the small difference is model-dependent and is not interpreted as statistically established thermal superiority. The proposed controller does not achieve the best SOC prediction RMSE, and its higher SOC RMSE illustrates a trade-off between prediction accuracy and risk-aware control. The reported host-side MATLAB latency is approximately 0.00233 ms and is treated only as a preliminary computational indicator, not as an embedded automotive execution benchmark. The results therefore support the feasibility of studying Digital Twin prediction and uncertainty-aware decision-making in a simulation-based EV battery energy-management framework, while the limitations of the reduced aging model, nominal thermal range, statistical evidence, and simulation-only validation are explicitly acknowledged.