Real-Time Self-Learning Digital Twin for Lithium-Ion Battery Energy Storage Systems in Smart Grids

In this paper, we propose a self-learning digital twin (SLDT) architecture that incorporates real-time battery degradation modeling and optimum operational management for grid-scale lithium-ion battery energy storage systems (BESS). This work extends the Adaptive Real-Time Degradation Model (ARDM) framework to allow real-time updates of the parameters based only on live operational data without pre-cycling experiments and further improves its robustness under various depth-of-discharge (DoD), charging/discharging current (C-rate), and temperature conditions. The ARDM is incorporated in a real-time digital twin that maintains synchronized health, state of charge (SoC), and degradation cost predictions. The digital twin is linked to an Optimization and Control Layer (OCL), which plans the charge/discharge day-ahead in advance based on dynamic power rates. The Musical Chairs Algorithm (MCA) is used for parameter identification and scheduling due to its better convergence characteristics compared to swarm-reduction forms of benchmark optimization algorithms. Experimental validation is carried out on two commercial 48 V Li-ion modules with various cycling patterns, and sub-millipercent root-mean-square error (RMSE) is achieved in capacity-fade tracking. The economic analysis for a 5-MW/10-MWh system indicates that dynamic tariff scheduling results in about nine times greater arbitrage revenue compared to fixed rates, 41–58% higher yearly net income, and lower degradation costs. The results confirm that the SLDT is a practical and accurate platform for degradation-aware operational planning in modern smart-grid environments.