Predictive optimization of a battery and flywheel storage system for cost-effective reduction of peak electricity demand

From both economic and sustainability perspectives, the effective operation of modern storage systems is increasingly important. The ETA Factory at TU Darmstadt operates a hybrid energy storage system comprising a lithium-ion battery and a flywheel for research. In this work, a model predictive control (MPC) approach is developed for peak shaving that minimizes electricity costs while accounting for storage losses and degradation. To reduce computational complexity, the MPC optimization is implemented in both single-stage and two-stage mixed-integer linear programming (MILP) formulations. Forecasting is achieved via a transparent monthly median-profile method and a lightweight machine learning (ML)-based forecast. Using 30-s power measurements from April–September 2021, the ML-based forecast achieves the highest accuracy among deployable forecasts and delivers robust closed-loop economic performance, while the median forecast remains a low-overhead, interpretable fallback. Across all cases, the two-stage MILP further reduces computation time and improves the cost outcome. In this study, total electricity costs are reduced by up to 47.07% with a perfect forecast, 22.67% with the ML-based forecast, and 14.15% with the median forecast. • Predictive control and mixed-integer linear programming optimize hybrid storage. • Two-stage optimization lowers compute time with similar cost savings to single-stage. • Perfect forecast saves >43% of monthly costs; median forecast still >26%. • Finite element-based losses, battery aging, and state-of-charge limits modeled