Using Deep-Learning based Probabilistic Forecasting for Multi-Use Operation of Battery Energy Storage Systems

This work investigates the representation of uncertainties in optimizing the multi-use operation of stationary Battery Energy Storage System (BESS) across multiple electricity markets, including the Day Ahead (DA), Intraday Auction (IDA), and Intraday Continuous (IDC) market. Uncertainties in price developments are addressed by integrating forecasting models into a data-driven uncertainty-aware optimization framework and comparing results with a deterministic optimization approach using real data.The study evaluates forecasting models trained on historical time-series data, including deep learning and weighted ensemble methods. Model performance is assessed using error metrics, revealing that no single model performs best across all markets. It proposes different data-driven risk-aware strategies for handling the uncertainty in the objective function of a sequential decision-making process for the multi-use operation of a BESS.Probabilistic forecasting optimization demonstrates that integrating uncertainties through probability distributions, combined with forecasting models, enhances practical applicability and results in higher economic viability by 14.09%.