Point and Probabilistic Forecasts for Day-Ahead Battery Scheduling

The global energy transition towards renewable energy sources (RES) poses challenges due to their inherent uncertainty, which reduces the energy system's flexibility and stability. Battery energy storage systems (BESS) offer a solution by increasing flexibility and making aggregated RES and loads dispatchable, enabling them to meet day-ahead dispatch schedules communicated with the grid through effective BESS management. This concept is implemented in a so-called dispatchable feeder (DF), which represents a system consisting of RES generation, loads, and a BESS, working together as a single controllable unit. The operation of the DF is formulated as a stochastic optimisation problem (OP) using forecasts of RES generation and loads as inputs. This thesis focuses on two main aspects affecting the performance of the DF: the forecasts and the mathematical formulation of the OP. In the first half of this thesis, the impact of different point forecasts with different characteristics on the performance of the DF via the so-called forecast value is investigated. Using a dataset of 300 residential buildings, the results show that the forecast characteristics affect the performance of the DF depending on problem-specific parameters. Based on these results, a meta-learning framework is proposed that automates the selection of the value-oriented forecast for the DF. The evaluation shows that the proposed framework outperforms existing forecast method selection heuristics in performance and accuracy. Additionally, it achieves comparable results to a manual selection with noticeably lower computational effort. In the second half of this thesis, probabilistic forecasts with different characteristics are evaluated with respect to the forecast value for the DF, showing again that the forecast characteristics in combination with the problem-specific parameters affect the performance of the DF. Additionally, a head-to-head performance comparison is made between the DF considering point forecasts and the DF considering probabilistic forecasts. This evaluation is carried out using both standard and the best value-oriented forecasts resulting from the aforementioned forecast evaluations. This allows for a comprehensive analysis of the impact of the forecast characteristics and the integration of the uncertainty of the RES generation and loads on the DF's performance. Finally, a modification to the mathematical formulation of the DF considering probabilistic forecasts is introduced, incorporating a line restriction to prevent line overloads in low-voltage distribution grids. Evaluations show that the modified DF can improve line usage and mitigate grid overloads.