Akademik Veri Yönetim Sistemi
14th INTERNATIONAL CONFERENCE ON RENEWABLE ENERGY RESEARCH AND APPLICATIONS, Vienna, Avusturya, 27 - 30 Ekim 2025, ss.1-6, (Tam Metin Bildiri)
Accurate ultra-short-term forecasting of photo voltaic power is essential for grid stability and market operations. This study establishes a reproducible framework for ultra-short term photovoltaic power forecasting, electricity price modeling, and anomaly detection using the Open Power System Data time series dataset. The dataset includes multivariate series relevant to power system modeling, such as electricity prices, demand, and solar generation capacities. Photovoltaic power forecasting models are evaluated across 1-h, 24-h, and 168-h horizons. In addition, 24-step-ahead trajectories are generated using an iterative forecasting approach. On the price modeling task, the Ridge regression baseline is benchmarked against tree-based alternatives. Model performance is assessed using MAE, RMSE, sMAPE, and R2 metrics. Results from the most recent seven days show that tree-based models consistently outperform persistence baselines. They capture intraday ramp-up and ramp-down dy namics more accurately, with marked error reductions during peak load periods and spike days. These findings demonstrate the potential of tree-based models to improve operational decision making in renewable-integrated power markets.