Data-Driven Advanced Forecasting, Planning, and Market Operation of Hybrid PV-BESS Power Plant

Abstract

This study aims to develop a unified, data-driven framework that integrates probabilistic photovoltaic (PV) forecasting, risk-aware battery energy storage system (BESS) planning, and market-oriented operational control under uncertainty. The research addresses the following key questions: (i) how forecast uncertainty can be effectively utilized in operational decision-making, (ii) how it influences optimal BESS sizing, and (iii) how it affects economic performance and risk in electricity markets. The proposed framework enables end-to-end propagation of probabilistic information across decision layers, transforming forecast uncertainty into a quantifiable decision variable. The framework is validated using real data from a utility-scale PV plant in Finland. Probabilistic forecasting based on quantile regression achieves reliable calibration (prediction interval coverage ≈ 0.88–0.93). Forecast accuracy degrades with increasing prediction horizon defined as day-ahead (D1), two-day-ahead (D2), and three-day-ahead (D3) with root mean square error (RMSE) increasing from approximately 8.4 MW (D1) to 12.9 MW (D3). This degradation significantly increases the probability of extreme imbalance events, referred to as tail-risk exposure. Incorporating this uncertainty into planning, a Conditional Value-at-Risk (CVaR)-based optimization where CVaR represents the expected loss under worst-case conditions consistently identifies a compact BESS configuration within the evaluated design space and considered market conditions (2 MW / 2 MWh). This design maximizes economic value (≈ €4.18 M annual profit, ≈ €35 M net present value (NPV), where NPV represents long-term investment profitability) under controlled risk (≈ €2.38 M CVaR), indicating that optimal sizing is governed by marginal value saturation rather than capacity scaling. Operational results show that uncertainty-aware dispatch reduces extreme imbalance exposure and reserve volatility, at the cost of moderate reductions in short-term profit (≈10–20%) and increased curtailment. A critical nonlinear regime is observed at intermediate horizons (D2), where imbalance cost peaks (~450 k€), indicating maximum sensitivity to forecast error. Long-term evaluation demonstrates that optimized control reduces battery cycling by approximately 50%, significantly extending asset lifetime. The main contribution of this study is the development of a unified, decision-oriented PV–BESS framework that explicitly links forecast uncertainty to economic value, operational risk, and storage efficiency. The results demonstrate that optimal system performance emerges from selective, risk-aware utilization of flexibility rather than maximum energy throughput, providing a scalable foundation for real-time digital twin applications in renewable-dominated power systems.