Forecasting the Photovoltaic Power Output via Two-Dimensional Time-Series Imaging

Photovoltaic (PV) power systems have experienced significant expansion in recent years. However, the inherent intermittency of PV power presents challenges to grid reliability, operational planning, stability, and economic viability. Accurate PV power forecasting is essential to ensure secure and stable grid operation. This paper introduces an image-based approach for photovoltaic power forecasting, leveraging U-Net to enhance prediction accuracy and mitigate solar variability. PV time-series data is first encoded into a 2D representation, then processed through U-Net to predict the next frame, and finally reconstructed into time-series format to generate the PV forecast. We evaluate our proposed method against five well-established benchmark models. The results show that our approach effectively captures the temporal and seasonal patterns in PV data, making it well-suited for forecasting across all four seasons. The model achieves superior performance across all accuracy metrics, with RMSE and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex> scores as follows: Winter (RMSE: <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1.4519, R^{2}: 0.92849$</tex>), Spring (RMSE: 0.6375, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}: 0.98460$</tex>), Summer (RMSE: 0.1031, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex>: 0.99984), and Autumn (RMSE: <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.6293, R^{2}: 0.99032$</tex>).