Hybrid VMD-RF Approach for Probabilistic Photovoltaic Power Forecasting

Accurate photovoltaic (PV) power forecasting is essential for ensuring grid stability and enhancing renewable energy integration. This paper proposes a hybrid forecasting approach combining Variational Mode Decomposition (VMD), correlation-based feature selection, and Random Forest (RF) to deliver precise deterministic PV power forecasts. Additionally, probabilistic forecasting methods employing Quantile Regression (QR) and Kernel Density Estimation (KDE) are introduced to quantify forecast uncertainty. First, the volatile PV power time series is decomposed into stable modes using VMD, thereby reducing data complexity. Next, correlation analysis is applied to identify critical meteorological features closely related to PV power output. A deterministic forecasting model is then constructed based on RF, followed by probabilistic modeling using QR and KDE to generate prediction intervals at various confidence levels. Case studies demonstrate that the proposed methodology effectively improves both the accuracy and reliability of PV power forecasts, providing robust decision-making support for photovoltaic energy management and operational planning.