Stability-Assured ARMA-Kalman Filter Based Adaptive Short-Term Solar PV System Forecasting

Photovoltaic (PV) plant output forecasting is a key factor for the efficient operation and decision-making of modern renewable energy-based systems. Numerous research has focused on improving prediction accuracy; however, these models often fail to maintain performance over time due to changing system dynamics, necessitating frequent retraining. Retraining hampers real-time application as it is a time-consuming process. Furthermore, forecasting models are typically selected based on performance accuracy, which does not ensure long-term stability. Since these models are not mathematically robust, their performance on future unknown data remains uncertain. This paper proposes a stability-guaranteed ARMA-Kalman filter-based shortterm forecasting model. The ARMA model develops the forecasting framework, while the Kalman filter is employed to predict and update the states to enhance prediction accuracy. The models developed by the ARMA approach are selected based on spectral radius stability analysis for discrete systems. When applied to multiple-horizon solar PV plant forecasting tasks (15, 30, and 60 minutes), the developed algorithms demonstrated that smaller time horizons yield better accuracy with smaller datasets. In contrast, longer forecasting horizons require larger datasets, resulting in lower accuracy than shorter ones. This approach ensures stability and improved PV forecasting performance, addressing traditional models’ limitations.