Combined Fractional Derivative Approach for Very Short-Term Photovoltaic Power Forecasting

This paper deals with very short-term forecasting of solar power and proposes a combined approach that leverages the strengths of two established methods from the literature to enhance the accuracy of solar power forecasting, namely the Fractional Linear Predictor and the Caputo Derivative. Both approaches are based on fractional derivatives, which inherently account for memory effects and past behaviour, characteristics that contribute to improved accuracy and predictability in solar power forecasting. A proper weighted combination of the methods enhances their different performance in capturing the variability of solar irradiance and, consequently, improves the accuracy of PV power production forecasting. Numerical applications are carried out, including comparative analysis among the methods. Evaluation through metrics, including mean absolute error and root mean squared error, demonstrates the accuracy of the proposed approach.