An adaptive interval power forecasting method for photovoltaic plant and its optimization

With the high photovoltaic (PV) access ratio, high precision PV power prediction is of great significance for the large-scale PV plants. The existing deterministic prediction methods are not completely effective in dispatching decision making. PV output power and its prediction error have obvious non-linearity and fluctuation, the fixed model is not capable, which creates unstable performance in PV power forecasting. The PV power interval forecasting method with dynamic adaptability provides a new way to solve the above problems. The main work of this paper is as follows: The forecasting interval and error distribution of PV output power are analyzed, which shows obvious differences with time. A self-adaptive model is then established to calculate the PV power forecasting interval by using the kernel density estimation algorithm. The optimization setting of dynamic time window length and kernel density estimation window width, the advantages of dynamic interval method compared with fixed method are illustrated through experimental verification of. The innovation of this paper is to find the seasonal distribution characteristics of photovoltaic power prediction error, and a dynamic interval prediction method is proposed. The verification shows that the proposed optimization method has 5% PICP improvement than other methods.