An Insolation Forecasting Method by Partial Least Squares and a Confidence Interval Estimating Method

Abstract

This paper presents an insolation forecasting method using numerical weather forecasting data and a confidence interval estimating method. It is planned to introduce photovoltaic (PV) system on a large scale into the power system in order to achieve the low carbon society. Thus, it is necessary to forecast the insolation and evaluate the confidence interval in terms of power system operating and planning. This paper proposes the insolation forecasting method by Partial Least Squares (PLS) and the confidence interval estimating method considering the forecasting error distributions. PLS handles multicollinear data among input variables in order to construct proper forecasting models. The proposed method can estimate continuous confidence intervals by linearly-approximating for representative values of some weather conditions extracted features from forecasting error distributions. The effectiveness of the proposed method is demonstrated using numerical weather forecasting data and actual data observed by Japan Meteorological Agency. The proposed method by PLS can forecast the hourly insolations more accurately than conventional multiple regression equation. Moreover, the proposed method can evaluate continuous confidence interval.