Solar irradiance modeling and forecasting using novel statistical techniques

Abstract

This thesis focuses on the solar irradiance modeling and forecasting through statistical approaches. A local clear sky model is first developed. Univariate and multivariate time series forecasting techniques including autoregressive integrated moving average model, time-forward kriging and vector autoregressive model are explored and extended in a solar engineering context. To facilitate the spatio-temporal data collection, inverse transposition models are proposed to convert solar irradiance on a tilted plane to that on a horizontal plane. The transposed stationary spatial-temporal data can thus be used in forecasting. Finally, two new parameter shrinkage methods are proposed based on the threshold distance in the irradiance random field. Such practice not only increases forecast accuracies but also significant reduces the computation burden owing to the large covariance matrix. As PV systems proliferate and become an important part of the global energy mix, the above mentioned works provide valuable information to the electricity grid operators.