Solar radiation prediction model based on Empirical Mode Decomposition

Accurate solar radiation data are a key factor in Photovoltaic system design and installation. Efficient solar radiation time series prediction is regarded as a challenging task for researchers both in the past and at present. This paper deals with solar radiation time series prediction. To date an essential research effort has been made and various methods are proposed that have different mathematical backgrounds, such as artificial neural networks, fuzzy predictors, evolutionary and genetic algorithms. In the present study the solar radiation time series prediction combines the Empirical Mode Decomposition (EMD) and Support Vector Regression (SVR) models. The EMD is an adaptive signal processing technique that decomposes the nonstationary and nonlinear signals into a set of components with a different spatial frequency content. It results in a small set of new time series that are easier to model and predict. The SVR is applied to the new solar radiation time series. Since support Vector Machines provide great generalization ability and guarantee global minima for given training data, the performance of SVR is investigated. Simulation results demonstrate the feasibility of applying SVR in solar radiation time series prediction and prove that SVR is applicable and performs well for solar radiation data prediction.