Design and performance of two decomposition paradigms in forecasting daily solar radiation with evolutionary polynomial regression: wavelet transform versus ensemble empirical mode decomposition

Design and performance of two decomposition paradigms in forecasting daily solar radiation with evolutionary polynomial regression: wavelet transform versus ensemble empirical mode decomposition

Due to highly destructive effects of on the environment and the steep growth in the global energy demands, renewable energy resources have been the focus of researchers. Solar energy, a renewable energy resource, is available all over the world. In this study, DSR has been forecast using an AI approach called EPR. To achieve this goal, six daily inputs (i.e., TA, RH, VP, SLP, PE, and SD) and one output (DSR), measured from 2000 to 2016, have been decomposed into new variables using two preprocess processes, WT and EEMD.

The results of EPR, WT-based EPR, and EEMD-based EPR models have been compared using comparative statistics containing NSE, RMSE, MAE, WI, and Legates-LMI. A holistic evaluation via statistical assessment and diagnostic plots indicates that the EEMDEPR model generates superior forecasting compared with the standalone EPR model and WT-based EPR models. The comparison reveals that the EEMD-EPR model provides the best performance at Seoul and Busan (calibration and validation stages) stations. The performances of single EPR and hybrid EPR models are evaluated based on the error size and the uncertainty analysis of model forecasting. The forecasting errors and uncertainties associated with the proposed EEMD-EPR model are smaller than those associated with the W-H-EPR, W-D-EPR, and W-C-EPR models.

Files associated with this item cannot be displayed due to copyright restrictions.