Adoption of Ensemble Empirical Mode Decomposition Algorithm and Back Propagation Neural Network in Net Surface Solar Radiation Prediction

Abstract

Abstract To improve the prediction accuracy (PA) of net surface solar radiation (NSSR), a net surface solar radiation (NSSR) prediction model, named EEMD-BPNN, is proposed by adopting the ensemble empirical mode decomposition (EEMD) algorithm along with back propagation neural network (BPNN). In this paper, EEMD is used to extract the signals to reduce the influence of noise with physical significance from the time series of the original NSSR, so as to obtain the intrinsic mode functions and residual terms of different frequencies. As well as BPNN is used to establish a corresponding prediction model for each component of mode function. The proposed model has been applied and verified in test the daily total NSSR in Aksu region, Xinjiang. In the case study, the mean percentage error (MPE), mean bias error (MBE), root mean square error (RMSE), and correlation coefficient are taken as evaluation indexes, and the accuracy and applicability of the EEMD-BPNN for NSSR prediction are analyzed by comparing the prediction results of the EEMD-BPNN with the BPNN and H-S model. The results show that the predicted values (PVs) of EEMD-BPNN are closer to the actual data and have better correlation coefficient (R2=0.9615) compared with the prediction results of BPNN (R2=0.8703) and H-S model (R2=0.8373), and the error analysis indexes of the predicted results are all small. It indicates that the PA of EEMD-BPNN is improved obviously, which means the EEMD-BPNN has superiority in NSSR prediction and provides a new reference method for NSSR prediction.