A double decomposition-based modelling approach to forecast weekly solar radiation.

To meet the future energy demand and avert any looming crises, efforts are being carried out to utilize sustainable and renewable energy resources. In this paper, the naturally occurring non-linearity and non-stationarity deficiencies within the climatological predictors to forecast solar radiation (Rdn) are resolved via a multivariate empirical mode decomposition method (MEMD). First, a set of antecedent weekly lags at timescale (t -1) of input datasets were collated and then were divided into training and testing subsets. The MEMD method is restricted to dissolve the training and testing climatic data independently into intrinsic modes functions (IMFs). As the numbers of total IMFs were very large, the singular value decomposition (SVD) algorithm is accustomed for dimensionality reduction simultaneously capturing the most relevant oscillatory features embedded within the IMFs. Finally, the random forest (RF) model is applied to forecast Rdn at selected solar-rich regions in Australia. The resulting hybrid MEMD-SVD-RF model was established as a consequence of the aforementioned modelling strategy. The results are benchmarked with other comparative models. The hybrid MEMD-SVD-RF model generates better and reliable forecasts having significant implications for renewable and sustainable energy applications and resources management. • A double decomposition based solar radiation forecasting model incorporating climate drivers. • Multivariate decomposition splits climate drivers into intrinsic mode functions (IMF). • Singular value decomposition reduces dimensionality to pertinent features from a pool of IMFs. • Random Forest (RF) algorithm used to design newly integrative hybrid model. • Hybrid model provides significant energy management implications. [ABSTRACT FROM AUTHOR]