Solar irradiance short-term prediction under meteorological uncertainties: survey hybrid artificial intelligent basis music-inspired optimization models

Abstract

Short-term predictions of solar radiation are essential for enhanced management and control of solar energy systems. However, developing artificial intelligent models is often a challenging task due to the high variances in solar and meteorological data records. In this study, several hybrid artificial intelligent models are discussed for predicting the global horizontal irradiance in 5-min time intervals for four stations located in the Northern Territory of Australia. The multilayer neural network (MLNN) is hybridized with six music-inspired optimization algorithms, namely harmony search, improved harmony search, global-best harmony search (GHS), improved GHS (IGHS), Gaussian GHS (GGHS), and dynamical GHS (DGHS) algorithms for comparison of solar radiation prediction. The GGHS and DGHS are developed using two adjusting processes, which are applied to provide the new position of the particles. The results show that GGHS- and DGHS-MLNN models provide superior predictions in terms of accuracy and tendency, compared to the other harmony search-based models, as well as the corresponding MLNN models, conventionally trained using the backpropagation algorithm, at the four studied locations. The recommended DGHS-MLNN model was further evaluated under different sky conditions, where all root mean square errors and mean absolute errors were found to be lower than 76 and 57 W m−2, respectively.