Application of improved version of multi verse optimizer algorithm for modeling solar radiation. (November 2022)

• Record Type: Journal Article
• Authors: Ikram, Rana Muhammad Adnan; Dai, Hong-Liang; Ewees, Ahmed A.; Shiri, Jalal; Kisi, Ozgur; Zounemat-Kermani, Mohammad
• Is Part Of: Energy reports. Volume 8(2022)
• Page Start: 12063
• Page End: 12080
• Publication Date: 2022-11
• Subjects: Solar radiation prediction -- Least square support vector machine -- Genetic algorithm -- Gray​ wolf optimization -- Sine–cosine algorithm -- Multi-verse optimizer -- Improved multi-verse optimizer; Power resources -- Periodicals; Energy industries -- Periodicals; Power resources; Periodicals; Electronic journals; 621.04205
• DOI: 10.1016/j.egyr.2022.09.015
• Languages: English
• ISSNs: 2352-4847

Abstract

Abstract: For better estimation of renewable environmental friendly and carbon-free energy resources, precise prediction of solar energy is very essential. However, accurate prediction of solar energy is a challenging task due to its fluctuations and due to climatic factors those make it very nonlinear in nature. Therefore, in this study, the novel robust soft computing method is applied to predict solar​ radiation of two stations located in the southeast region of China. For modeling solar radiation of selected stations, the improved version of multi verse optimizer algorithm (IMVO) is utilized with integration of least square support vector machine (LSSVM) for better tuning the hyperparameters of LSSVM model. For validation, the newly developed method is compared with other algorithms integrated with LSSVM models, such as LSSVM with genetic algorithm (LSSVM-GE), LSSVM with gray wolf optimization (LSSVM-GWO), LSSVM with sine–cosine algorithm (LSSVM-CSA) and LSSVM with multi verse algorithm original version (LSSVM-MVO). It is found that newly developed method, LSSVM-IMVO, provided more accurate results in comparison to other models. For better visualization of data and model application, three different training testing data splitting strategies are used. It is found that the increase in training sample size considerably improved the models' accuracies.