Artificial neural network forecast of monthly mean daily global solar radiation of selected locations based on time series and month number. (10th April 2019)

• Record Type: Journal Article
• Authors: Ozoegwu, Chigbogu Godwin
• Is Part Of: Journal of cleaner production. Volume 216(2019)
• Page Start: 1
• Page End: 13
• Publication Date: 2019-04-10
• Subjects: Renewable energy assessment -- Global solar energy -- Time series -- Artificial neural networks; Factory and trade waste -- Management -- Periodicals; Manufactures -- Environmental aspects -- Periodicals; Déchets industriels -- Gestion -- Périodiques; Usines -- Aspect de l'environnement -- Périodiques; 628.5
• DOI: 10.1016/j.jclepro.2019.01.096
• Languages: English
• ISSNs: 0959-6526

Abstract

Abstract: Prediction and forecast of monthly mean daily global solar energy available at selected locations are carried out using a nonlinear autoregressive, a nonlinear autoregressive (exogenous) and a hybrid time series methods. The methods are implemented with artificial neural networks. The proposed hybrid method is based on a combination of nonlinear autoregressive and structural artificial neural networks where the structural input is the current month number. The aim of the hybridization is to improve prediction accuracy of the nonlinear autoregressive methods and simplify the input layer of the nonlinear autoregressive (exogenous) method. Based on various statistical criteria, the hybrid method is verified to predict and forecast global solar energy with noticeably higher accuracy. For example, while a nonlinear autoregressive predicts the global solar energy availability at Abuja with R 2 -value of 0.78 and correlation coefficient of 0.89, the hybrid method improves the prediction to 0.96 and 0.98. The hybrid model is verified via one-way analysis of variance to consistently perform better than the nonlinear autoregressive method in months ahead forecasting of solar energy. The proposed hybrid method is capable of long-term forecast of up to two years ahead within a typical mean percent forecast error of 5.67%, therefore, is applicable for designing/planning solar energy integration and simulation of agricultural food security, especially in developing countries. A cases study of power output and sizing of standalone PV options showed that decisions based on forecasted and actual solar energy are the same demonstrating the very useful application of the hybrid method. Highlights: ANNs based on time series inputs were used to predict monthly mean daily global solar energy. A proposed hybrid ANN was shown to predict with higher accuracy than NAR and NARX. The hybrid ANN is capable of a long-term forecast of up to two years ahead. The hybrid ANN predicts and forecasts reliably across different climatic regions of Nigeria. The solar energy forecast can be used in deciding future power output and sizing of solar systems.