A Comparative Assessment of Time Series Forecasting Using NARX and SARIMA to Predict Hourly, Daily, and Monthly Global Solar Radiation Based on Short-Term Dataset

Abstract

Several hourly, daily and monthly global solar radiation prediction models have been designed, to overcome the weakness of the previous models. Many previous models have used a long-term global solar radiation of Seoul to predict the consecutive years. Unfortunately, many countries do not have an enough history to build such prediction models, in addition many researchers suggested that seasonal autoregressive integrated moving average (SARIMA) is better than nonlinear autoregressive exogenous (NARX) neural network in predicting global solar radiation. Therefore, this research comes to fill the gaps in previous work, develop prediction model based on short-term global solar radiation, and test the best model between NARX and SARIMA by using global solar radiation of Seoul. The methodology divided the developed models into two parts including train phase and test phase. Train phase used dataset between 2007 and 2013, where test phase used dataset between 2014 and 2015. Afterward, the developed models are validated and tested using determination coefficient (R2) and different error function and the results are compared to two previous model that used long-term dataset namely ANFIS model and SARIMA. The results showed that the determination coefficient (R2) and RMSE of NARX model based on hourly data are 0.95 and 0.23 MJ/m2, respectively, besides the best daily and monthly average solar radiation predictors are obtained when NARX and hourly data are used. The results revealed that using hour, day, month and year as independent variables and less history with NARX model is efficient to predict two consecutive years.