Improving Solar Energy Prediction in Complex Topography Using Artificial Neural Networks: Case Study Peninsular Malaysia

This research assesses the feasibility of using artificial neural networks (ANN) to predict and improve the spatial distribution of solar radiation data, using Peninsular Malaysia as a case study. This peninsula has seas to the east and west that control cloud formation and rain throughout the year. A rugged mountain range bisects the length of the peninsula creating a complex topography. These features make it difficult to develop effective empirical solar radiation models to cover large areas in Peninsular Malaysia. In this article, several different solar radiation prediction models were designed using the ANN tool in MATLAB. Geographical and meteorological data from 24 solar energy stations were used to predict the solar radiation in 341 cities. Standard multilayer, feed-forward, and back-propagation neural networks were used for the 12 solar radiation models with different numbers of neurons, training functions and activation functions. Predicted solar radiation results were actively used to develop monthly solar radiation maps. The results show that the mean absolute percentage error is less than 6.07% for both the training and testing datasets. This shows that the models are highly reliable predictors of solar radiation values, even in the selected locations that have deficient or unavailable solar radiation databases. The maps show that Peninsular Malaysia receives a monthly average daily solar radiation of between 3.82 and 5.23 kWh/m(2)-day, and that the extreme northern region in Peninsular Malaysia has the highest solar radiation intensity throughout the year. 2015 American Institute of Chemical Engineers Environ Prog, 34: 15281535, 2015