Hourly irradiance forecasting for Peninsular Malaysia using dynamic neural network with preprocessed data

Abstract

Accurate irradiance forecasting is one of the essential factor that helps facilitate the proliferation of grid-connected photovoltaic (GCPV) integration. In Malaysia, this topic has not been substantially explored. This paper attempts to investigate the use of neural network by using data obtained from meteorological condition measurement in Sepang, Malaysia to forecast hourly values of solar radiation. The data is preprocessed to eliminate defective values and help achieve convergence in a faster and reliable manner. The methodology uses Nonlinear Autoregressive (NAR) network which utilises historical irradiance values of annual, quarterly, and monthly durations to predict future hourly irradiance. The result shows that the NAR network can predict hourly irradiance with satisfactory result and, in order to produce better forecasting, longer data timeframes is preferable.