Artificial Neural Network Based Technique for Lightning Prediction

Abstract

Malaysia has high lightning and thunderstorm occurrences throughout the year. A vast amount of its data have been recorded which allows various lightning-related studies to be conducted. This paper presents the application of artificial neural network (ANN) in predicting the occurrence of lightning events based on historical lightning and meteorological data. ANN, which was inspired by the way biological nervous systems process information, is utilized in this study due to its strong pattern recognition capabilities; implemented through learning patterns and relationships in data. A two layer back-propagation neural network has been developed to predict the occurrence of lightning at least four hours prior to its arrival. Several network structures, training algorithms and activation functions have been rigorously tested in order to obtain the most suitable network with high accuracy and convergence capability, while the perfection of the developed network was conducted through postprocessing, indicated by the closeness of correlation coefficient to unity. The computation burden experienced in this study in achieving the converged solution has been alleviated by the introduction of indicator module to the original features of the training and testing patterns.