Hybrid meta-EP-ANN technique for lightning prediction under Malaysia environment

This paper presents the development of hybrid Meta-Evolutionary Programming (EP) and Artificial Neural Network (ANN) prediction system for lightning occurrence based on historical lightning and meteorological data from a Malaysian environment. It involved the development of the ANN design and embedding EP optimization technique for optimizing learning rate and momentum constant with the aim to improve the performance of the prediction system in terms of its generalization capability. ANN is employed as it has the ability to learn by example and do tasks based on training experience, making it profoundly suitable for pattern recognition and forecasting tasks. However, due to its heuristic nature, the process of finding suitable network architectures could be burdensome and time-consuming. There is also no systematic way to design the optimal architecture for a given task automatically. For that reason, an efficient optimization technique such as EP is employed in the study to find the best ANN architectures systematically which leads to lightning prediction performed accurately with less computation time. As a result, the developed lightning prediction system is able to generalize well when presented with new sets of input data. Comparative studies conducted with respect to heuristic search indicated that hybrid-meta-EP-ANN is more accurate in its prediction capability.