Prediction of rainfall using improved deep learning with particle swarm optimization

Abstract

Rainfall is a natural factor that is very important for farmers or certain institutions to predict the planting period of a plant. The problem is that rainfall is very difficult to predict. Trials to get optimal rainfall prediction have been carried out by BMKG through research with variety of methods in various fields, including meteorology, climatology and geophysics. The results of the study unfortunately obtained a less optimal success rate in predicting rainfall. Today, there are many new methods for predicting events. These methods include Deep Learning (DL) and Particle Swarm Optimization (PSO). The use of the Deep Learning method is very susceptible to initial weights that are less than optimal, so it requires a process of optimization using a metaheuristic technique, which is the PSO algorithm, because this algorithm has a level of complexity that is much lower than genetic algorithms. In this study, this method is utilized to predict rainfall by determining the exact regression equation model according to the number of layers in hidden nodes based on the size of the kernel and the weight between the layers. This research is approved achieved get more optimal rainfall prediction results that those of previous research that without optimization with PSO.