Implementation of convolutional neural networks to determine lightning location

Abstract

Lightning produces not only electromagnetic signals but also thunder sound signal. A new method was proposed to determine lightning location by measure time of arrival (TOA) of lightning electromagnetic signals and thunder sound signal to get lightning distance using Convolutional Neural Network (CNN). These signals were caught by using a set of cell phones. These cell phones collect three main data i.e. the TOA of lightning electromagnetic signals, the TOA thunder sound signal, and the coordinates obtained from Global Positioning System (GPS). In this study, CNN determines thunder sound signal pattern which is recorded by certain cell phone. More than one hundred samples of thunder sound signals were used as data set. Whole data set divide into three categories i.e. training data, validation data, and testing data. The model runs few hours to train the network and then produce a confusion matrix. The matrix consists of several column represents a set of samples that previously predicted by their labels and several rows represents actual labels. The labels are thunder sound and the other sound. After the model created, it is exported to the cell phone so that every time cell phone record a sound signal, the model will predict whether the thunder sound or not. When the model predicts the sound as thunder sound, the cell phones save the exact occurrence time. This determination is very important when the cell phone is going to measure lightning distance between lightning location and cell phone location.