A Deep Learning Model for Lightning Location and Peak Current Estimation from Induced Overvoltages

A novel methodology for the determination of lightning return stroke strike position and peak current using Deep Learning (DL) techniques is introduced in this contribution. The underlying concept consists in exploiting the voltage transients produced by indirect lightning strikes on distribution lines as input features for the regression of both the strike location and the peak of the channel base current. Starting from the results of prior investigations, and following a model selection phase, a DL model operating directly in the time domain is proposed. Data generated through the Lightning Power Electromagnetic Simulator for Transient Overvoltages (LIGHT-PESTO) are employed for the training of the model. Initially, the method’s performance is evaluated using the same test scenario presented in a previously published study, showing an improvement in accuracy. Subsequently, the model is applied to a more realistic configuration, i.e., a three-phase distribution line protected by surge arresters, to demonstrate that its effectiveness is preserved.