Data Driven Artificial Neural Network Fault Classification by Bayesian Optimization

Abstract

Intelligent fault diagnosis is a method that involves the utilization of advanced computational tools, which are frequently based on artificial intelligence (AI) and machine learning, in order to diagnose defects or malfunctions in systems or processes with the purpose of resolving issues. As a means of achieving the goal of preventing catastrophic failures by the early detection and resolution of defects, the optimization of hyper-parameters in machine learning models is an essential stage in the process of achieving higher performance and generalization. An intelligent fault detection technique for micro-grids that is based on Artificial Neural Network and is driven by data is presented in this study. The technique makes use of a model that utilizes hyper-parametrical tuning to optimize Machine learning (ML) performance. Within the framework of the suggested method, three-phase currents and voltages are utilized as variable features that are fed into classifier layers, using hyper-parameters that have been optimized. The Bayesian optimization strategy is utilized during the model learning phase in order to maximize the hyper- parameters. Last but not least, multi-class classification of the five distinct types of faults can be accomplished through the utilization of a ML classifier model whose hyper-parameters has been optimized.