Challenges Posed by Renewable Energy Source Integration to Machine Learning Based Power System Fault Diagnosis

Nowadays machine learning (ML)-based power system fault diagnosis is being researched to a great extent. ML techniques have many advantages over conventional fault diagnosis techniques. However, with the advent of renewable energy sources (RES) integration on a large scale it is necessary to study whether ML models will be able to adapt to new RES integration. In this paper, a performance investigation of a few ML techniques has been done for power system fault classification and localization on new solar PV plant integration to the IEEE 9 Bus system. The proposed performance investigation is an impact analysis of new RES integration on ML model performance when fault data for RES integrated system is unavailable. Thus, fault classification and localization of RES integrated power system faults are predicted from pre-trained ML models, that were trained from conventional power system fault data. The results revealed that ML models' performance degrades severely with RES integration.

I. Introduction

The use of renewable energy sources (RES) to reduce power generation via coal-based power plants for sustainable devel-opment and reducing greenhouse gas emissions is increasing worldwide [1]. Renewable energy has gained popularity as energy demand rises and fossil fuels become less abundant. Renewable energy technologies include solar, wind turbines, microturbines, tidal, and geothermal power plants with capac-ities ranging from 1 KW to several MW [2]. Integrating RES plants into conventional power systems (CPS) requires optimal sizing and placement study for overall voltage stability of the network and minimization of power loss. Many works have already been reported in the literature for optimal sizing and placement [3]. Another heated topic for RES integration is power generation forecasting. This helps to estimate power generation from RES plants for short, medium, and long-term that help in planning day-to-day operations, and scheduling maintenance [4]. However, one of the neglected aspects of RES integration is the impact of new RES integration on fault classification and localization techniques.