Best-Fit Machine Learning Classifier for Early-Stage Photovoltaic Hot-Spots Detection

Dhimish, M. (Speaker), Schofield, N. (Contributor to Paper or Presentation)

Activity: Talk or presentation typesOral presentation

Description

Photovoltaic (PV) hot-spots is a reliability problem in PV modules, where a cell or group of cells heats up significantly, dissipating rather than producing power, and resulting in a loss and further degradation for the PV modules’ performance. Therefore, in this article, we present the development of a novel machine learning-based (ML) tool to diagnose early-stage PV hot-spots. To achieve the best-fit ML structure, we compared four distinct machine learning classifiers, including decision tree (DT), support vector machine (SVM), K-nearest neighbour (KNN), and the discriminant classifiers (DC). Results confirm that the DC classifiers attains the best detection accuracy of 98%, while the least detection accuracy of 84% was observed for the decision tree. Furthermore, the examined four classifiers were also compared in terms of their performance using the confusion matrix and the receiver operating characteristics (ROC).
Period5 Nov 2020
Event title10th Solar & Storage Integration Workshop: International Workshop on Integration of Solar Power and Storage into Power Systems
Event typeWorkshop
Conference number10
Degree of RecognitionInternational