Title:	Modeling a Pothovoltaic Fault Detection Approach Considering Machine Learning for the EU PVSEC 2021
Author(s):	P. Akharath, J. Altkrüger, H. Sahota, V. Herbort, H. te Heesen
Keywords:	Monitoring, PV System, Energy Performance, System Performance
Topic:	PV Systems – Modelling, Design, Operation and Performance
Subtopic:	Operation, Performance and Maintenance of PV Systems
Event:	38th European Photovoltaic Solar Energy Conference and Exhibition
Session:	5CV.2.20
Pages:	1234 - 1237
ISBN:	3-936338-78-7
Paper DOI:	10.4229/EUPVSEC20212021-5CV.2.20
Price:	0,00 EUR
Document(s):	paper

Abstract/Summary:

---

Photovoltaics play a central role in transforming the global energy system into an emission-free energy supply. For example, about 1.8 million installed photovoltaic roof systems in Germany with a nominal output of up to 30 kWp contribute to electrical power generation. Nevertheless, especially in this class of plants, it is evident that technical malfunctions can occur due to a lack of remote monitoring of electricity production and insufficient quality assurance concepts, resulting in a reduction of the yield of these photovoltaic plants [1,2,3]. So far, manual methods have been used to detect such anomalies for unlabeled data; machine learning concepts can provide an automated and adaptive alternative. For this purpose, the time and memory saving anomaly detection approach of Isolation Forests is applied. The results are compared to the ground truth provided by a manual method introduced by Leloux et al. [4]. This evaluation reveals that the approach of Isolation Forest can correctly evaluate the functionality of a plant. Nevertheless, a high rate of undetected anomalies could be caused by a misjudgment by the Ground Truth or the Isolation Forest. Therefore, it must be manually assessed until labeled data is available.