Anomaly Detection of Photovoltaic Systems Installed in Renewable Energy Housing Support Project Sites by Analyzing Power Generation Data

Abstract

In this study, we proposed a new method of detecting abnormalities by analyzing power generation data of photovoltaic (PV) systems installed in renewable energy housing support project sites. The study site is north of Gakbuk-myeon, Cheongdo-gun, Gyeongsangbuk-do, Korea, where 63 PV systems have been installed and operated. Based on the system design and surrounding environment, the 63 PV systems were clustered into 6 groups using the K-means clustering method, which is an unsupervised machine learning algorithm. The power production data from the PV systems in each group were analyzed and set as abnormal values if they deviated from the range of ±2.58 times the standard deviation from the mean (assuming a normal distribution and 99% confidence interval). As a result, several abnormalities were detected in the PV systems in November 2020. The cause of the abnormalities was confirmed through site investigation. The proposed method is expected to accelerate the diagnosis of PV systems in renewable energy housing support project sites.