Anomaly Detection of Mountain Photovoltaic Power Plant Based on Spectral Clustering

Abstract

Mountainous photovoltaic (PV) power plants cover a large area and are distributed dispersedly. The construction surface is complex and the slope is large. It is difficult to find and locate faults when dealing with defects. Effective anomaly detection and fault location technology can not only improve the reliability and stability of the power plant but also reduce the operation and maintenance costs. Because of the terrain and the influence of construction, the radiation received by each group string at the same time point may be different, which brings great difficulties to anomaly detection. In this article, a method based on spectral clustering is proposed for anomaly detection of mountain PV power plants. The data is preprocessed by filtering and reconstruction. By sorting the time series current data, the influence of dip angle and azimuth is eliminated to a certain extent. Further, the normal and abnormal strings can be distinguished by spectral clustering. Using current time series data, the proposed method can detect anomalies effectively in different weather without additional irradiance and temperature data. Moreover, since the high-dimensional data can be processed by spectral clustering, the process of feature extraction can be greatly simplified. The proposed algorithm is validated by the actual data of PV power plants, which nicely verifies that the anomaly diagnosis method can effectively identify anomaly clusters.