A Study of Anomaly Detection Method for Wind Power Generation Facilities Independent of Operating Conditions

Abstract

In wind power generation, early anomaly detection is important from the perspective of stable energy supply. Previous research has focused on detecting anomaly under constant operating condition (e.g., rated operation). However, wind power generation, which uses natural wind, is not always operated under constant operating conditions, and it is difficult to detect anomaly continuously using conventional methods, and only some of the data has been used. Therefore, in this paper, we propose a method to detect anomaly in wind turbine generators regardless of the operating conditions. The proposed method performs unsupervised learning by variational autoencoder (VAE) with convolutional layer using features generated from vibration data of normal gearbox under various load conditions to detect anomaly considering the diversity of normal data. In this paper, we show that the proposed model can detect anomalies independent of the operating conditions of the equipment.