Unsupervised Anomaly Detection Method Based on Deep Learning and Support Vector Data Description

Anomaly detection in unlabelled and highly imbalanced high-dimensional monitoring data is one of the most urgent and challenging industry problems in the energy industry. Based on the powerful high-dimensional data analysis capabilities of autoencoders, the use of autoencoders for anomaly detection is becoming more and more popular. This paper proposes an anomaly detection method based on deep learning and support vector data description. First, feature engineering is built based on an optimized serial deep autoencoder; second, different feature combinations are studied and compared; finally, anomaly detection based on support vector data description. In this paper, experiments are carried out on the actual operating data of a real steam turbine to verify the effectiveness and accuracy of the proposed method.