dTASD: A Novel Online Detection Method for Anomalous State of Dry-type Transformer

Abstract

Due to the advantages of dry-type transformers such as safety, no pollution, and low power consumption, they are widely used in shopping malls, hospitals, data centers and other places. Therefore, anomalous state detection for dry-type transformers is of great significance. However, the traditional detection methods are generally based on the hard threshold judgment method, which is difficult to ensure timeliness and may cause irreversible damage to the device. In this paper, we present dTASD, Dry-type Transformer Anomalous State Detector, a framework that can timely detect the anomalous state of dry-type transformer online. dTASD consists of an offline training model stage and online detecting stage. In offline training model, dTASD adopts the semi-supervised mode, and applies self-organizing map to discretize the three-phase temperature data to solve the challenge of three-phase data fusion. In online detecting, we propose a novel calculation method for anomaly scores to measure the degree of transformer operation deviating from the normal state. The experimental results using Real monitoring data of dry-type transformers installed in a large data center demonstrate dTASD can effectively solve the problem of anomalous state detection for dry-type transformers, and outperforms the existing anomaly detection approaches.