Power Quality Problem Classification Methods Based on Active Transfer Learning

Abstract

In this paper, we propose approach for classification of power quality problem using active transfer learning. Active learning is suitable for solving problems with real industrial data because even a small amount of labeled data can derive good performance. Active learning can derive performance by querying human data labelers for data the model does not learn or trained. Active transfer learning utilizes transfer learning for data sampling of the trained model. However, querying too much data can put pressure on human data labelers. Thus, we suggest the solution of this problem by limiting query in two ways. One is to limit the number of queries to 100 and the other is dynamically controlled the number of queries according to the accuracy. As a result, we derive better predictive performance with suggest method and compare the results with the traditional method.