Unsupervised Detection of Non-Technical Losses via Recursive Transform Learning

This letter addresses the problem of detecting non-technical losses in an unsupervised fashion. Most prior studies in this area proposed supervised means (assumed the losses to be labeled); getting supervised data for such a problem seems impractical. For a practical scenario, non-technical losses should be detected in an unsupervised fashion. This work proposes a new dynamical model called recursive transform learning for online unsupervised detection of non-technical losses. Results show that our proposed method is better or at par compared to the state-of-the-art.