Mohammad Hossein Zarei
Leila Rahman Setayesh
Hamidreza karimifard
Mohammad Mohammadi
Ayoub Ravan

This article described a three-step method for detecting non-technical losses in the distribution network by using data mining. Atfirst, a data analysis process is performedon the data. Then, using robust principal component analysis (PCA) method, features are extracted to improve the recognition performance. In the third step, the outlierdata is provided by an improved k-means clustering method, which identifies subscribers suspected of nontechnical losses. For the evaluation ofthe proposedplan,the real data obtainedfrom the distribution company of Fars province has been used, which includes the data of ordinaiy andAMI meters.

non-technical losses, principal component analysis method robust (PCA) , improved K.-means clustering