Nontechnical Losses detection: A Discrete Cosine Transform and Optimum-Path Forest based approach

This paper presents a Discrete Cosine Transform (DCT) and Optimum-Path Forest (OPF) based approach for Nontechnical Losses (NTL) detection. NTL decrease the economic efficiency of distribution utilities, which harms the entire society since energy prices increase as a consequence. Pattern Recognition based approaches have been applied to identify these losses so that the problems related to NTL may be corrected through on-site inspections. One of the main characteristics of NTL is the decrease in the customer's measured monthly-consumed energy, caused by frauds or failure of the energy meter, which is called “consumption step”. This record of kWh consumption can be treated as a time-series and then benefit from the knowledge available in this area. In this paper DCT and OPF were used as a means of implementing automatic feature extraction. The pro-posed method was compared with an OPF based approach previously proposed by the authors. Test results show that feature extraction using DCT can provide a more compact and efficient way of representing data, improving the performance of previously developed methods by the authors.