NTL DETECTION: A INTELLIGENT SYSTEM USING A NON-SUPERVISED CLUSTERING OF CONSUMPTION PROFILES

Abstract

One of the main concerns of power distribution companies is energy loss. A non-technical loss (NTL) is defined as any consumed energy or service that is not billed by some type of anomaly. This paper presents an intelligent system for the detection of NTL of electrical energy in power utilities. The use of machine learning classifiers has been an attractive option for NTL detection. It enhances data-oriented analysis and high hit ratio along with less cost and manpower requirements. This proposal has three stages: a non-supervised clustering of consumption profiles based on a hybrid algorithm using Self-Organizing Maps (SOM) and Discrete Cosine Transform (DCT). A second stage compute several features criteria based on locality, infrastructure, and consumption profile (feature engineering). The final stage is supervised classifiers to detect NTL in the costumers. The proposed intelligent approach was trained and tested with real data from Ceará-Brazil (149000 customers), where the utility reports energy losses of around 13% of the total energy purchased by the company during the three last years. The results show an average overall performance of 85% and AUC = 0.758 in the detection process of NTL. The results increasing the effectiveness compared to other approaches previously applied by the utility in the region.