Abnormalities and Fraud Electric Meter Detection using Hybrid Support Vector Machine & Genetic Algorithm

K.S. Yap, I.Z. Abidin, A.R. Ahmad, Z.F. Hussien, H.L. Pok, F.I. Ismail, and A.M. Mohamad (Malaysia)


Support Vector Machine, Genetic Algorithm, Dynamic Crossover Point, PrePopulated Database, Dual Lagrangian Optimization


This paper presents an intelligent system to reduce Non Technical Loss (NTL) using hybrid Support Vector Machine (SVM) and Genetic Algorithm (GA). The main motivation for this research is to assist Sabah Electricity Sdn. Bhd. (SESB) to reduce their distribution loss, estimated around 15% at present in Sabah State, Malaysia. The hybrid algorithm is able to preselect customers to be inspected on-site for abnormalities or potential fraud according to their consumption patterns. SVM is a classification technique developed by Vapnik [1] but a practical difficulty of using SVM is the selection of parameters such as C and kernel parameter, σ in Gaussian RBF kernel. The purpose of choosing parameters is to get the best generalization performance. Genetic Algorithm (GA) is used to search for the best parameter of SVM classification by using combination of random and pre-populated genomes from Pre-Populated Database (PPD). It provides an increased convergence and globally optimized solutions. The algorithm has been tested using actual customer consumption data from SESB. 10 fold cross validation method is used to confirm the consistency of the detection accuracy. The paper also highlights comparison results between typical SVM and SVM-GA. The highest fraud detection accuracy for SVM GA is 94%.

Important Links:

Go Back