Electricity Theft Detection to Reduce Non-Technical Loss using Support Vector Machine in Smart Grid

Among the various reason behind Non-technical losses in smart grid, losses due to electricity theft have become major apprehension in power system industries. A significant amount of consumption of electricity in a fraudulent manner decrease the supply quality, increase generation load and the real consumers have to pay excessive electricity bill that finally affects in overall economic condition. The advance metering infrastructure (AMI) has the ability to monitor the consumption detail of every consumer, record the consumption pattern, billing them as well as find any types of abnormalities. The communication capabilities of smart grid have facilitated the utilities to store their consumers’ consumption patterns. With this database it is possible to formulate a theft detection model by machine learning algorithm by analyzing the recorded data of smart meter. In this paper, Support vector machine (SVM), one of the prominent machine learning classifiers applied with principle component analysis to train the data collected from smart meter and calculate the prediction accuracy with the test data. Here the principle component analysis reduces the dimension of the data to make the turtuer processing less complex. After that, by using grid-search method the best-suited meta-parameters for SVM has been selected where the almost 90% accuracy rate is achieved. The classification results with respect to different metaparameters of SVM (regularization parameter C and γ) have presented. Moreover, the obtained result specifies that the applied techniques possesses higher accuracy and less false positive rate for real time consequences.