Anomaly Detection method of Smart Meters data based on GMM-LDA clustering feature Learning and PSO Support Vector Machine

Abstract

Aiming at the new risk of power data loss brought by the introduction of new technology into traditional power grid by Advanced metering infrastructure(AMI), an adaptive method for detecting abnormal data of smart meters based on Gaussian mixture model linear discriminant aualysis (GMM-LDA) clustering feature learning and particle swarm optimization support Vector Machine (PSO-SVM) is proposed. In order to detect abnormal electricity consumption behavior in AMI. Firstly, GMM-LDA clustering feature learning algorithm is used to cluster some data sets to obtain the optimal feature representations of normal pattern feature database and abnormal pattern database. Then, an adaptive updating mechanism of pattern database is introduced to make the abnormal detection model adapt to the dynamic changes of network environment, and the labeled data set is compiled. The training samples are sent to the PSO-SVM classifier for learning. Finally, the unknown label data sets are classified, and the abnormal electricity data can be traced back to the suspected users, so that realize the key monitoring of the suspected users. Experiments on existing data sets show that the proposed detection model not only avoids the dependence on manual classification in supervised training samples, but also has higher detection accuracy and lower false alarm rate than single or other algorithms.