NON-INTRUSIVE ABNORMAL USER BEHAVIOR DETECTION BASED ON DOUBLE-LAYER FUSION OF SEQ2SEQ AND SVM

The paper studies the abnormal power consumption behavior of customers on the basis of non-intrusive load disaggregation. K-means clustering algorithm was used to extract the state features of the loads, and the sequence-to-sequence model of deep learning algorithm was used to disaggregate the total power consumption data of power users into power consumption data of a single equipment. Combining with SVM algorithm, the abnormal data of many kinds of household electrical appliances were analyzed. Experiments on UKDALE dataset show that the proposed model can not only improve the disaggregation accuracy and reduce the disaggregation error, but also can monitor the abnormal behavior of users.