Transient state features selection method in the non-intrusive load monitoring

Abstract

The article presents selected feature selection methods to determine their optimal set in the task of transients identification in non-intrusive appliance load monitoring system. The architecture of such a system is proposed. In the paper, the method of appliance pattern calculation is introduced. Fisher linear discriminative analysis and correlation analysis are applied to find the most significant features. The approach is verified on real transient state data recorded in laboratory. 10-fold cross validation method exploiting discriminant analysis and decision tree identifiers was employed to verify effectiveness of adopted methods. Experimental results show usefulness of proposed approach allowing to reduce the set of employed feature without decrease of identification accuracy.