Feature selection for the non-intrusive electrical appliances identification

Piotr Bilski, Wiesław Winiecki

Abstract

The paper presents the application of selected feature selection methods to determine their optimal set in the task of non-intrusive appliances identification. The process detects devices operating in the specific location based on the analysis of voltage and current signals. The informativeness calculation methods and Fisher linear discriminative analysis are proposed to find significant features for determining, which appliances changed their state. The approach is verified on two sets of features. The scheme was tested in the laboratory conditions with multiple appliances working simultaneously, exploiting the decision tree as the appliance identifier. Experimental results show usefulness of the approach maximizing the identification accuracy with decreased complexity of computations.

Keywords

NIALM, artificial intelligence, feature selection

Download

IMEKO-TC4-2016-37.pdf

DOI

-

IMEKO TC

TC4 - Measurement of Electrical Quantities

Back to the proceedings