Vector Projection Classification for home appliances recognition: A load signature comparative analysis

Non-intrusive methods for recognize appliances are widely studied at the state of art since 90's, but the use of human faces pattern recognition techniques to solve this kind of problem is a novel research field. Vector Projection Classification (VPC) is a brand new method in this area and its approach is settled upon finding the similarity between images by proximity among vectors projected onto a vector space. This paper presents a depth comparison between two load signatures types (current and power) using VPC. Besides, domain analysis is performed (time and time-frequency applying Stockwell transform), observing the identification rate for the optimal parameters values of VPC. Sixteen residential loads were sampled for this work (thirty samples each). Results showed that for time domain VPC originates unsatisfactory identification rates (less than 20% for current and less than 60% for power). Best results are achieved for time-frequency domain, with 97% of identification rate for the optimal parameters values and current signature.