Reasoner design based on HYPO for classification of lighting loads

Abstract

Nonintrusive Load Monitoring (NILM) provides information about the electrical power consumption per appliance in a house to manage the energy consumption. NILM requires measurements in only one point and algorithms to make load disaggregation. One approach is classifying characteristics of the appliance through machine learning techniques such as Support Vector Machines (SVM) and Artificial Neural Networks (ANN). These techniques have limitations in the database use and the disregard of the information context. In this paper a reasoning technique based on the Case Based Reasoning (CBR) reasoner called HYPO is proposed. This reasoner creates hypothetical cases to classify new cases based on the solution of previous experiences. The study is focused on lighting appliances which represent meaningful power consumption in the houses. Electrical measurements lamps in steady state were acquired in the Laboratory, for individual and combined operation. Additionally, characteristics are computed to build the CBR HYPO models. The performance of CBR HYPO is evaluated and compared to the one of SVM. As a result, CBR HYPO outperforms the SVM for combined operation of lamps, while it fails behind SVM for individual operation.