Modified Cross-Entropy Method for Classification of Events in NILM Systems

Non-intrusive load monitoring is an algorithm or process that disaggregates the total power in a facility to identify consumption of individual appliances. In this paper, a new algorithm is proposed to classify events of appliance states based on modification of the cross-entropy (CE) method. The main contribution is a formulation and solution of the problem with the CE method as a constrained optimization problem. This new technique is called the modified CE method. The proposed algorithm is simple, as it operates in real time, using low rate sampling of the active power. In addition, there is no need to train the system or to use complex hardware. The algorithm is tested using the REDD and AMPds datasets and the results are compared to several state-of-the-art techniques.