Non-Intrusive Load Disaggregation Considering User Behavior and Appliances States Combination

Non-Intrusive Load Monitoring (NILM) is one of the fundamental problems in intelligent power consuming networks. With the aid of NILM, the power consumptions of all the household appliances become available for users, who are able to save energy via optimization and automatic control methods. In this paper, we propose a load disaggregation algorithm based on time probability distributions of appliances. First, we modify the iterative K-means clustering algorithm to extract appliance states. Then, each state time probability distribution is extracted using the historical data. Super states are constructed based on appliance state combinations, and further reduced via learning from historical data. Super states clustering is performed considering power overlap. While performing load disaggregation, optimal solution is searched by maximizing the time probability. The effectiveness and performance of the proposed method is assessed via data set test.