States and Power Consumption Estimation for NILM

This paper presents Nonintrusive Load Monitoring (NILM) for electrical home appliances network which consists of a known set of devices. Hidden Markov Model (HMM) is used for system modeling. The proposed method enhances determining and defining all states for each device. First we classify each device states into a set of states not only the ON and OFF states in the form of variations in its active power ranges. AMPDS collected dataset is used in training and testing for six selected home devices in a certain household and is also compared to GREEND dataset showing the advantage of the variable observed power readings with those of constant power readings. Each device has different number of states. Then the proposed mechanism is used to minimize these states after understanding the behavior of each state into OFF and ON states only. This method provides high accuracy on the system level, the device level, state inference, power and state sequence estimation.