A new approach for supervised power disaggregation by using a deep recurrent LSTM network

This paper presents a new approach for supervised power disaggregation by using a deep recurrent long short term memory network. It is useful to extract the power signal of one dominant appliance or any subcircuit from the aggregate power signal. To train the network, a measurement of the power signal of the target appliance in addition to the total power signal during the same time period is required. The method is supervised, but less restrictive in practice since submetering of an important appliance or a subcircuit for a short time is feasible. The main advantages of this approach are: a) It is also applicable to variable load and not restricted to on-off and multi-state appliances. b) It does not require hand-engineered event detection and feature extraction. c) By using multiple networks, it is possible to disaggregate multiple appliances or subcircuits at the same time. d) It also works with a low cost power meter as shown in the experiments with the Reference Energy Disaggregation (REDD) dataset (1/3Hz sampling frequency, only real power).