Neural NILM

Abstract

Energy disaggregation estimates appliance-by-appliance electricity consumption from a single meter that measures the whole home's electricity demand. Recently, deep neural networks have driven remarkable improvements in classification performance in neighbouring machine learning fields such as image classification and automatic speech recognition. In this paper, we adapt three deep neural network architectures to energy disaggregation: 1) a form of recurrent neural network called 'long short-term memory' (LSTM); 2) denoising autoencoders; and 3) a network which regresses the start time, end time and average power demand of each appliance activation. We use seven metrics to test the performance of these algorithms on real aggregate power data from five appliances. Tests are performed against a house not seen during training and against houses seen during training. We find that all three neural nets achieve better F1 scores (averaged over all five appliances) than either combinatorial optimisation or factorial hidden Markov models and that our neural net algorithms generalise well to an unseen house.

An Empirical Study on Energy Disaggregation via Deep Learning

“…Similar to [1], we refer to the power over a complete cycle of an appliance as an appliance activation. For a short-duration appliance, for example a kettle, an activation usually lasts for several minutes, while for long-duration device such as washing machine, an activation may be as long as several hours.…”

“…We also used the five appliances as in [1], the kettle, dish washer, fridge, microwave oven and washing machine to perform experiments. These devices consume the majority of energy and each of them exists in at least three houses in the dataset.…”

“…Disaggregated electricity consumptions can be used to produce itemized electricity bills to help customers to identify and improve their consumption behavior. Moreover, energy disaggregation can help operators to better manage power grid, and detect faulty or improperly used devices [1].…”

“…So researchers now start to adapt deep neural networks to energy disaggregation task. Kelly and Knottenbelt [1] applied three types of deep neural network structures, a recurrent neural network using Long Short Term Memory units (LSTM) [12], a denoising auto-encoder, and a regression model that predicts the start time, end time and average power demand of each appliance. However, they did not achieve state-of-the-art disaggregation performance with deep learning approaches.…”

“…It can also be considered as a non-symmetric auto-encoder. The second type of network is a multi-layer RNN (recurrent neural network) using LSTM units, which is similar to the structure used in [1]. For both structures, we use multiple parallel convolutions with different filter sizes to transform the raw power signals.…”