Machine learning approach for non-intrusive load monitoring in smart grids: new deep learning method based on long short-term memory and convolutional neural networks

Abstract

Non-intrusive load monitoring (NILM), problem is a method that translates consumer’s electricity consumption data into the electricity consumption for each appliance in smart grids. New approach for NILM has been proposed in this article that uses combination of long short-term memory (LSTM) networks and convolutional neural networks (CNN). Due to the complexity of NILM problem, application of deep neural networks could be advantageous because of its performance and flexibility. In this article proposed method significantly enhances the efficiency of NILM due to implementation of both deep neural networks. It applies sequence to sequence learning, which predefined window of the consumption is fed as an input and specified appliance consumption data is considered as output. Real-world household energy data set REFIT" has been used for training and testing the proposed method. In this study the RFIT data set has been used, and electricity consumption data of 20 households with nine appliances measured at 8-second intervals. The electricity usage data have been recorded regularly over a two-year period for 20 British households. The proposed method managed to reach accuracy, Fl-score and estimated energy measures by 95.93%, 80.93% and 93.67%, respectively that validates accuracy and performance. Comparison of the proposed method’s results and recently published studies has been presented and discussed based on accuracy, number of considered appliances and the size of the deep neural network’s trainable parameters. The proposed method shows remarkable performance compared to past studies.