Appliance Usage Status Classification Using Long Short-Term Memory Technique

Abstract

This paper explores the challenge of non-intrusive load monitoring (NILM), which involves disaggregating the total electricity consumption of a household into individual usage without the need for intrusive monitoring equipment. The main objective of this research is to classify appliance status using a long short-term memory (LSTM) technique to improve the accuracy of NILM. This work encompasses the development of LSTM, the collection and pre-processing of raw data, and the evaluation of the predictive performance. Extensive data pre-processing was conducted, including handling missing values, normalizing, and segmenting data into sequences suitable for LSTM training and testing. LSTM was trained using pre-processed data and achieved impressive prediction accuracy, with a consistency error rate below 5%, translating into a prediction accuracy more than 95%. The performance was assessed using the root mean square error (RMSE) and showed a nearly to zero value, indicating a high level of accuracy. This performance was compared to other techniques such as convolutional neural networks (CNN) and recurrent neural networks (RNN), where LSTM demonstrated superior accuracy and reliability. The key findings underscore the effectiveness of the LSTM in capturing long-term dependencies within the data, resulting in higher prediction accuracy for individual appliance usage. Implementing threshold logic further enhanced the predictions by reducing noise and increasing accuracy. Despite the high performance, the study acknowledges limitations related to data quality, computational complexity, and generalizability.