A Comparative Study on the Performance of Hidden Markov Model in Appliance Modeling

Abstract

Load modeling using data-driven algorithms is a widely used technique in applications like load identification. It is also one of the fundamental concepts which enable Non-Intrusive Appliance Load Modeling (NIALM). This paper develops a load modeling framework using Hidden Markov Models (HMMs) to identify a two-state home appliance. Unlike previous studies, the training and testing dataset is derived from different monitored domestic houses to analyze the effect of the training data trends on the model’s accuracy. We used the Reference Energy Disaggregation Dataset (REDD) in the load modeling process. The developed system utilizes adaptive measures to construct HMM models that can identify foreign variants of the same two-state appliance. We measured the accuracy of our proposed methodology by comparing a known state sequence with a Viterbi-generated one. The accuracy results are up to 96%, depending on the nature of the used training dataset.