Load identification based on Factorial Hidden Markov Model and online performance analysis

Abstract

Load identification is important for the tasks such as load forecasting, demand response and energy management in smart buildings. The accuracy of the traditional methods depends on the dimension of load signatures, the sampling frequency and the stability of load profile. In this paper, a Factorial Hidden Markov Model (FHMM)-based method is proposed to analyze the aggregate load profile and identify the individual device. We extend the Viterbi algorithm to solve the FHMM directly, and this process is more efficient than the solution of the equivalent HMM by using the conventional Viterbi algorithm. The proposed method is insensitive to the stability and accuracy of power data, so it is suitable for the devices in buildings, even for the continuously variable loads. Two experiments with real power data are evaluated to illustrate the proposed method. Meanwhile, we focus on the online performance of the Viterbi algorithm. It is found that the states decoded by Viterbi are unreliable when the observed data are inside a confusing zone. Through analyzing the mechanism of the Viterbi algorithm, the judgment conditions the boundary of the confusing zone are given. We hope this work brings insight to the research on load identification and HMM.