Feature extraction based hellinger distance algorithm for non-intrusive aging load identification in residential buildings

Although steady-state power features such as real power (P), reactive power (Q), and total voltage/current harmonic distortions (VTHD/ITHD) may contain sufficient information, adopting them directly for non-intrusive aging load monitoring (NIALM) identification process requires longer computation time and larger memory. To effectively reduce the number of steady-state power features representing load aging signals without degrading performance, a Hellinger distance (HD) algorithm for extracting the power features of NIALM is proposed and presented in this paper. To minimize the training time and improve recognition accuracy, a Particle Swarm Optimization (PSO) is adopted in this paper to optimize parameters of training algorithm in Artificial Neural Networks (ANNs). The proposed methods are used to analyze and identify the load characteristics of aging loads in residential buildings. The recognition result shows that the accuracy can be improved by using the proposed method.