A Novel Method for Power Quality Disturbance Detection Based on LSTM Network Model Residuals

Abstract

In order to accurately detect the time when the disturbance of power quality (PQ) occurs and the variation of voltage amplitude, a novel method of power quality disturbance detection that is based on the residual of the long short-term memory (LSTM) network model is proposed. Firstly, an offline model is developed based on historical voltage signals from a power grid under the normal condition by using the LSTM theory. Then, in order to detect PQ events online, the grid voltage signal is taken as the input of the model to obtain the residual component between the input signal and the normal signal. Finally, the amplitude change of the residual component reflects the amplitude change of the voltage. The starting and ending time and duration of the disturbance are located by detecting the mutation points of the residual component. Experiments on simulation data and engineering data show that the proposed method can well reflect the amplitude change when the voltage is disturbed, and the disturbance positioning accuracy is high.