Power Quality Disturbance Classification via Deep Convolutional Auto-Encoders and Stacked LSTM Recurrent Neural Networks

Power quality disturbances are one of the main problems in electric power systems. The perturbations or disturbances occur due to many factors such as non-linear loads, lighting controls, power electronic equipment, system faults and switching events. In this sense, if monitoring and control actions are not developed to mitigate properly these disturbances, an overall interruption of the power transmission and distribution networks could be generated, causing an important social impact and huge economic losses. To date, several strategies have been proposed in the literature, including signal processing based feature extraction approaches, machine learning based classifiers and heuristic optimization techniques to identify and detect power quality disturbances. However, limitations in terms of computational load, efficiency and performance are still important issues to solve. In this paper, a novel technique of automatic classification of power quality disturbances is presented. The proposed ensemble algorithm consists of a Convolutional Auto-Encoder compression architecture, which is included to reduce the signal size, and a stacked Long Short Term Memory (LSTM) Recurrent Neural Network (RNN), to conduct the classification process keeping high accuracy metrics. The performance of the proposed algorithm is evaluated through the process and analysis of a synthetic database created with the characteristic mathematical models formulated for nine disturbances. The results show a significant reduction of time processing and a higher accuracy rate compared with traditional LSTM RNNs.