Power Signal Disturbances Analysis based on Transfer Learning in Deep Architecture

Abstract

A deep learning model based on transfer learning for power quality (PQ) disturbances detection and classification has been proposed in this article. The power disturbances are non-stationary and random so deep learning is a best choice to deal with such signals as it can estimate various random parameters of power quality disturbances such as phase, frequency and amplitude. Real time power quality disturbances are created in laboratory and accuracy of proposed model in compared with seven CNN-based DL architecture models (viz., VGG19, ResNet101, ResNet152V2, Xception, InceptionV3, MobileNet, and MobileNetv2) under serve noise conditions. Low computational overhead and high accuracy in severe noise conditions make the proposed model suitable for real time applications. The results indicated that the proposed model could realize and achieve more than 99% of accuracy. Furthermore, the proposed novel architecture could be readily used for early detection and classification of power disturbances and can be further extended to various smart cities based IoT models to realize an accurate, timely, and faster detection & classification of power disturbances in smart grid.