Recognition of pollution layer location in 11 kV polymer insulators used in smart power grid using dual-input VGG Convolutional Neural Network

Abstract

This paper portrays the application of a Partial Discharge (PD) signal combined with the dual-input VGG Convolution Neural Network (CNN) to predict the location of the pollution layer on 11 kV polymer insulators subjected to alternating current for smart grid applications. First, a non-uniform pollution layer artificially created for HV insulator with three straight shed ball end fitting in a laboratory setup and corresponding PD readings are measured. The wavelet transform is employed to represent the measured PD signal as scalogram patterns. In general CNN uses a single input pattern for feature extraction. If the pattern quality is low, it is easy to cause misclassification. Hence in this proposed work, the feature fusion of a dual-input Visual Geometry Group (VGG) based CNN is used for the classification of contamination layer. VGG 19 is a pretrained deep learning network used for extracting the rich features from the patterns. In continuation to that, hyperparameter (HP) play a vital role in deep learning algorithms because they directly manage the behaviours of training algorithms and have a significant effect on the performance of deep learning models. Hence, Bayesian Optimization (BO) is used for tuning the HP. At last, to check the practicality of the proposed algorithm, a new dataset is created for 11 kV polymer insulator with three alternate shed clevis end fitting and different pollution levels—acceptable results obtained by using dual-input CNN with the minimum quantity of data.

Keywords

• Dual-input CNN
• Convolution Neural Network
• Bayesian optimization
• Feature fusion
• Training optimizer
• High voltage insulator