A novel three-dimensional deep learning algorithm for classification of power system faults. (May 2021)

• Record Type: Journal Article
• Authors: Srikanth, Pullabhatla; Koley, Chiranjib
• Is Part Of: Computers & electrical engineering. Volume 91(2021)
• Publication Date: 2021-05
• Subjects: 3D CNN -- Deep learning -- Power system faults -- Classification -- Signal processing -- Artificial intelligence; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2021.107100
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: First time a 3D CNN is proposed for fault identification in power system (FIPSNET). Validation accuracy of 93.75% & 100% for a dropout of 0.4 and 0.5 respectively has been achieved. Proposed 3D FIPSNET has taken only 210 s to complete training & validation. The 3D FIPSNET is lighter, faster and simple in comparison with AlexNet. Proposed approach has achieved 100% accuracy with improved speed and reduced memory. Abstract: A three-dimensional (3D) deep learning algorithm (DLA) has been proposed to classify Power System Faults. The proposed network is novel and requires fewer data to identify the power system faults with very high accuracy. The proposed network overcomes the issue of overfitting due to fewer layers and dropout provision. Also, it can be designed with basic knowledge of deep learning. The input to the DLA is a 4D image split into RGB channels. The 4D image is obtained after transforming the fault currents recorded in an IEEE-9 Bus system to a time-frequency domain. The individual color channel matrix is stored as a 4D matrix and then fed to the DLA to identify the power system fault type. The proposed 3D CNN is trained multiple times by changing the number of epochs and learning rates. The proposed model can classify the type of power system faults with an accuracy of 93.75 and 100% for a dropout value of 0.4 & 0.5. The training-validation sample size is considered as 1600 4D-images for both the dropout values. Graphical abstract: Image, graphical abstract