Fault classification in power system distribution network integrated with distributed generators using CNN

•Proposed a deep learning based CNN model for fault classification in Power System distribution network integrated with distributed generators (DGs).•In this proposed work, fault classification is done without any pre-processing or feature engineering steps.•All types of fault in 3-phase power system network including no-fault is classified using proposed model with superior performance.•Proposed model is compared with conventional machine learning methods for fault classification in terms of both performance and computational cost.•The performance of proposed model is also tested on a mixed transmission line and distribution network with PV as DG. Fault detection is the critical stage of the relaying system and their successful completion in minimum time is expected for fault clearance. With the increasing usage of distributed generators (DGs) in a distribution network, the conventional relaying methods are becoming inappropriate due to changing fault current levels. This paper presents a deep learning algorithm i.e. Convolutional Neural Network (CNN) customized for fault classification in the distributed networks integrated with DGs. This is first time that CNN has been used for fault detection using raw and sampled-data of three-phase voltage and current signals of various fault classes and no-fault class. The 10-fold cross-validation is used to demonstrate the performance of the proposed model in terms of different metrics such as accuracy, sensitivity, specificity, precision, and F1 score. The proposed model has attended an average 10-fold cross-validation accuracy of 99.52% for all the tested fault cases. This featureless proposed method has been compared with conventional approaches from literature and has shown better performance in terms of accuracy and computation burden. Further, a similar fault study is conducted on a mixed transmission line and distribution network with PV as DG using the proposed method and found performance accuracy of 99.92% and 99.97%, respectively.