Fault Location and Faulty Line Selection in Transmission Networks: Application of Improved Gated Recurrent Unit

Abstract

Wide-area measurement systems (WAMS) are efficient tools to estimate the fault location in the large-scale power systems. In this study, a fast and accurate deep learning based structure is proposed to detect the faulty lines and locate the fault based on the measured data of WAMS including phasor measurement units. To this end, a modulated gated recurrent unit deep learning based network named improved gated recurrent neural network has been proposed. The designed network, which has a recurrent neural network based structure, benefits from one update gate that can capture sudden changes in transient voltage and current signals and also enhance computational efficiency. Furthermore, to enhance robustness of the designed network under noisy conditions, a new informative-based loss function is formulated. The proposed loss function utilizes generalized form of mutual information to enhance robustness of designed network in the presence of Gaussian and non-Gaussian noises. The numerical results on IEEE 68-bus system verifies the effectiveness and superiority of the proposed method considering several operational conditions and comparison by two combined shallow-based networks and one deep learning based networks.