Hybrid CNN-LSTM approaches for identification of type and locations of transmission line faults

Timely and accurate detection of transmission line faults is one of the most important issues in the reliability of the power systems. In this paper, in order to assess the effects of impedance and location of the fault in iden-tifying and classifying it, the frequency response analysis (FRA) method is utilized. This method clearly shows the smallest effects of the faults on voltage and current signals in the frequency domain. Interpretation of the results associated with the FRA procedure is considered a weakness of this method. To overcome this issue and accu-rately categorize types and locations of various transmission lines faults such as asymmetric faults and symmetric faults, machine learning, and deep learning applications called support vector machine (SVM), decision tree (DT), k-Nearest Neighbors (k-NN), convolutional neural network (CNN), long short term memory (LSTM), and a hybrid model of convolutional LSTM (C-LSTM) are utilized. Introduced faults are applied with various imped-ances in 6 segments of an IEEE standard transmission line system. Then, the frequency response curves (FRCs) for them are computed and selected as input datasets for the suggested networks. After categorizing the types and locations of faults, the results for each network are analyzed via different statistical performance evaluation metrics. Finally, in order to early detection of faults, the new high impedance faults (7000 and 9000 O) are applied based on the previous routine in the transmission line. At this stage, evaluations demonstrate the capability of the C-LSTM followed by SVM, DT, k-NN, CNN, and LSTM in categorizing the type and location of transmission line faults.

Automated summary

The timely and accurate detection of transmission line faults is crucial in power system reliability. This paper utilizes the frequency response analysis (FRA) method to assess the effects of impedance and fault location, and introduces machine learning and deep learning applications to accurately classify types and locations of transmission line faults. The results show that faults have minimal effects on voltage and current signals in the frequency domain, and the proposed networks demonstrate strong capability in categorizing fault types and locations.