Transmission Line Fault Classification Based on Dynamic State Estimation and Support Vector Machine

Abstract

The transmission line is an essential part of the power system, while the faults in the transmission line are inevitable due to a number of reasons, such as lightning, tree limps, wind, etc. This paper presents a technique to classify the short circuit faults in transmission lines, which distinguishes fault phases and fault types, such as phase A to ground fault and phase A to phase B fault. The proposed method combines the dynamic state estimation (DSE) based protection technique and Support Vector Machine (SVM) to achieve an accurate classification model with physical foundation. DSE utilizes measurement data from merging units and the model of transmission line to obtain estimated states of the line. The differences of measurements and estimated values are referred as residuals. Different fault types possess different patterns of residuals and include features that can be used in a classification scheme. Therefore, these residuals are applied to SVM to set up classification models. A feature selection method based on Lasso regression is deployed to increase the training speed for SVM. The performance of the technique is evaluated using simulation test cases. Different fault types with different fault impedance and fault locations are included in the test cases. The test results show that the classifiers are able to accurately categorize the fault types, which is very useful for protection functions.