Data Mining and SVM Based Fault Diagnostic Analysis in Modern Power System Using Time and Frequency Series Parameters Calculated From Full-Cycle Moving Window

This paper proposes a complete diagnostic analysis of faults in a typical modern power system's transmission line using the support vector machine (SVM) with time-series parameters and frequency series parameters as features. The training and testing data of the proposed method are collected by simulating all types of faults with all possible variations on a transmission line (TL) in the IEEE-9 bus system using the PSCAD/EMTDC software. While simulating one type of fault, fault resistances and fault inception angles are also varied to account for the various behaviours of the fault. The three-phase instantaneous currents and voltages on both sides of TL are recorded at 32 samples per cycle. A thirty-two sample moving window is used to compute time-series and frequency-series parameters applied as features to the SVM. Ten-fold cross-validation is used to evaluate the performance of the proposed algorithm with evaluation metrics such as accuracy, precision, recall and F1 score. Features generation, training and testing of the proposed method, and performance comparison are done using PYTHON software. The proposed method has achieved an average accuracy of 99.996%, even in the most contaminated environment of 30 dB noise. Compared with the performance of the other popular machine learning algorithms, the proposed method has achieved more accuracy. The performance of the proposed method is also tested with different noise levels, which account for the measurement errors of 30 dB, 35 dB and 40 dB.