장지연(Jiyeon Jang) ; 김인수(Insu Kim)
In power systems, faults, such as ground faults and short circuits, and non-fault disturbances, such as large load fluctuations and unbalances, occur frequently. However, the power system responses to power faults and non-fault disturbances are different.
Therefore, it is essential to accurately distinguish between faults and non-fault disturbances in power systems. Previous studies have collected large-scale data by monitoring real-time parameters of the power system and detecting the occurrence of power system faults. However, this study does not focus only on diagnosing power system faults but also on accurately distinguishing between faults and non-faults disturbances and uses various classification models to train the data and evaluate and analyze the prediction results. Collecting power system parameters when faults and non-faults disturbances occur is not easy. Therefore, this study used DIgSILENT PowerFactory software to simulate faults and non-faults disturbances in the power system and collected 120 small data sets. The data collected for the 120 cases consists of various metrics such as voltage, current, frequency, rotor speed, and HVDC parameters. This study used seven classification models for training and prediction: decision tree, gradient boosting classifier, k-nearest neighbors, logistic regression, naive Bayes classification, and random forest regression. In addition, this study introduced an importance-based data reorganization method to improve the performance of the best-performing classification model and analyzed its effectiveness.