|
Abstract
|
Collecting the optimal transient features (OTFs) from multivariate transient time series data is crucial for ensuring precise and prompt transient analysis (TA) in power systems. Realizing the OTFs requires designing a comprehensive hybrid feature selection scheme (FSS). Hence, this work offers a four-group permutational hybrid FSS called FGPHFSS to select OTFs from transient trajectory data. The FGPHFSS includes four hybrid FSS (4HFSS) groups decorated by filter-wrapper mechanisms. The two 4HFSS groups comprise a relevance-based filter and another two 4HFSS groups derived from a conditional relevance-based filter. The 4HFSS groups includes incremental wrapper methods namely incremental wrapper subset selection (IWSS) and IWSS with replacement (IWSSr), which are fed by the support vector machine (SVM) and twin SVM (TWSVM) classifiers. The nonlinear space of transient data causes the SVM and TWSVM to be equipped with elastic and non-elastic kernels. Generally, we have IWSS Relevance SVM RBF DTW TWSVM RBF POL, IWSSr Relevance SVM RBF DTW TWSVM RBF POL, IWSS Cond Relevance SVM RBF DTW TWSVM RBF POL, and IWSSr Cond relevance SVM RBF DTW TWSVM RBF POL, which are permuted in a 24-way manner. After selecting OTFs, the OTFs' performance in transient stability prediction (TSP) is evaluated by a cross-validation technique. The results show that FGPHFSS has a prediction accuracy of 99 % and a processing time of 101.819 milliseconds for TSP.
|