A Fourfold Bi-Filter and Permuted Bi-Quad-Wrapper Feature Selection Method for Finding Optimal Moments of Multi-Trajectory Transient Records in Transient Analysis

Fast and accurate transient analysis (TA) requires a meticulously explored nonlinear transient excursion space (NTES). Hence, designing a polyhedral feature selection scheme (FSS) to fix the TA bottleneck raised by feeding non-discriminative transient records into predictive models is a significant task to avoid discordant TA indices results. Hence, we offer a fourfold bi-filter and permuted bi-quad-wrapper (F2FP2QW) FSS to find optimal transient moments (OTMs) from multi-trajectory transient records (MTTRs). The F2FP2QW model exerts varied fourfold repetition of the linked four-stage, including the bi-filter (relevancy rate (RR) and conditional-based RR (CRR) calculations) and bi-wrapper (incremental wrapper subset selection (IWSS) and IWSS with replacement (IWSSr)). The RR and CRR mounted on mutual information (MI) and conditional MI (CMI) in the filter stages. In the case of wrapper methods, hyperplane-based classifiers (HCs), namely support vector machine (SVM) and twin SVM (TWSVM), play the main role in IWSS and IWSSr tree formation. Due to kernel-based learning necessity in NTES, embedding the radial basis function (RBF), dynamic time warping (DTW), and polynomial (POL) kernels into dual HCs is on the agenda. Generally, wrapper stages include ${ }_{TWSVM}^{SVM} \textrm {IWSS}_{RBF/\,POL}^{RBF/\,DTW\,} $ and ${ }_{TWSVM}^{SVM} \textrm {IWSSr}_{RBF/\,POL}^{RBF/\,DTW\,}$ in a 24-way permutation. By applying the multifaced filter-wrapper scenario of F2FP2QW on 28-trajectory transient data, the OTMs are obtained. Finally, the efficacy of OTMs survived by the F2FP2QW model in TA is evaluated via a cross-validation scenario. The results show that F2FP2QW has a prediction accuracy of 98.87 % and a processing time of 102.587 milliseconds for transient stability prediction.