Power System Transient Stability Assessment Based On Characteristic Redundancy

Real time measurement data has the characteristics of high dimension, feature redundancy, and insufficient weight of boundary information, which is not conducive to transient stability assessment. Feature redundancy and fuzzy boundary information are not conducive to transient stable boundary fitting; Category imbalance leads to the model paying more attention to most class samples. In this paper, the improved random forest assessment feature importance is used to obtain the sample stability, and the samples with high stability are deleted to increase the boundary sample weight; Combining the feature independence determined by improved k-means++to obtain the optimal feature subset to reduce feature redundancy. The example shows that the processed samples have the characteristics of no redundancy of features and heavy weight of boundary samples, thus effectively improving the transient stability evaluation performance of the model.