Feature selection of power system transient stability assessment based on random forest and recursive feature elimination

A feature selection algorithm based on random forest and recursive feature elimination is proposed in this paper, which can extract the key feature subset of transient stability assessment. Firstly, the weighted random forest is adopted to pay special attention to the unstable samples, which constitute only a very small minority of the power system operation samples. Secondly, random forest combined with the recursive feature elimination strategy is used to determine the most reasonable threshold to select important features. The threshold that distinguishes a feature whether important or not is determined by the change of OOB error rate. Finally, the classification performance of the reserved feature subset is tested by support vector machine. Case study conducted on the New England 39-bus power system exhibits the validity of the proposed algorithm.