In order to improve the accuracy of power system transient stability assessment, this paper proposes a feature selection method based on the combination of maximum correlation minimum redundancy (mRMR) and random forest out-of-bag error (RFOE). method (mRMR-RFOE), and use the random forest algorithm to build a power system steady-state evaluation model, and use mRMR and RFOE to rank the features respectively. The final ranking result is obtained by combining the two kinds of feature sorting, and the optimal feature subset required by the random forest algorithm is selected according to the optimal number of features obtained from the experiment. Finally, the random forest classification model is trained using the optimal feature subset. The experimental results show that compared with other classification methods, this method can improve the accuracy of power system steady-state assessment, which is of practical significance.