In the process of making full use of the massive historical operation data and simulation analysis data of large power grid to evaluate the transient stability situation, the proper selection of features closely related to stability and the extraction of key factors affecting the abnormal stability situation are the basis. In this paper, an intelligent stability situation assessment based on time-series feature selection is proposed. The online generation method of neighborhood samples based on the future operation section and the stability situation levels are given, and then, the transmission section is determined as the main features. With the support of the neighborhood information measurement algorithm of the power grid time-series data, the features are arranged in descending order according to the cumulative contribution rate, and then, the redundant features are eliminated by the calculation based on the temporal neighborhood mutual information and the feature subset search based on SVM to form a stable feature subset. In the scenario of applying the feature subset, the extraction of stability-related information (SRI) based on MSPCA algorithm is adapted, and the abnormal situation identification of key features is realized through the ranking of SRI contribution rate. The simulation results of IEEE 39-bus verify the effectiveness of the proposed method.