In view of the shortcoming of k -nearest neighbor（kNN） classification for neglecting sample structure information， a pattern discovery algorithm is adopted using the proposed extended k -nearest neighbor（EkNN） algorithm to obtain the spatial distribution knowledge from sample space which is then utilized to classify the unknown sample instead of the training set. The technique has eliminated the interfering samples unfavorable to classification while improving its accuracy and speed. The stability assessment scheme employs EkNN to identify the stability levels of input operation states based on pre-contingency steady state parameters. The simulation results of two IEEE test systems show the effectiveness of the proposed method. The EkNN algorithm as a knowledge acquisition tool can be applied to a wide range of engineering domains.