Abstract: The strength reduction method is widely used for slope stability analysis, and different instability criteria have their own characteristics. Among them, a simple and practical one is the mutation in the displacement reduction coefficient curve. This study uses cluster analysis to select the displacement mutation feature point, and identifies its mutation using the self-organizing neural network change point analysis method. Then, we suggest an improved criterion for this mutation. Such a fusion method－that clusters such an analysis－is more objective and mathematically based. For typical examples, safety factors calculated by this analysis method are closer to the judging values than those of the strength reduction method built in FLAC3D.

Key words: slope, clustering analysis, self-organizing maps, change point analysis, strength reduction method