The invention discloses a super computer operation failure active prediction method based on application similarity, and belongs to the field of super computers, and the method comprises the steps: S1, extracting feature data from an operation log, adding operation path data, preprocessing the feature data and the operation path data, and taking the preprocessed data as input features of a machine learning algorithm model; and S2, after the machine learning algorithm model processes the input feature data, the operation failure state is actively predicted. According to the method, the characteristics capable of accurately describing the job application attributes are mined, and a good prediction improvement effect is achieved; a machine learning algorithm is adopted to find an operation failure prediction method, the robustness of a prediction model is improved, and the method is especially suitable for nonlinear data; the clustering calculation overhead is remarkably reduced and the error is reduced by adopting a clustering method for the job application attributes; the prediction efficiency is high, and the method can be practically applied to large supercomputers.