Abstract: In order to accurately predict the degree of large extrusion deformation of deep roadway in different stages, taking the deep roadway in Jinchuan No.2 mining area as the research object, the large deformation was divided into four grades. According to the S-shaped deformation law of large extrusion deformation, the monitoring data were regressed and predicted by Logistic model, and the overall development trend and final deformation of large deformation were obtained, which provided a reference for the grade prediction of roadway production and maintenance stage. At the same time, comprehensively considering the factors such as stress state, geological environment, construction conditions and surrounding rock properties, the large deformation prediction model based on AHP-SPA was established by using the weight of analytic hierarchy process. Based on the maximum membership principle and confidence criterion, the grade identification was carried out, and the potential large deformation grade in the stage of roadway survey and design was predicted in advance. The results show that the prediction grade of the prediction model for different stages is consistent with the actual situation, which can provide a basis for the design and maintenance of roadway support in deep mines.