In this paper, in the process of building an online-offline integrated independent learning model, we propose to better adapt suitable learning resources for learners using knowledge clustering. Based on the characteristics of CURE clustering algorithm, which has strong generalization ability but high complexity, the optimized ACURE algorithm is designed by making improvements in two stages: preprocessing and class merging. Secondly, to test the performance of the optimized ACURE algorithm, decentralized representative points are used to represent the clustering of the ACURE algorithm. Next, a questionnaire survey was conducted among college students and teachers using random and stratified sampling to analyze their perceptions of this merged autonomous teaching model. Finally, the combined autonomous teaching model was applied to actual teaching and its effectiveness was analyzed through teaching control. The results show that the ACURE algorithm presents perfect clustering from the contraction factor 0.3. The overall percentage of teachers’ cognition exceeds 75%, and the overall percentage of students’ cognition exceeds 50%. In the second stage test result of practical application, the experimental class is higher than the control class up to 3.87, and the effect of the integrated autonomous teaching model is distinct.