Abstract: The spaced repetitive learning method has played an important role in assisting students in self-directed learning. However, traditional spaced repetition algorithms are too rigid in spaced control, resulting in significant differences in daily learning tasks for students, which in turn affect learning efficiency. In order to improve the efficiency of self-directed learning, we propose a cognitive spaced repetitive learning method based on ACT-R. On the one hand, by basing on ACT-R programming learning process, this paper simulates student learning behavior and extracts activation parameters for the dominant model's memory changes. On the other hand, a forgetting curve cutting algorithm is proposed, which reflects the forgetting characteristics in learning planning and extracts parameters such as memory retention rate and recommended review interval. Finally, based on the learning parameters obtained from both approachs, this paper dynamically generates spaced repeat learning plans for specific learning tasks. Experimental analysis shows that, compared to traditional spaced repetitive learning algorithms, the cognitive spaced repetitive learning method based on ACT-R can arrange autonomous learning tasks more reasonably and effectively. It achieves a more balanced daily amount of learning tasks and a more reasonable distribution of learning time for each task.