Abstract: The learner's historical practice sequence has varying degrees of influence on the current answer, and the existing deep knowledge tracking model is relatively insufficient to consider the learner's learning transfer process. Aimed at this problem, a deep knowledge tracking optimization model based on production transfer theory is proposed. Based on the theory of production transfer, the model used a knowledge growth matrix to represent the knowledge and skills acquired by learners after practice. It took the historical knowledge growth matrix sequence as input and used the self-attention mechanism to construct the learner's learning transfer process. The influence value matrix predicted the probability that the learner would answer the next question correctly. Experimental results show that the model improves the prediction accuracy of knowledge tracking, and the model structure is more interpretable.