Abstract: Knowledge Tracing (KT) is the core function of intelligent tutoring systems. It estimates the knowledge states of a student on each time step based on historical interaction logs, and then predicts the student's perfor-mance in solving new exercises. A stable knowledge tracing model based on learning transfer (SKT-LT) is proposed to address the problem of existing knowledge tracing methods not considering the stable evolution of students' individual concept knowledge states and overall knowledge states between adjacent time steps. On the one hand, the learning transfer effect between knowledge concepts is utilized to optimize the knowledge tracing process. On the other hand, by introducing the stability constraint of single concept knowledge state of student and the stability constraint of overall knowledge state of student in the knowledge tracing process, the predicted knowledge state of the model does not underwent abrupt changes in adjacent time steps, thereby improving the accuracy of the model's predictions. Finally, comparative experiments were conducted between the SKT-LT model and the DKT, CKT, ContextKT, DKVMN, SPARSEKT, GKT, and SKT models on two publicly available datasets (ASSISTments 2015 and ASSISTments 2009). The experimental results show that the AUC and F1-Score values of the SKT-LT model on the ASSISTments 2015 dataset improved by 3.45% and 22.80%, respectively, compared to the best performing baseline model SKT. Meanwhile, ablation experiments demonstrate the effectiveness of each module in the SKT-LT model, while visualization experiments have shown that the SKT-LT model can trace stable students' knowledge states.