At present

deep learning is widely used in transient stability assessment of power systems

yet there is a lack of continuous assessment schemes for the constantly evolving grid data flow. Moreover

with the continuous emergence of diverse samples

data management and class imbalance problems have become urgent challenges. Therefore

this paper proposes a continuous learning framework for transient stability assessment in power systems based on knowledge distillation. The framework uses knowledge distillation to achieve lightweight model initialization during offline training endows the model with continuous learning capabilities during online updates

improves its generalization for different operating scenarios

overcomes historical forgetting

and compromises sensitivity to historical and new cases. To address the issues of insufficient online sample representation and data accumulation

a multidimensional time-series clustering sampling algorithm based on intercorrelation distance is used to select the most representative samples for model updates. Additionally

to mitigate misjudgment risks from class imbalance

a class imbalance corrector is built after the Transformer classifier. Case analysis confirms that the proposed framework offers good continuous assessment in transient stability assessment.