Electrical tree cracks are one of the main forms of failure in insulating materials of electrical equipment. Their propagation path and speed are influenced by various environmental factors and exhibit high uncertainty, making it difficult to accurately describe using traditional physical models. This paper proposes a prediction model for electrical tree crack propagation based on graph representation, combining the phase-field method and Graph Neural Networks (GNNs), aiming to address the complexity and efficiency issues in predicting electrical tree crack growth. Firstly, the phase-field method is used to construct a dataset of electrical tree propagation under multiple stress conditions to represent crack morphology. Then, based on a graph-driven deep learning method, a spatiotemporal feature extraction model is designed to predict the propagation path of electrical tree cracks. Experimental results show that the proposed model can effectively capture the spatiotemporal correlation of electrical tree propagation, significantly improving prediction accuracy and supporting model deployment. This study provides a new solution for the intelligent perception and health management of electrical equipment insulation performance, with significant engineering application value.