Abstract: In recent years, the data-driven methods have been widely used in transient stability assessment. However, the traditional data-driven methods are mostly used to analyze European data, and the description of power system topology is relatively limited, leading to the insufficient generalization ability of traditional methods in the case of frequent topological changes. In this paper, we introduced the message passing neural network (MPNN)-based approach for fast transient stability assessment of power system which is based on the steady state data. Through graph data processing and topological modeling, a transient stability assessment model which can describe the topological changes of power system was obtained. The proposed method has promising performance on the dataset with frequent topological changes, and can generalize to new topologies better than other data-driven methods. A comprehensive case study covering more than 600 topologies with more than one million simulation samples on New England 39-bus system validated the proposed method.