To solve the problems of poor stability and low accuracy of traditional machine learning in transient stability assessment and the limitations of offline training, an online transient stability assessment model based on multi-model fusion Bagging ensemble learning method is proposed. First, in combination with research on the frontier theory of artificial intelligence, the principles and implement methods of seven machine learning algorithms commonly used in transient stability assessment are analyzed, and the Bagging method is used to integrate them to give full play to the advantages of each model. Secondly, the mathematical method of Bagging ensemble learning is given and a simulation experiment is carried out. When the topological structure of the original system changes, a Boosting algorithm and transfer component analysis are used to carry out sample and feature transfer of the original grid historical data to complete the online update of the proposed model. IEEE10-machine 39-bus system and IEEE16-machine 68-bus system are used in the simulation analysis, and the results show that the proposed method is more accurate than the traditional machine learning model. It can maintain stable operation when the data is mixed with noise, and accurately evaluate transient stability by transferring the historical data when the system topology changes. This work is supported by the National Key Research and Development Program of China (No. 2017YFB0902600).