Power system transient stability assessment is the key to the safe and stable operation of power grids. As the complexity and volatility of power grid operation modes continue to climb, the application of both deterministic and probabilistic assessment methods in different scenarios is limited. Transient stability assessment in conventional scenarios using deterministic physical models is too slow and the calculations are too absolute for decision making such as risk assessment, while in extreme cases, where there is less historical data and the mode of operation fluctuates, the use of probabilistic models can lead to large biases. Therefore, this paper proposes a hybrid model-data driven approach for transient stability assessment of power system. Firstly, scenario identification is performed based on K-means clustering. Secondly, in general scenarios, the probabilistic assessment method of transient stability based on gaussian process regression (GPR) for particle swarm optimization (PSO) is used; and in extreme scenarios, the deterministic assessment method based on time-domain simulation is used. Finally, the validity of the method proposed in this paper is verified in IEEE39 node system.