Vehicle motion prediction is of great significance in human-machine shared driving, as it is a basis for the virtual driver to make suitable driving strategies so as to better assist the human driver or even complete its own autonomous driving. While recent studies have achieved good results by applying an intention prediction module into the multi-modal trajectory generation strategy, they mostly give supervision between intention proposals and final predicted points with insufficient alignment method. This leads to a relatively low accuracy of the vehicle motion prediction. To tackle this challenge, we propose a graph-based model with mixture of experts (MOE) based intention prediction module, introducing an efficient scenario-based intention prediction mechanism to improve the performance of intention prediction task. Several testing was conducted on the Argoverse motion forecasting dataset, which showed that our model excels in predicting trajectories for multiple agents. And the ablation experiments have verified the efficency of our method.