Typhoons can cause significant damage along their paths upon landfall, making accurate typhoon trajectory prediction crucial for disaster prevention. Traditional numerical forecasting methods, while comprehensive, require substantial computational resources and lack real-time capabilities. Existing learning-based methods primarily focus on extracting historical typhoon tracks and environmental information for short-term prediction, but their performance on long-term prediction remains suboptimal. To deal with this issue, we propose a physics- and data-driven hybrid approach for long-term typhoon trajectory prediction. Specifically, Weather Research and Forecasting (WRF) is used to obtain long-term typhoon path prediction features, which are combined with statistical features and spatial environmental features. The combined features are processed through a comprehensive Multi-ConvGRU model to extract spatio-temporal characteristics and predict typhoon trajectories. Experimental results demonstrate that the proposed method significantly improves long-term prediction accuracy by 62.96% compared to the original model.