Renal artery embolization is an interventional procedure used to treat conditions such as renal artery rupture and renal cancer. It offers advantages such as minimal damage, fast recovery, and low side effects. The implementation of robotic wire navigation in interventional surgery can effectively assist doctors in performing the procedure. Deep learning and reinforcement learning methods have been widely used for wire navigation tasks. However, they face challenges such as overly simplistic simulation environments, single reward functions, and slow convergence speed. To address these issues, we propose the use of a virtual training environment that models real vascular projections, thereby closely resembling the real environment. We incorporate the distance information between the wire tip and the target point into the reward function and utilize real-time images as input states. We accelerate the convergence of the algorithm using the multi-threaded Proximal Policy Optimization (PPO) algorithm and adopt a multi-stage training approach. The results demonstr-ate that our method effectively achieves wire navigation in the virtual environment, reducing training time and improv-ing the success rate of wire navigation and the robustness of the algorithm.