It was an important problem for the intelligent navigation of unmanned vehicles that planning the optimal path in the actual traffic environment. At present, many researches about global path planning of unmanned vehicle mainly focused on the improvement of algorithm solution speed in the simulation environment. Most of them just only considered the optimal path distance or the current road conditions, also ignored other factors and future changes in the actual environment. In order to complete the complex task that competing global path planning of unmanned vehicle in dynamic road network, this research put forward a framework of unmanned vehicle driving system for practical environment based on the thought of planning after prediction, and put forward DP-DQN which was a fast global path planning method combined with deep Q learning and deep prediction network technology. This method used the road characteristic data such as time and space, weather et al to predict the future traffic situation, and then competed the global optimal path. Finally, experimental results based on open datasets showed that the proposed method reduced driving time 17.97% at most than Dijkstra, A*, algorithm et al.