Unmanned aerial vehicle (UAV) clusters offer significant potential in civil, military, and commercial fields due to their flexibility and cooperative capabilities. However, characteristics such as dynamic topology and limited energy storage bring challenges to the design of routing protocols for UAV networks. This study leverages the Deep Double Q-Learning Network (DDQN) algorithm to optimize the traditional Greedy Perimeter Stateless Routing (GPSR) protocol, resulting in a multi-objective optimized GPSR routing protocol (DDQN-MTGPSR). By constructing a multi-objective routing optimization model through cross-layer data fusion, the proposed approach aims to enhance UAV network communication performance comprehensively. In addition, this study develops the above DDQN-MTGPSR intelligent routing algorithm based on the NS-3 platform and uses an artificial intelligence framework. In order to verify the effectiveness of the DDQN-MTGPSR algorithm, it is simulated and compared with the traditional ad hoc routing protocols, and the experimental results show that compared with the GPSR protocol, the DDQN-MTGPSR has achieved significant optimization in the key metrics such as the average end-to-end delay, packet delivery rate, node average residual energy variance and percentage of node average residual energy. In high dynamic scenarios, the above indicators were optimized by 20.05%, 12.72%, 0.47%, and 50.15%, respectively, while optimizing 36.31%, 26.26%, 8.709%, and 69.3% in large-scale scenarios, respectively.