Due to the rapid response and flexible deployment capabilities of Flying Ad-Hoc Networks(FANETs), it holds significant potential in the specific areas without ground communication infrastructure such as post-disaster relief and battlefield communication. However, FANETs face significant challenges in ensuring fast, efficient, and reliable data transmission due to the highly dynamic network environment and the sharp increase in data traffic. Therefore, to tackle the issues of data reliability and low-latency transmission within FANET environments, characterized by high dynamism and significant fluctuations in data traffic, we propose a FANET routing approach based on a Deep Reinforcement Learning (DRL). This approach employs a Multilayer Perceptron (MLP) model to acquire each Unmanned Aerial Vehicle’s (UAVs) location, hovering time, Round-Trip Time (RTT) of historical messages, and the amount of data awaiting transmission in the network, adjust the distribution of UAV positions and hover time based on these network information to maintain the stability of the network transmission links. Then, considering multi-metric factors of current network performance, We formulate the routing decision issue as an optimization problem of FANET link performance. Experimental simulation results have verified the effectiveness of the Multilayer Perceptron-based Deep Reinforcement Learning FANET Routing approach (MDR). Compared to AODV, its load balancing and latency were reduced by 17.2% and 47.81%, respectively. The packet loss rate and bit error rate were decreased by 50.7 and 42.6%, effectively dispersing network traffic and reducing transmission delay, enhancing the reliability of data transmission in the network.