<![CDATA[Flying Ad Hoc Network (FANET) is a mobile wireless network formed by multiple Unmanned Aerial Vehicles (UAVs) with highly dynamic topology. This rapid topology change makes routing more complex than in conventional Mobile Ad Hoc Networks, since UAV movement can continuously affect link quality, disconnection probability, and packet delivery delay. This study applies Long Short-Term Memory (LSTM) to optimize FANET routing using time-series network metrics, including signal-to-noise ratio, delay, throughput, energy, packet loss rate, jitter, and bandwidth utilization. The LSTM model learns temporal relationships among network conditions, enabling next-hop selection to consider not only current link status but also its evolution over time. The proposed method is evaluated against AODV, OLSR, and the Stochastic Probability Algorithm (SPA) using Packet Delivery Ratio (PDR) and end-to-end delay under different numbers of UAVs and UAV speeds. Results show that LSTM consistently achieves the highest PDR across all scenarios. For UAV number variation, LSTM improves PDR from 0.166 to 0.380, outperforming AODV, OLSR, and SPA. For UAV speed variation, LSTM maintains PDR between 0.89 and 0.73, remaining superior to the comparison methods. In addition, LSTM produces the lowest delay, ranging from 0.60 to 0.70 s for UAV number variation and 0.35 to 0.61 s for UAV speed variation. These results demonstrate that LSTM effectively captures the temporal dynamics of FANET and is suitable for adaptive routing support.]]>
Copyrights © 2026
