Bidirectional Q-Learning based Multi-objective optimization Routing Protocol for Multi-Destination FANETs

Flying Ad-hoc Networks (FANETs), with their superior characteristics, offer solutions for many complex and hazardous scenarios in both military and civilian domains. Multi-destination FANETs, which include several destination nodes that can communicate with each other, are important forms of FANETs and widely used in the field of emergency communications. However, current researches on routing protocols for FANETs primarily consider scenarios with only a single destination in the network, and these protocols often exhibit poor performance in multi-destination scenarios. This paper proposes a Bidirectional Q-learning based Multi-objective optimization routing protocol (BQMR) to tackle challenges in multi-destination FANETs, including the higher network load and the enhanced routing complexity. In BQMR, the load occupancy of neighbor nodes is additionally incorporated into the reward function to increase the resistance to network congestion under the higher network load. We also propose the bidirectional Q-value updating mechanism which enables the relay node update its Q-value for both source and destination nodes within a single transmission. This mechanism accelerates the network's perception of topological changes, which in turn significantly enhances the routing efficiency among multiple destinations. Simulation results show that compared with QMR, BQMR can provide lower delay, higher packet transmission ratio and throughput in multi-destination FANETs.