Optimized FANET Routing Algorithm with Reinforcement Learning Based on Function Approximation

Abstract

The high-speed movement of nodes in Flying Ad-Hoc Network(FANET) has caused difficulties in maintaining the links of the FANET routing protocol.To address the problem,an algorithm named QLA-OLSR is proposed based on Reinforcement Learning(RL) for adaptive optimization of link state routing.By sensing the changing number of the node neighbors and the service loads in the dynamic environment,the Q-learning algorithm in RL is used to construct a value function.On this basis,the optimal HELLO time slot is solved to improve the performance of the node in link detection and maintenance.Then the State Similarity Mechanism(SSM) of the improved Kanerva coding algorithm is used to reduce the complexity of the algorithm while increasing its stability. Simulation results show that the QLA-OLSR algorithm can significantly improve the network throughput,reduce the overhead of routine maintenance,and is capable of self-learning.It is suitable for FANET in a highly dynamic environment.