| In recent years,the application of UAVs has gradually penetrated into various fields such as civil,industrial and military.In order to realize the application requirements of multi-UAV collaborative operation,Flying Ad hoc Network(FANET)has been widely used.The drone network uses an open wireless channel,which is vulnerable to attacks in special fields such as military applications.Malicious nodes will intentionally drop packets,resulting in network performance degradation and even network paralysis.Based on a research project related to FANET that studies the application scenarios where UAVs are divided into multiple groups to perform different tasks,this thesis studies the security routing mechanism in the FANET environment in which communication clusters are divided by groups.The malicious node detection algorithm under the network and the security routing algorithm based on the detection algorithm are researched as follows.Malicious nodes that intrude into the network mainly use intentional packet loss as an attack method.In order to bypass malicious nodes in the routing process,this thesis first studies how to accurately detect malicious nodes in the network.In FANET,packet loss often occurs due to link interference,topology changes,etc.These normal network packet loss phenomena are easily confused with malicious node intentional packet loss behaviors,which makes detection difficult? in addition,when obtaining the forwarding behavior of nodes,malicious nodes will also deceive by sending forged information,making detection more difficult.In view of the above problems,this thesis proposes Long Short-Term Memory based Malicious Node Detection(LSTM-MND)algorithm based on long shortterm memory artificial neural network.The algorithm first combines the self-certification mechanism with the neighbor monitoring mechanism in the information collection stage,and obtains the characteristic information describing the forwarding behavior of nodes from the perspective of itself and the perspective of neighbors.By comparing the information from the two perspectives,it solves the problem that it is easy to confuse different packet loss phenomena by only using the neighbor monitoring mechanism,and it is easy to be deceived by only using the self-certification mechanism.Then based on the LSTM algorithm,the characteristics of the node in the latest multiple cycles are detected to reduce the short-term forwarding capability fluctuations of the node resulting in misjudgment by the detection algorithm.The simulation results show that in the FANET environment,compared with several other existing malicious node detection algorithms,the LSTM-MND algorithm proposed in this thesis has better results under various evaluation indicators.In order to avoid forwarding data packets to malicious nodes that intentionally drop packets and improve the end-to-end packet delivery ratio,based on the LSTM-MND algorithm,this thesis further studies and designs Secure Q-learning based Multi-objective optimization Routing(SQMR)algorithm.The algorithm evaluates the link stability and selects a more stable link to reduce packet loss.In terms of security,firstly,by introducing a direct trust target in the reward function,the security of neighbor nodes is quickly pre-evaluated,and the message is avoided as far as possible to forward the message to the node suspected of intentionally losing packets? however,the accuracy of pre-evaluation is limited,cannot be used to completely eliminate malicious nodes,and malicious nodes will move to other locations in the network to attack,so this thesis uses the LSTM-MND algorithm to evaluate security from a global perspective,and further optimizes routing decisions based on the evaluation results,avoiding malicious nodes during forwarding,and the security is further improved.The simulation results show that compared with several other existing routing algorithms,the SQMR algorithm proposed in this thesis can improve the delivery ratio of data packets and improve the security of the network. |
PDF Full Text Request