| Underwater sensor networks is an important means of underwater detection because it is easy to deploy and practical.However,if the corresponding spatial location of the sensor node collecting data is not known,the monitoring station will not be meaningful even if it receives the data collected by the sensor node.Therefore,as the core technology of underwater sensor networks,underwater node positioning technology has received wide attention from researchers.However,the multipath effect,ocean noise,sound velocity variability in the underwater environment make the positioning process of underwater nodes appear many non-linear changing factors.Depth learning is an effective method to solve complex non-linear problems.In this paper,depth learning algorithms are applied to range-based positioning technology and range-free positioning technology respectively,and the following research is carried out:Range-based underwater node positioning algorithm is common,but the accuracy of ranging between nodes is affected by underwater sound velocity variability,which affects the positioning effect of underwater nodes.To solve the above problems,this paper presents a location algorithm for ranging nodes based on GA-BP(Genetic Algorithm Back Propagation)neural network.This algorithm relies on GA-BP network and makes use of the global Argo grid dataset to build a deep learning prediction model that can obtain accurate underwater sound velocity in different space-time and different sea areas.This model,combined with echo ranging and spatial four-point location algorithm,can locate unknown nodes.On this basis,a location correction mechanism based on multipoint average predicted sound speed is proposed,which replaces the single point fixed predicted sound speed with the multipoint average predicted sound speed along the anchor node and the node to be corrected,and further improves the positioning accuracy.Underwater node positioning algorithm without ranging is often dependent on the connectivity of sensor networks and the positioning accuracy is low.The main reason for this is that the communication range of each underwater node is different and the positioning process is susceptible to the influence of complex underwater environment factors.To solve this problem,a range-free node location algorithm based on RBF(Radial Basis Function)neural network is proposed.The algorithm learns the hops and position relationship between anchor nodes and establishes a distance prediction model based on RBF neural network.The model can predict the distance between nodes based on the number of hops of communication between nodes.Substituting the distance predicted by this model into the maximum likelihood estimation method,the exact location of the node to be located can be solved.This algorithm does not need distance measurement and inference,and the calculation process is simpler and more practical.Both algorithms presented in this paper solve the node positioning problem caused by complex underwater environment factors through deep learning technology.The simulation results show that the location algorithm of ranging nodes based on GA-BP network can improve the accuracy of sound speed and the accuracy of ranging and positioning between underwater nodes.The location algorithm based on RBF neural network without ranging nodes has higher positioning stability and accuracy under different communication range of nodes and different proportion of anchor nodes. |
