Node Localization And Trajectory Prediction For Marine Sensor Networks

An oceanic sensor network consists of various sensors with wireless sensing capabilities distributed in the ocean.It has advantages like high coverage,low power consumption,and low cost,which provides significant application values for oceanic resource development,military defense,and other related areas.However,due to the special nature of the oceanic environment,node localization and trajectory prediction are facing severe challenges.The non-linear oceanic acoustic velocity leads to large ranging errors based on acoustic signals.Moreover,the sparse anchoring problem occurs when the oceanic sensor range is too large,resulting in a significant decrease in the node localization coverage rate.The dynamic property of the oceanic environment necessitates frequent restart of the localization system,which increases network overhead.Therefore,new node localization and trajectory prediction algorithms must be designed to meet the demands of the oceanic sensor network in complex environments.In response to the urgent issues that the oceanic sensor network needs to solve,two new node localization and trajectory prediction algorithms are proposed: Conv LSTM modified ranging based Node Localization Algorithm(CML)and CNN-GRU based mobile Node Localization and Trajectory Prediction Algorithm(CGL).The CML algorithm extracts features from historical ocean data using the Conv LSTM neural network model and predicts oceanic acoustic velocity hierarchically to correct ranging errors.By using the corrected ranging values in the improved Least sqaure method(LSM)based localization algorithm,the localization precision of underwater nodes can be further improved.Besides,for the sparse anchoring problem that may arise when the network scale is too large,the CML algorithm designs an iterative localization algorithm based on located nodes to lower deployment costs and improve network localization coverage.The CGL algorithm can be divided into two phases: node localization and trajectory prediction.In the node localization phase,a dynamic scheme is used to select nodes to solve the problem of insufficient reference node numbers and improve the localization coverage rate.In the trajectory prediction phase,a hybrid Convolutional Neural Networks(CNN)and Gate Recurrent Unit(GRU)neural network model is designed to perform trajectory prediction of the node’s movement.With this approach,the location information can be updated through the movement prediction model without frequent restarts of the localization process,thereby reducing system energy overhead.This thesis conducted simulation experiments on CML and CGL algorithms,respectively.The results show that the CML algorithm can improve the node localization precision and has good localization stability.The CGL algorithm can improve the node localization coverage rate and reduce energy consumption.Furthermore,this thesis separately tested the anti-interference ability of the two algorithms,further validating the reliability of the proposed algorithms.