The Research On AUV Track Prediction Method Based On BP Neural Network

Track prediction plays a crucial role in the navigation safety of AUV(Autonomous Underwater Vehicle). Track prediction is the best way to assist AUV in evading known risks when AUV sails in complex ocean environment. However, former studies of track prediction were mainly aiming at aerospace and vessel operation, leaving a gap for effective research on AUV track prediction. Moreover, methods of building kinematics or dynamical equations,which were applied in most of former research on track prediction, might be complex or even infeasible due to high-level randomness of ocean environment’s influence on AUV motion.Therefore, a precise and feasible method is necessary to predict AUV track.Considering the practical need of AUV track prediction, this paper built a predictive model based on BP neural network and achieved precise AUV track prediction from the sample data of AUV positional information and oceanographic hydrological element.Detailed content and procedures are listed as follows.Firstly, effects of ocean environment (including current, tide, and tidal) on AUV navigation and AUV spatial motion mathematical model were analyzed. Then the prediction was promised feasibility by introducing multivariable time series. and using three-dimensional ocean reanalysis data as input data of ocean environment predicting.Secondly, according to track prediction’s requirement on sample data of AUV positions,abnormity data examination and Grey Prediction were applied to eliminate exceptional value and complement missing value. Also, interpolation to data of current, tide, and tidal was made in the light of cubic spline interpolation, in order to enhance density of data points.In addition, BP neural network was analyzed in depth. Giving consideration of BP network’s features and practical need of AUV track prediction, the network was then improved with the help of Principle Component Analysis, adaptive learning rate, and combined forecast. Therefore, this research established three complexity-increasing predicting models: prototype model,combined model,and ocean environment model.Finally, this paper took data-processed AUV positional information and ocean reanalysis data as input data, achieving simulation of the three predicting models in Matlab. The precise AUV track prediction was finally accomplished with the exertion of ocean environment predicting model.