Research On Prediction Method Of Drift Trajectory Of Crashed AUV

Autonomous Underwater Vehicle(AUV)plays an important role in ocean development and defense due to its high flexibility and low noise.However,AUV is unmanned and operates in complex marine environments,and accidents cannot be avoided.Therefore,conducting research on the drift trajectory prediction of crashed AUVs is of great significance for salvage and rescue missions,as well as for the detection of fault causes in the later stages of AUVs.It is worth noting that due to the complex drift motion of the crashed AUV,the mutual interference between multidimensional position information makes it difficult to establish model relationships,and the layering phenomenon in the ocean can lead to feature drift and low prediction accuracy in the prediction process.To address the above issues,this article will conduct research on predicting the drift trajectory of crashed AUVs from two aspects: depth and latitude and longitude.The main research work is as follows:(1)A depth measurement model is proposed to address the issue of limited effective features and lack of representativeness in the depth trajectory data of crashed AUVs.Through feature fitting modules and feature extraction modules,a more accurate mapping relationship between depth and features is constructed from both spatial and temporal perspectives.Regarding the construction of feature fitting modules,based on two ocean state equations with different ocean layers,the relationship between depth trajectory and ocean parameter conversion is studied,and two depth measurement functions are proposed.Secondly,by setting the depth measurement function as the connection function of the hidden layer,the prediction model has higher function fitting ability and convergence speed,resulting in more accurate prediction results.(2)A modified neural network model based on emotion regulation mechanism is proposed to address the feature drift problem of AUV in the process of longitude and latitude prediction.By studying the brain tissue structure and function in the emotional regulation mechanism,the data matching module and memory bank module in the model are constructed.And for the construction of the memory module,a sequential data segmentation method is proposed to obtain several trajectory subsequences with significant data distribution differences,thereby improving the diversity of model parameters in the memory and achieving more accurate matching with predicted data.Then,according to the different data segmentation situations of different ocean layers,based on the long path and short path adjustment methods in the emotion regulation mechanism,two matching mechanisms for sequence data are established to search for the optimal model that matches the data from the memory and improve the accuracy of prediction.(3)An efficient AUV trajectory prediction system based on the proposed crash AUV trajectory prediction method is designed.The system is constructed using the Pyqt5 graphical framework,which mainly includes functions such as trajectory prediction type selection,prediction model selection,data segmentation module and memory library module construction,and helps researchers observe the training results and trajectory prediction results of the model intuitively and conveniently through the system visualization area.