Motion Attitude Prediction Of Unmanned Surface Vehicle Based On Deep Learning

In the process of building China into a maritime power,Unmanned Surface Vessel(USV)plays a vital role.During the voyage at sea,USV will be subject to a lot of random and uncertain external interference such as sea breeze,waves,swells and so on,resulting in a strong six-degree-of-freedom swing motion,which will cause great safety risks during the sea operation process.Therefore,the research on the motion attitude prediction of USV is of great significance for marine safety and the efficiency of USV operations.The six-degree-of-freedom sway motion of USV at sea is a complex time-varying nonlinear and non-stationary dynamic system,which varies with time-varying environmental disturbances as well as various sailing conditions.The conventional methods have the disadvantages of low accuracy,poor robustness,and insufficient practical application ability.The rise of deep learning provides new opportunities for USV motion modeling and prediction.In this paper,a large number of sensor data stored by USV during the mission are used to establish a model for predicting the motion attitude of USV based on deep learning.Experimental studies are carried out based on two case from “Jing Hai-VI” and “Jing Hai-III” USV of Shanghai University.Firstly,in this paper,we propose a prediction model for USV roll motion prediction based on a coupled Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM).CNN is utilized to extract spatial correlation and local temporal characteristics of the sensor data,using the LSTM modeling USV roll motion and to predict.The experimental result shows that the prediction effect of this model is significantly better than traditional methods.Secondly,in this paper,we integrate the attention mechanism,optimize the structure of the prediction model based on CNN-LSTM,and extract context information through the attention mechanism.LSTM weights the features extracted by CNN to obtain significant fine-grained features with high importance.For some local peak intervals and non-stationary change intervals,the prediction model incorporating the attention mechanism has a better prediction effect and makes up for the shortcomings of the CNN-LSTM-based prediction model in this regard.Finally,in this paper,we use these two models to model and predict the pitch and sway motion of USV,and the experiments verify the applicability of the prediction model in this paper to the prediction of USV motion attitude.