Research On Analysis And Prediction Method Of Safe Navigation State Of Unmanned Craft

The navigation of high-speed unmanned surface craft has the characteristics of nonlinearity,strong coupling and susceptibility to sea conditions.At the same time,the hull is usually small and the speed is high.Often due to improper control,the safety of the unmanned surface craft is seriously reduced.A safety incident occurred.Analyze the safe navigation status of the unmanned craft.If the movement of the unmanned craft can be predicted in advance,and early control and intervention of the unmanned craft based on the prediction results,the seaworthiness and safety of the high-speed unmanned surface craft will be greatly improved..The subject is to carry out research on time series forecasting based on deep learning,aiming to obtain a practical online intelligent forecasting method based on the intelligent forecasting of unmanned craft’s heel and pitch,which can meet the real-time performance while ensuring the accuracy of the forecast,and is easy to transplant and promotion.Firstly,analyze the safe navigation status of unmanned crafts,focus on roll and pitch that affect the safety of unmanned crafts,explore roll and pitch time series prediction methods based on deep learning,and give a deep learning time prediction structure;Based on actual ship data,complete the roll and trim time series data preprocessing and experimental environment construction,and clarify the prediction accuracy evaluation index of the roll and trim time series of the unmanned craft based on deep learning.Secondly,aiming at the problem that the existing traditional prediction algorithms and a single neural network model have low accuracy in the ship attitude prediction,a combined deep learning model of the ship attitude prediction method based on the bidirectional long short-term memory network(Bi-LSTM)fusion temporal pattern attention mechanism(TPA)is proposed..The temporal pattern attention mechanism extracts the temporal pattern of the deep features from the output features of Bi-LSTM,which are beneficial to the ship attitude prediction,and ignores other features that contribute less to the prediction.In order to objectively demonstrate the effectiveness of the algorithm,real ship attitude data is used.Compare the prediction models of SVM,LSTM,and Bi-LSTM-TPA respectively,Bi-LSTM-TPA ship attitude prediction model has better prediction accuracy than other algorithms.The experimental results of real ship data show that the proposed Bi-LSTM-TPA combined model has a significant reduction in MAPE,MAE,and MSE indicators compared with the LSTM and other model,which verifies the effectiveness of the proposed algorithm.Again,aiming at the problem that the shallow network model does not have high accuracy in ship attitude prediction,a deep bidirectional feature network based on multiple inputs of ship roll,roll angle speed,relative wind speed,relative wind direction,slew rate and rudder angle is proposed.The network introduces two Bi-LSTM branch structures to explore deep features between multiple input data to improve the accuracy of ship attitude prediction.Finally,four error indicators are used to evaluate the proposed multi-input ship roll prediction algorithm on real ship data.The applicable conditions of the single-input and multi-input ship attitude prediction algorithm are given,and the effectiveness of the algorithm is verified.Finally,the Bi-LSTM-TPA and deep bidirectional feature network model proposed in this article are deployed on the TX2 platform,realize the landing of the real-time prediction system of ship attitude.The ship posture prediction system can predict the posture of the ship at the next moment,and can also visually demonstrate the position of the ship attitude at the next moment.When the ship’s roll and trim may exceed the threshold,the system will prompt.