Online Prediction Algorithms For Ship Attitude Data Stream

The advancement of the national navigation strategy accelerates the ship monitoring system,with particular importance of forecasting and mining the data stream of movement obtained in the monitoring system effectively.During the course of journey,the ship will sway uncontrollably under the disturbance of wind and wave.Especially when navigating in harsh sea conditions,the excessive swaying motion will pose a serious threat to the lifting and landing of the carrier aircraft,the timely roll reduction of the ship and the accurate launching of the carrier missile.If the ship's movement trend can be predicted in advance for a few seconds or a dozen seconds,it can respond to the negative effects of swaying in a timely manner,which is of great significance to the safe operation of ships.The on-line prediction of ship's data stream of motion attitude generally refers to the real-time forecast of the future trend of the data stream,i.e.the real-time prediction of ship's attitude in a period of time(3-15s)by certain scientific means.Since the application of historical data to prediction belongs to the research category of data mining,this thesis focuses on online mining of the movement attitude data stream acquired in the monitoring system.Considering the common characteristics of both motion attitude and data flow in ship motion attitude data stream,this thesis presents an improved chain rewritable sliding window technology based on data stream mining,and applies it to extracting the outline data structure in data stream.The data set in the outline data structure is denoised by wavelet transform,and then the Volterra kernel estimation based on Kalman filtering algorithm is employed in the real-time prediction of continuous motion attitude data stream collected in monitoring system.Experiments show that this method has obvious advantages in accuracy and computational efficiency.In view of the common characteristics of both motion attitude and data flow in ship motion attitude data stream,this thesis proposes an improved chain rewritable sliding window technology to extract outline data structure from data stream.The data set in outline data structure is denoised by wavelet transform,and then the Volterra kernel estimation and prediction based on Kalman is used.The method realizes the real-time prediction of the continuous motion attitude data stream collected in the monitoring system.The results show that the method has obvious advantages in accuracy and computational efficiency.The research is as follows: Firstly,this thesis introduces the uniqueness of data stream and its main corresponding query models,and elaborates on the purpose,process,framework and common algorithms of data and data stream mining to achieve a better comprehension of the core idea of data mining and make it suitable for subsequent researches.Secondly,considering the chaotic characteristics of the ship's motion attitude,two algorithms,Volterra series and Kalman filter,as well as their combinational algorithms,are used to predict the motion attitude,verify the advantages of the combinational algorithm,and realize the real-time processing of data stream combined with the traditional sliding window method.Finally,the existing problems of the traditional sliding window technology are analyzed,and an improved chain rewritable sliding window technology is proposed for data stream processing accordingly.Considering the noise problem in the motion sequence,the improved chain rewritable sliding window is used to obtain the outline data structure,then the data set is denoised by wavelet threshold,and then the Volterra kernel estimation based on Kalman filtering algorithm is employed to achieve the goal of prediction.The practical verification in ship motion attitude prediction shows that this algorithm can solve the problem of online prediction of motion attitude data stream.