Research On Ship Motion Attitude Estimation And Prediction Methods

The vigorous development of the shipping industry and the continuous expansion of maritime trade have made ship research a hotspot in the academic circle.Among them,the research on ship's swing motion has attracted much attention.The fierce ship swing motion will seriously affect the crew's comfort and the safety of the ship,and may even lead to overturn.Once these security hazards erupt,they often cause incalculable losses.Therefore,effective and accurate estimation and prediction of the ship's swing posture has become an important issue.The specific content includes the following aspects:Firstly,in order to describe the ship's motion characteristics more realistically and effectively,the mathematical model of ship motion and the ship's steering gear model are theoretically analyzed.A six-degree-of-freedom nonlinear mathematical model that can effectively reflect the actual motion of the ship is established,and a mathematical model of ship plane motion is derived.The model of marine environment disturbance,including wave disturbance force and moment,sea winds and ocean currents,is established.According to the random sea wave of long-crested wave theory,the surge force,sway force and yawing moments of the wave in three different situations are simulated and analyzed.Secondly,aiming at the problem of abnormal mutation of measurement values in ship motion attitude system,this paper proposes a ship motion filter method based on Modified Unscented Kalman Filter(MUKF).An outlier detection function based on measurement residual statistics is designed to determine whether the measurement value of the system is abnormal.Subsequently,according to the measurement residual covariance,the filter gain is modified,the nonlinear filter of ship motion is designed to reduce the influence of measurement anomaly on ship motion attitude system,making the estimation of ship motion attitude have more accurate precision and better timeliness.Thirdly,among the six-degree-of-freedom of the ship's motion attitude,the rolling,pitching,and heaving motions largely determine the seaworthiness of the ship.Therefore,a detailed study is conducted on the extremely short-term prediction of the motions on these three degrees of freedom.The Long Short-Term Memory(LSTM)neural network model is deeply analyzed,and the extremely short-term prediction model of LSTM ship motion attitude based on Particle Swarm Optimization(PSO)is proposed.A Multi-layer Heterogeneous Particle Swarm Optimization(MHPSO)algorithm is proposed to address the problem that particles in PSO are easy to aggregate to their local optimal position and fall into local extremum.By establishing a particle behavior pool,the premature particles are randomly selected for behavior,which enhances the information interaction ability between particles and particles during the running of the algorithm,as well as improves the optimization ability of the algorithm.By taking into account the nonlinear and nonstationary characteristics of actual ship motion,this paper designs an extremely short-term prediction model of ship motion attitude based on the combination of Empirical Mode Decomposition(EMD)and improved LSTM.Through EMD,a finite number of independent Intrinsic Mode Functions(IMFs)can be obtained,which can highlight different local feature information of the original data.An improved LSTM neural network prediction model is established for each IMF component,and the results of each prediction component are summed,thus the final prediction result is output.Finally,an online recursive modeling prediction method for ship motion attitude is studied.Ship motion is a dynamic process that changes with time.During the prediction process,new data are generated continuously.The correlation between older samples and current and future motion features will be smaller and smaller,but more relevant to newer samples.The prediction model using the off-line training method does not consider the dynamic characteristics of the sample during training,resulting in a long-term prediction accuracy degradation and poor real-time performance.Therefore,for this problem,a combined recursive modeling prediction model is proposed,which integrates Fast Sparse Approximation Least Squares Support Vector Machine(FSALS-SVM)with Bidirectional Long Short-Term Memory(Bi LSTM)neural network.By training the sample data using Bi LSTM neural network,training residuals are obtained,which are then modeled and predicted using FSALS-SVM algorithm.The error compensation mechanism is used to automatically update the parameters of the Bi LSTM model to predict the ship's motion attitude in real time.This modeling method can adjust the expression of the prediction model during different periods of the motion process,and has certain adaptive control ability.