Online Modeling And Predicting Of Ship Motions In Waves Based On Wavelet Neural Network

Ship motion prediction is an important issue in the field of ship navigation performance research,which is related to ship navigation safety and operational efficiency.Precisely predicting the motion attitude of the ship and then using the motion compensation and control technology to control ship motion can improve the safety of aircraft landing on carriers,the aiming accuracy of ship fire control system,the efficiency of cargo transfer between ships and so on.It is of great significance for improving the safety and efficiency of offshore operations.The mathematical model for ship motion prediction is the basis of ship motion prediction and controller design.Due to the disturbances from environmental factors such as wind,waves,currents,as well as the change of ship's loads,ship motion at sea is a complex nonlinear time-varying system,and it is very difficult to establish an accurate mechanism model for ship motion.Therefore,many methods for ship motion modelling and prediction are based on statistical theory.In the early days,the classical time series analysis methods,such as autoregressive models,autoregressive moving average models,have been used.In recent years,with the development of artificial intelligence technology,machine learning,such as neural networks and support vector machine,have also been widely used in ship motion modeling and prediction.Neural network can approximate any nonlinear function with arbitrary precision.Wavelet neural network is proposed on the basis of wavelet theory.Wavelet neural network composed of wavelet activation function has not only the advantages of traditional neural network but also the advantages of wavelet function.The wavelet function has local characteristics in both the time domain and the frequency domain.Especially for fitting strong nonlinear systems,wavelet neural networks usually converge faster.In this thesis,for the highly nonlinear ship motion,adaptive wavelet neural network and fixed grid wavelet neural network are used to identify the prediction model of ship motion in waves.Adaptive wavelet neural network is generally trained by the gradient descent type algorithm,which easily makes the wavelet network converge to local optimum.In order to avoid the wavelet neural network converge to local optimum,the combination of particle swarm optimization algorithm and error backpropagation algorithm is used to train the neural network.It overcomes the shortcoming being sensitive to the initial position of the neural network caused by the error back propagation algorithm,in the meanwhile improves the robustness of the trained wavelet network.In this thesis,the Mexican Hat wavelet activation function is used to construct a neural network model,and the model is applied to prediction of ship roll motion in regular waves and irregular waves.In order to predict ship motion with multiple degrees of freedom,the modeling method is extended to multiple input multiple output systems.The coupled heave-pitch motion in waves is studied,and the effectiveness of the modeling method is verified by comparing the simulation data of ship motion with the experimental data.There are many kinds of wavelet activation functions,and different wavelet activation functions can construct different wavelet neural networks.Different wavelet network models can be used for different application environments.In this thesis,two adaptive wavelet neural networks are constructed by using Mortlet and Gaussian wavelet activation functions.Based on the simulation data of ship roll motion and coupled heave-pitch motion in irregular waves,the training method combining particle swarm optimization algorithm and error back-propagation algorithm is adopted to establish respectively the Mortlet and Gaussian wavelet neural network based prediction models for ship roll motion and coupled heave-pitch motion in waves.This embodies the flexibility of model construction of wavelet neural network.Based on particle swarm optimization algorithm,the model of adaptive wavelet neural network can,to a certain extent,avoid the neural network being trapped in local optimum.However,when modeling the multiple input multiple output system,since the objective function aims at the overall optimization effect of the system output,there may be a large difference in the optimization results of different dimensional outputs during the training optimization process.To avoid this phenomenon,in this thesis,a modeling method based on fixed grid wavelet neural network is adopted.As long as the training conditions of the fixed grid wavelet neural network are the same,the training result of this modeling method is unique,and the model can converge to the global optimal.Meanwhile,the wavelet neural network model can clearly indicate the relationship between different input variables and their contributions to system output,as well as the coupling relationship among input variables.Based on the data of ship roll motion and coupled heave-pitch motion in waves,the modeling method is applied to the online identification modeling and prediction of ship roll motion and coupled heave-pitch motion in waves.In order to improve the computational efficiency of the online modeling method based on fixed grid wavelet neural network,an online modeling method called coarse and fine tuning fixed grid wavelet neural network is proposed in this thesis.In the modeling method,coarse tuning can change the structure of neural network model,while fine tuning only adjusts the parameters of neural network model by means of Givens Rotation algorithm.This modeling method can not only adjust the model structure flexibly,but also improve the computational efficiency of modeling.Based on the simulation data and experimental data of ship roll motion and coupled heave-pitch motion in waves,the method is used to identify the models of ship roll motion and coupled heave-pitch motion.The online prediction models of ship motion in waves are established for the multiple input single output system and multiple input multiple output system.The simulation results show that the model can online predict the ship motion well.In this thesis,the online modeling method of adaptive wavelet neural network based on particle swarm optimization is introduced for online modeling and predcting ship motion in waves.Based on fixed grid wavelet neural network,a new online modeling method of coarse and fine tuning fixed grid wavelet neural network is proposed,which provides an effective method for online modeling and prediction of ship motion in waves.