Prediction Of Ship Movement Based On Parallel Grey Neural Network

The short-term prediction of ship vertical movement posture is very important to enhance the accuracy of the ship-weapon system and the safety factor of capturing planes. Because of the influence of ocean waves, winds and other interactions, ships have complex movement in six free degrees, which is very random and nonlinear. So, it is much difficult to predict the ship movement posture.By comparative analysis of the history background and the research actuality home and abroad of the short-term prediction for ship movement posture, we find the traditional time-series analysis and predicting theory, which is based on the linear models, has good prediction effect for linear systems, but not for nonlinear system time series.On the basis of the grey system theory and neural network theory, from the point of view of combinatorial optimization, the paper establishes an online real-time prediction model for nonlinear system, that is, equal dimension recursive parallel grey neural network model(EDRPGNN) by the concept of effective degree. And the model is applied to the short-term prediction for ship movement posture for the first time. The main work in this thesis is as follows:1. According to the character of historical data of shipping movement, and considering the GM(1,1)prediction model is fit for the monotonous incremental sequence, the function transformed GM(1,1) prediction model is chosen.2. Around the modeling and prediction schemes based on neural network, we studied the different neural network models. Considering the request of prediction reliability, the real time nature, time property, we make choice among different neural network models, and decide to use the RBF neural network, which has faster training speed and better generalization ability.3. Through the introduction of the concept of effective degree, we combine a function transformed form of grey system theory prediction--GM (1,1) model and the forecasting model of RBF neural network well.4. With the powerful software Matlab, we determine all parameters in the model, the data number for GM(1,1)modeling, the speed spread parameter for RBF network and the prediction time on the basis of the experimental analysis of many data.By the analysis of simulated results from quantities of experiments through the EQRPGNN model, it is known that, the prediction model can increase the prediction precision and expand the prediction time, which implies the prediction model present in the paper is reasonable and feasible.