On The Neural Network Theory And Its Application In Ship Motion Prediction

There are a large number of nonlinear systems in practice, so the study of nonlinear systems and prediction method are very important. Large-scale ships can have 6-motions complex movements because of the ocean waves, the sea breeze and other disturbances which have the very randomness and the non-linearity, this caused to extremely difficulty to the ships movement posture short-term prediction. This article leaves from the neural network theory, fully studied one kind to be suitable for the nonlinear system modeling prediction neural network -Diagonal Recurrent Neural Network. This paper mainly includes:1. The collection and collation of the correlation data about the ships motion posture prediction technology in domestic and foreign, and proposed one improvement classical spectral estimate method about the ships motion in this foundation. And has proven this method usability and the reliability through the actual simulation.2. Introduced the elementary theory on neural network, research to the BP neural network thoroughly, used the improved conjugate gradient algorithm to the BP neural network, therefore enabled to improve the network performance and increase the prediction precision, and carried on the simulation computation with the actual example.3. Research and discussed to the approximating ability thoroughly on neural network using the MATLAB software and the BP neural network toolbox, elaborated the approximation of function and it's MATLAB realization method based on the BP neural network in detail through an actual example.4. Research to the Recurrent Neural Network(RNN) theory, strictly inferred the dynamic reversing dissemination algorithm aboutthe Diagonal Recurrent Neural Network (DRNN) and theGroup Diagonal Recurrent Neural Network (GDRNN), the rule of renewing the power value and proof the stability of the training algorithm. For the more have produced a more precise result of the study-rate 77.5. Established the structure of the diagonal recurrent neural network and the Recurrent Prediction Error (RPE) algorithm applying on the foundation of the time series prediction method in large-scale ship motion , analysis to this algorithm's stability, convergence and unbiasedness, comprised to the result from the autocorrelation law and from the periodogram prediction method, the prediction result showed this algorithm feasibility.