Hybrid Model of Wavelet Transform and Wavelet Neural Network and Its Application on Time Series Prediction of Ship Roll Motion

摘要

Based on the multi time and frequency scale and the nonlinear character of ship roll motion time series, a hybrid model prediction approach combining wavelet transform (WT) and wavelet neural network (WNN) is presented in the paper. By using wavelet transform, multilevel 1-D wavelet decompositions of nonlinear roll motion are completed to obtain relative simple and regular period signals, then the multiple-input single-output (MISO) WNN is employed as prediction model for the main decomposed signals above, finally, a prediction result is given through linear combination. The simulation experiments demonstrate the proposed method can effectively reduce the prediction difficulty and obtain better prediction precision. This method can also be used in ship pitch and heave motion prediction.