Research On Volterra Series Forecast Method And Application In Ship Motion Prediction

Research on model forecast method of nonlinear system is vitally significant, because there are plenty of nonlinear systems in the application of practical engineering. Volterra series nucleus is the essential characteristics of nonlinear system. Based on chaotic time series in the adaptive filter forecast method of nonlinear functional transform, Volterra series can describe nonlinear behavior that possesses the function of response and memory, which has been widely used. Therefore, this paper focuses on the identification method of Volterra series nucleus and applying it into the nonlinear prediction model constructing ship motion chaos time series so as to realize the prediction of ship motion. The main research contents are as follows:First of all, According to the time sequence analysis of characteristics of chaos as well as chaotic time series of phase space reconstruction theory, the paper analyzes the chaotic characteristic of ship motion sequence and proves that the ship roll motion time series have chaos characteristics. Meanwhile, it also systematically analyzes and researches the Volterra series adaptive prediction models laying a theoretical basis on the chaos time series prediction of the ship motion research.Secondly, the paper deeps in the analysis of the LMS (Least Mean Square, LMS) algorithm theory and the related LMS algorithm, studies the LMS algorithm to identify the Volterra series nucleus estimation algorithm profoundly, and puts forward NLMS (Normalization Least Mean Square, NLMS) and VSS-LMS (Variable Step Size Least Mean Square, VSS-LMS) to identify the Volterra series of nucleus estimate algorithm. Based on the forecast of ship motion, the paper shows that the forecast accuracy of VSS-LMS is higher than NLMS, and validates the practicability and good effects of this method.Thirdly, The system introduces ANN model and the BP neural network model, and studies principle relationship between Volterra series and the BP neural network model. This paper proposes to apply a single output of three layers of BP neural network to identify the estimation algorithm of Volterra series nucleus in order to realize the multistep forecast of ship movements, and proves that the forecast accuracy of this method excels the estimation of the related algorithm of LMS to identify Volterra series nucleus.Fourthly, through combining GA whole-searching optimization and the BP neural network local-searching optimization, the paper solves the problem of BP neural network in the training process into the local minimum value, and puts forward the threshold value and single figure of GA optimizing BP weights of the neural network to output three layers of BP neural network to identify Volterra series nucleus estimation algorithm. The GA optimizing BP neural network gains the best weights of the threshold value and single figure, based on Taylor progression to decompose to obtain Volterra series nucleus in all levels, and realizes the multi-step ship motions prediction.Finally, with the part of forecast method of ship motions of Kalman Volterra series nucleus identification, the paper analyzes the Kalman filter principle and state estimation, then it focuses on the study of the Kalman identifying the estimation algorithm of Volterra series nucleus, lodges ship motion forecast method of the Kalman identifying Volterra series nucleus, and realizes multi-step ship motion prediction.This paper raises the adaptive algorithm to identify Volterra series nucleus, through the modeling simulation of ship motion prediction, theoretically proves the feasibility and validity of methods, and provides the theory basis of real-time online prediction.