This article presents a method to build a T-S fuzzy model based on Generalized Dynamic Fuzzy Neural Network (GD-FNN) of Elliptical Basis Function in order to solve the problem of ship motion model's uncertainty and nonlinearity. The proposed method needs neither prior fuzzy neural networks structure knowledge nor prior training phase, it can be used to build the nonlinear and uncertain part through online adaptive learning algorithm. The fuzzy rules could be generated and pruned on-line by learning. The ship vertical (heave and pitch) dynamic linear adaptive CARMA model can be got by local dynamic linearization at each sampling point. Then Generalized Predictive Control (GPC) law is deduced by combining adaptive linear model with generalized predictive control. Simulation experiment shows that this algorithm is effective and efficient, its anti-pitching effect reaches 82.2%.