Ship motion prediction based on ConvLSTM and XGBoost variable weight combination model

The marine environment changes complexly, and the navigation of hovercraft is easily affected by waves and currents, etc. Due to the randomness and non-stationary nature of the ship motion, it is difficult to obtain the temporal and spatial characteristics of the ship movement with traditional ship motion prediction methods, resulting in low prediction accuracy. In this paper, a ConvLSTM-XGBoost variable weight combination model is proposed to extract the temporal and spatial characteristics of the ship motion, and the motion of the ship is predicted in one step. First, the genetic algorithm is used to optimize the nonlinear function offline, so as to obtain the optimal weights of the fitted values of the ConvLSTM and XGBoost models on the training set. Then, the k-nearest neighbour algorithm is used to find the corresponding weights of the predicted values of the two models and combine them into the final predicted value. Taking the roll and pitch in the real ship motion data as the the data set. Compared with the single model ConvLSTM and XGBoost, the roll and pitch predictions of the ConvLSTM-XGBoost variable weight combination model are significantly reduced on both RMSE and MAPE. The experimental results verified the effectiveness of the proposed algorithm.