Ship Attitude Prediction Based on VMD-PSO-LSTM

The marine environment is full of complexity, which makes the ship's posture highly stochastic, Traditional prediction models are less effective in predicting ship attitude Therefore, Using variational modal decomposition (VMD) to preprocess ship attitude data can effectively reduce the complexity of ship signals and improve the prediction model with a particle swarm algorithm-optimized long-short time (PSO-LSTM) neural network. First, the ship posture data are decomposed into several intrinsic modal functions (IMFs) as well as residual terms using the VMD algorithm. Then, neural network hyperparameter optimization and model training are performed for Each IMF component and the residual term separately by PSO-LSTM model. Finally, the trained model is used to predict each of these decomposition terms, and the above predicted values are summed to obtain the prediction results. The results show that the root mean square error values of the model are reduced by 0.0407m and 0.0121m, and its mean absolute percentage error is reduced by 27.57% and 24.68% compared to the LSTM and PSO-LSTM models, respectively. The effectiveness of the method in improving the prediction accuracy is demonstrated.