Multi-dimensional ship motion attitude combination prediction based on the SSA-BiGRU-Attention model

Ship motion attitude prediction is a key link in improving ship navigation safety and maritime operation efficiency. However, ship motion has non-stationary and nonlinear characteristics, and is affected by a variety of complex factors, resulting in low accuracy of existing prediction methods, which is difficult to meet actual needs. To address this problem, this paper proposes an integrated model (SSA-BiGRU-Attention) that combines the Sparrow Search Algorithm (SSA), the Bidirectional Gated Recurrent Unit (BiGRU) and the improved Attention Mechanism. After comparing and analyzing multiple models, the BiGRU model is selected as the basis for constructing an integrated prediction model. In the SSA-BiGRU-Attention model, the BiGRU network captures the basic characteristics of ship motion time series. Then, an improved attention mechanism is used to analyze the interaction between different degrees of freedom and time series data, which can assign higher weights to important features while reducing the interference of less important features on the prediction results. In addition, the sparrow search algorithm is used to optimize the hyperparameters of the integrated model to improve the efficiency of model parameter settings. The experimental results present that the minimum Root Mean Square Errors (RMSEs) of the integrated model for predicting roll, pitch, and heave can reach 0.06411, 0.05434, and 0.01159, which are reduced by 32 %, 18 %, and 10 %, respectively, compared with the single model.