Identification of ship motion model based on IPPSA-MTCN-MHSA

An accurate ship motion model is crucial for predicting ship attitude, preventing accidents, and facilitating autonomous navigation systems . This paper treats ship motion modeling as a sequence modeling task and introduces a hybrid prediction model, termed MTCN-MHSA. The model integrates a Multi-Channel Temporal Convolutional Network (MTCN) and a Multi-Head Self-Attention Mechanism (MHSA). The MTCN , comprised of parallel TCN channels and a Bi-LSTM, extracts and fuses multi-dimensional features from the sequence. The MHSA mechanism is incorporated to minimize feature loss during information transmission, capture dependencies within the sequence, and enhance the model’s expressiveness and generalization capability. To optimize the hyperparameters of the MTCN-MHSA model, an Improved Positional PID-based Search Algorithm (IPPSA) is proposed, which builds upon the PID-based Search Algorithm (PSA). IPPSA exhibits superior global optimization capabilities and convergence speed, effectively identifying the optimal hyperparameters. Extensive comparative modeling experiments utilizing KVLCC2 and KCS ship maneuvering data validate the notable effectiveness and superiority of the proposed IPPSA-MTCN-MHSA model in ship motion prediction. • A deep learning-based ship modeling strategy is proposed, thus eliminating the need for extensive prior knowledge. • A framework for automatic hyperparameter adjustment in deep learning is proposed. • A high-performance intelligent optimization algorithm inspired by the positional PID control algorithm is proposed. • The proposed model demonstrates generalization capabilities for ship maneuvering across various water domains. • The optimized model achieves universal applicability across different ship types. • The proposed innovative neural network model improves modeling performance by up to 82.63%.