The sea trials study of warp tension for trawlers based on dynamic and data-driven models

• A comprehensive towing model was developed using the lumped mass method, incorporating the otter board and fishing net as boundary conditions. • Four deep learning models (CNN, DT, RF, MLP) were trained, validated, and tested using sea trial data to predict trawler warp tension. • A method was proposed for selecting the optimal model based on the speed range for trawlers, combining the dynamic model with deep learning algorithms to create a comprehensive operational database under various sea conditions. • Extensive validation using sea trial data confirms the robustness and reliability of our proposed models and methods. • This study provides new insights and methodologies for optimizing trawl operations, benefiting the field of “Smart Trawlers”. As the importance of intelligence and automation grows across various domains, there is an increasing demand for intelligent applications in trawler fishing. This study begins with the collection of warp tension data through sea trials, followed by the development of a comprehensive towing model using the lumped mass method, incorporating the otter board and fishing net as boundary conditions. Four data-driven models—Convolutional Neural Network (CNN), Multi-Layer Perceptron (MLP), Decision Tree (DT), and Random Forest (RF)—are trained, validated, and tested using the sea trial dataset. These models were chosen to encompass two major categories of learning paradigms: deep learning models (CNN and MLP) and conventional tree-based machine learning models (DT and RF). The results demonstrate that the RF model exhibits superior predictive capability for trawler warp tension. By comparing the results of the data-driven models with the dynamic model, we propose a method for selecting the optimal model based on the speed range for trawlers that combines the dynamic model with data-driven models. This methodology enables the construction of a comprehensive operational database for trawlers under diverse sea conditions, enhancing their capability to assess real-time status and predict unknown sea states. This advancement is expected to significantly enhance the development of intelligent trawlers.