Research on ship safety risk early warning model integrating transfer learning and multi-modal learning
Abstract
An efficient risk warning model is crucial to the navigation safety of ships. However, existing researches are often limited by data acquisition, resulting in model training limited to specific sea areas, and data transferability among different sea areas has not been fully explored. Therefore, this paper studies the feasibility of some artificial intelligence methods in risk warning and the transferability of pre-trained models among different regions based on sizeable historical weather data sets, thus proposing a ship risk early warning model integrating transfer learning and multi-modal learning. This model integrates ship traffic flow data, weather data, ship accident data, ship parameters and other related data, and uses machine learning algorithms to pre-train the model to analyze possible risks to ship navigation safety driven by high-dimensional data. And discuss the generalization and transferability of the model based on transfer learning methods. Research results show that the ship navigation risk early warning model that integrates transfer learning and multi-modal learning can effectively identify ship navigation risks and has high accuracy in some transfer sea areas. This innovation has important practical significance in ship safety, improving global ship safety levels and providing more comprehensive and reliable decision-making support for risk management in the shipping industry.