Construction of a Machine Learning Model for Intelligent Ship Collision Avoidance in Intelligent Shipping Systems
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
To address the challenges of high uncertainty in dynamic environments, inadequate multi-source data fusion, and limited real-time decision-making in intelligent shipping systems, this paper proposes a multimodal fusion dynamic collision avoidance decision-making model (AMF-DCDM) based on an attention mechanism. This model implements end-to-end collision avoidance decision-making through a three-layer architecture: the multi-source data preprocessing layer employs Kalman filtering and adaptive median filtering to suppress noise and align spatiotemporal data from AIS, radar, and meteorological data. The cross-modal attention fusion unit dynamically assigns weights to multi-source features using a dual intra-modal and inter-modal attention mechanism to address feature redundancy caused by fixed weights. The dynamic decision-making layer combines STConv and GRU to capture spatiotemporal correlations among ships, embeds COLREGs rule constraints to ensure decision compliance, and outputs a heading adjustment angle and velocity ratio. Experiments using real Bohai Bay AIS data (50GB, 1200 ships) and simulated extreme weather/multi-ship intersection data were conducted within the PyTorch framework, comparing the model with support vector machines (SVM), CNN-LSTM, and Transformer models. Results show that in a multi-ship convergence scenario (8 ships), the AMF-DCDM collision avoidance success rate reached 98.2%, a 3.5 percentage point improvement over the Transformer, with a decision latency of 12.3ms, a 28.7% reduction over the CNN-LSTM. In an extreme weather scenario (20 m/s wind speed), the trajectory deviation was 0.95 m, a 36.7% reduction over the Transformer, and the rule compliance rate was 98.7%. This model meets IMO emergency collision avoidance requirements and provides technical support for the safe operation of intelligent shipping.
Fields of Study & Subjects
Maritime Navigation and Safety
Air Traffic Management and Optimization
Autonomous Vehicle Technology and Safety
Publication details
- Journal / source
- IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
- Pages
- 1383–1388
- Publication date
- 28 January 2026
- Language
- EN