Collision risk assessment for complex navigable waters based on ship dynamic cluster

To address the challenges of insufficient spatial-temporal granularity and limited real-time dynamics in perceiving collision risk within complex navigable waters, a collision risk assessment framework based on ship dynamic clusters was developed. After preprocessing of ship trajectory data, a ship traffic topology network model for complex navigable waters was constructed to depict the interplay of ship collision risk within the region. By taking into account the dynamic spatial-temporal characteristics of ships, the spectral clustering algorithm was introduced for traffic segmentation, defining ship dynamic clusters. Further optimization of the ship dynamic cluster partitioning strategy was performed through the Davies-Bouldin index. Based on the graph theory, measures for ship interference complexity were introduced to establish a ship dynamic cluster risk assessment method, discerning high-risk collision areas within the complex navigable waters. A study was conducted on a representative complex navigable waters, the core port area of Ningbo-Zhoushan Port, where one-month ship traffic flow data was collected to assess ship collision risk under multi-ship encounter scenarios. Research results show that the proposed adaptive method for ship dynamic cluster partitioning effectively divides ships within navigable waters into several clusters. This approach simplifies ship-to-ship collision risk relationship, while retaining the most critical collision risk across the entire spatial scope. The proposed ship collision risk assesment method preserves 84.98% of the global risk. Compared to traditional distance-based clustering algorithm and other classical clustering algorithms such as K-means, hierarchical clustering, and density-based spatial clustering of applications with noise (DBSCAN), the proposed method has a larger retained risk value and can reduce unnecessary risk relationships. It achieves optimal comprehensive performance in terms of accuracy, cluster compactness, and scale balance, effectively reducing false positive rates and discretizing the overall regional risk. By further utilizing the proposed risk assessment model, the risk states of 8 640 multi-ship encounter scenarios are computed through superimposition, identifying high-risk areas within the port area and align with actual survey results.