Evaluation Models for Safe Ship Operation through Data Driven Approach using AIS Information

Abstract

Herein, a study was conducted to derive quantitative methods for improving safety using Automatic Identification System (AIS) data. The study involved constructing algorithms using data mining techniques, including clustering analysis and regression analysis, based on a data-driven approach. The aim was to enable ships to operate safely. According to the 5th long-term development plan for the shipping industry, sea transport accounted for 99.8% of Korea's import and export cargo volume. The percentage of small container ships, with a capacity of 8,000 Twenty-foot Equivalent Units (TEU) or less, has been steadily declining, while the proportion of large container ships, exceeding 8,000 TEU, is rapidly increasing. It is expected that in the near future, there will be a rise in mega-sized container ships exceeding 400m in Length overall (LOA), driven by the rapid advancements in this field. The trend towards mega-sizing of ships has a significant impact on the design of port facilities, including the length of berths, the width of fairways, turning points, and construction of bridges across waterways. Additionally, offshore wind farms, a rapidly growing source of renewable energy, have emerged in recent years. It is crucial to ensure that these offshore facilities do not interfere with ship traffic. This study focused on four key factors related to ship operation safety: predicting future ship specifications, analyzing safety navigation width, assessing annual collision probability, and determining the optimal maritime traffic route for offshore wind farms, taking collision frequency into consideration. First, the future main specifications of a 30,000 TEU container ship were predicted using regression and cluster analysis among various data mining algorithms. A dataset of 5,497 container ships, registered with the International Maritime Organization (IMO) and up to 20 years old, was utilized for this analysis. The ships were categorized based on variations in their dimensions using the k-means clustering algorithm. This clustering allowed for the examination of Deadweight tonnage (DWT), TEU capacity, LOA, Length between perpendiculars (Lpp), Breadth (B), and maximum draft (d) of container ships, with a coverage rate of 75%. The results revealed that a container ship with a capacity of 30,000 TEUs is estimated to have an LOA of 428.4m, a B of 67.6m, and a d of 17.0m. Secondly, among the data mining algorithms employed, cluster analysis, dimensionality reduction classification through Principal Component Analysis (PCA), and statistical analysis were utilized to determine the safety navigational width for ships passing under bridges across waterways. This study focused on analyzing traffic distribution characteristics and deriving optimal traffic distribution for four cable-stayed bridges in Korea: Incheon Bridge, Busan Harbor Bridge, Mokpo Bridge, and Machang Bridge. The goodness-of-fit test was employed to achieve this. By assuming a safe passage range of 95% confidence interval and considering both lognormal and normal distributions, the analysis results were obtained. In the case of Incheon Bridge, the largest difference between the normal distribution and the lognormal distribution was observed in the range of 64m to 98m. On the other hand, a minimum difference of 10m was found at Machang Bridge. Consequently, it was determined that presenting the safety navigational width of traffic using the lognormal distribution was more suitable for Incheon Bridge, whereas for the other bridges, similar results could be obtained using either the normal or lognormal distribution due to the similarities in width between the two distributions. Thirdly, an analysis of annual collision probability was conducted from a probabilistic standpoint. In the Ministry of Oceans and Fisheries (MOF), the Maritime Traffic Safety Assessment (MTSA) evaluates the impact of maritime development, including bridges across waterways, on maritime traffic safety. However, the current system employs collision probability per 10,000 vessels as a safety standard based on the number of collisions per vessel, rather than considering the annual collision frequency during collision risk assessments using ship handling simulations. This approach is inadequate for verifying safety in areas with high maritime traffic. To address this issue, this study employed the annual frequency of collapse (AASHTO Method II) to convert collision frequency per ship into an equation representing the annual collision frequency for four bridges across waterways in Korea. A normality test was performed, and passing ships were clustered by tonnage using k-means clustering. The annual collision frequency was then calculated for each cluster. The analysis revealed that an annual collision frequency of once every 50 to 100 years was suitable, depending on the size of the ship. This finding aligns with the annual collision frequency reported by the IMO. Furthermore, the study examined the change in collision probability when a mega-sized ship, with the predicted main dimensions (428.4m x 67.6m) of the 30,000 TEU container ship, is introduced in the future. The analysis demonstrated that the annual collision probability decreased from 103.4 years to 70.9 years. Consequently, it is crucial to ensure the provision of a sufficiently wide main span when designing bridges across waterways, taking into account the potential presence of large ships in the future. Lastly, an analysis was conducted on the optimal route planning method for offshore wind farms from the perspective of collision frequency. The study examined the traffic density in the target area based on 20 months of Big AIS data, both before and after the installation of Floating-LiDAR (Light Detection and Ranging) devices, which are used for measuring wind resources in the sea area near Ulsan Exclusive Economic Zone (EEZ). The analysis revealed that after the installation of Floating-LiDAR, the traffic flow in the area was divided into three distinct flows. Subsequently, traffic distribution was modeled assuming the presence of a floating wind farm, based on the results of the traffic density analysis. Using this modeled traffic distribution, ship-to-ship collisions were analyzed using IWRAP Mk II, and ship-to-offshore wind farm collisions were comprehensively examined using the COWI A/S model. The optimal maritime traffic route was determined by applying the ALARP (As Low As Reasonably Practicable) principle. The results indicated that when establishing a corridor, the ALARP criteria set in this study were met when the corridor width was set at 4,000m, the buffer zone was set at 1,500m, and the outer separation distance was approximately 1,852m (1.0 Nautical Mile). When only detour traffic was allowed without a corridor, the ALARP criteria were satisfied with an outer separation distance of 2,037m (1.1 Nautical Miles) or more. According to the ALARP setting criteria, the proposed optimal maritime traffic method presented an appropriate safety range corresponding to a collision frequency of once every 126 to 128 years. Therefore, it was determined that securing a navigation route of approximately 4 km and a buffer zone of over 1.5 km would satisfy the ALARP criteria set in this study, potentially reducing the detour area. In other words, establishing a corridor was deemed the optimal route from the maritime traffic perspective. However, it should be noted that regardless of the existing navigation method, the installation of an offshore wind farm increases the collision risk by approximately three times compared to the existing traffic flow. To mitigate the frequency of accidents, additional safety operational measures will be necessary. The practical implications of this study can greatly contribute to the safety of ship operations. The methods developed in this study enable quantitative decision-making based on data, rather than relying solely on empirical factors, when it comes to ensuring the safety of ship operations. By adopting these data-driven approaches, we can enhance the safety operations of both port facilities and ships, providing a stronger foundation for overall maritime safety.