Multi-objective route selection for ice-class vessels using reinforcement learning and graph-based approaches
Route selection for ships in ice is a complicated problem in marine navigation. The navigators have to optimize many economic and environmental factors of the routes while adhering to all maritime regulations to ensure safety. The International Maritime Organization has introduced the Polar Operatio...
| Main Author: | |
|---|---|
| Format: | Text |
| Language: | English |
| Published: |
Memorial University of Newfoundland
2021
|
| Subjects: | |
| Online Access: | https://dx.doi.org/10.48336/e0v4-5x37 https://research.library.mun.ca/15235/ |
| Summary: | Route selection for ships in ice is a complicated problem in marine navigation. The navigators have to optimize many economic and environmental factors of the routes while adhering to all maritime regulations to ensure safety. The International Maritime Organization has introduced the Polar Operational Limit Assessment Risk Indexing System (POLARIS) as guidelines for all vessels operating in the Arctic Ocean. This research investigates a framework for finding an optimal route for different ice-class vessels using two methods: graph-based approaches and reinforcement learning. The system uses ice charts from the Canadian Ice Service to explore possible routes in a grid world. Reward and cost functions are formulated to achieve operational objectives, such as optimizing the distance travelled, voyage time, and fuel consumption while complying with POLARIS regulation. The graph-based method surpasses the Q-learning in deterministic cases. Despite the shortcoming of not handling the non-deterministic environment, it also shows similar routes compared to Q-learning in a stochastic context. The trial results show that the framework provides a means to identify an optimal route for vessels navigating through ice-covered waters. |
|---|