| dc.description.abstract | In all the hazardous accidents at sea, ship collision is always the most frequent one. It has been paid great attention, and International Regu- lations for Preventing Collisions at Sea (COLREG) are being complied
with. With the development of technology, autonomous guidance and navigation for ship collision avoidance have been developed and used in practice for several years. However, there are still challenges when introducing autonomous navigation with the compliance of COLREG in consideration of human interpretation. To narrow this gap between vague and complex human interpretation and precise navigation, a non- linear system of fuzzy logic is introduced to represent the human decision under different situations when ship encounters. However, the common approach of designing fuzzy logic verification and validation system of ship collision avoidance is still by utilizing domain expertise and based on assumptions and conjecture, which is good for interpretability and understandability, but is yet imprecise and lacks a certificate. In recent years, proposals have been brought that AIS can be included to have a thorough improvement of the fuzzy logic system, by the means of helping decide significant parameters and derive the shape of fuzzy membership functions, but have not been implemented and tested. To improve the knowledge-based fuzzy logic model to a novel data-driven fuzzy logic model, machine learning methods can be used, such as unsupervised learning and neural networks. One more concern is, it is also possible to not only improve the parameters and functions but integrate the whole system, which includes fuzzifier, If-Then rules, and defuzzier, into one fuzzy black box using fuzzy neural network. | |
