Abstract | Sudari na moru predstavljaju jedne od najtežih vrsta nezgoda s obzirom na to da mogu dovesti u opasnost ljude i okoliš te nanijeti veliku ekonomsku štetu. Međusobni odnosi brodova te potrebne radnje kako bi se izbjegli sudari uređeni su pravilima o izbjegavanju sudara na moru. Pravila, nastala u prošlom stoljeću, upućena su članovima plovidbene straže koji donose odluke na temelju vlastitog znanja, iskustva i vještina, imajući na umu sigurnost plovidbe. Međutim, pravila kao takva često koriste pojmove koji su nejasni ili neizraziti te njihova kvantifikacija predstavlja poteškoće razvoju modernih inteligentnih navigacijskih sustava. Neizraziti pojmovi mogu se jednostavno matematički opisati koristeći svojstva neizrazitih skupova i neizrazitih sustava. Budući da je procjena rizika sudara na moru stvar odluke člana plovidbene straže, u ovom se radu pristupilo modeliranju indeksa rizika sudara koristeći ekspertna znanja uz neizraziti pristup. Istraživanje u svom prvom dijelu određuje čimbenike koje eksperti smatraju bitnima pri procjeni rizika sudara. Rezultati ukazuju na to da su eksperti kao vrlo važne čimbenike pored uobičajenih odabrali i neke druge čimbenike koji imaju utjecaj na procjenu rizika sudara. Svi odabrani čimbenici, u svojoj naravi neizraziti, predstavljaju lingvističke varijable, te su se u drugom dijelu istraživanja kvantificirali koristeći teoriju neizrazitih skupova. Pri tome su se rasponi i oblik lingvističkih varijabli odredili koristeći ekspertno znanje ispitanika, upotrebom trapezoidnih i Gaussovih funkcija pripadnosti. Koristeći odabrane čimbenike, pristupilo se izradi modela neizrazitog sustava zaključivanja za određivanje neizrazitog indeksa rizika sudara. Kako bi se prikazale razlike pri procjeni rizika koje proizlaze iz različitih međusobnih odnosa brodova, potrebno je bilo odrediti više podmodela. Dodatno, s obzirom na mišljenje eksperata, bilo je potrebno kreirati modele u uvjetima smanjene vidljivosti, modelirajući tako poseban oprez kako je on propisan pravilima. Validacijom modela na neovisnom skupu uzoraka odabrani su oni koji su koristili trapezoidne funkcije pripadnosti kao povoljniji. Verifikacijom rezultata studijama slučajeva prikazane su mogućnosti modela neizrazitog sustava zaključivanja. Rezultati indiciraju kako je ostvareno unapređenje protusudarnog uzbunjivanja. Među ostalim, postignuto je pravovremeno uzbunjivanje u nekim situacijama u kojima postojeći sustavi ne bi alarmirali člana plovidbene straže na vrijeme. Dodatno prikazani indeks rizika sudara nije binaran, te dopušta usporedbu brodova s obzirom na neizraziti indeks rizika sudara. Na taj način, model pruža prikaz hijerarhije indeksa rizika sudara. Prikazani neizraziti sustav mogao bi se koristiti kao sustav potpore odlučivanju brodovima s ljudskom posadom. Sustav je također moguće primijeniti za upotrebu autonomnim brodovima. Karakteristike neizrazitih sustava dopuštaju u budućim istraživanjima uvesti dodatne ulazne varijable.Dodatno, moguće je još detaljnije prikazati razlike u procjeni rizika koje proizlaze iz međusobnih odnosa brodova koristeći finiju podjelu podmodela u budućim istraživanjima. |
Abstract (english) | Collisions at sea represent one of the most serious types of accidents, considering that they can put people and the environment at risk and cause significant economic damage. The mutual relations of ships and the necessary actions to avoid collisions are governed by the International Regulations for Preventing Collisions at Sea. The rules, created in the last century, are addressed to the members of the navigation watch who make decisions based on their own knowledge, experience and skills, keeping in mind the safety of navigation. However, rules often use unclear or vague concepts, and their quantification presents difficulties for the development of modern intelligent navigation systems. Vague concepts can be easily described mathematically using the properties of fuzzy sets and fuzzy systems. Since the assessment of the risk of collision at sea is a matter of the decision of a member of the navigational watch, in this research, the modelling of the index of collision risk was approached using expert knowledge with a fuzzy approach. In its first part, the research determines the factors that experts consider important when assessing the risk of a collision. The results indicate that the experts choose some other factors that impact the collision risk assessment as very important in the last century, are addressed to the members of the navigation watch who make decisions based on their own knowledge, experience and skills, keeping in mind the safety of navigation. However, rules often use unclear or vague concepts, and their quantification presents difficulties for the development of modern intelligent navigation systems. Vague concepts can be easily described mathematically using the properties of fuzzy sets and fuzzy systems. Since the assessment of the risk of collision at sea is a matter of the decision of a member of the navigational watch, in this research, the modelling of the index of collision risk was approached using expert knowledge with a fuzzy approach. In its first part, the research determines the factors that experts consider important when assessing the risk of a collision. The results indicate that the experts choose some other factors that impact the collision risk assessment as very important in addition to the usual ones. All selected factors, fuzzy in nature, represent linguistic variables and were quantified in the second part of the research using the theory of fuzzy sets. In doing so, the ranges and shapes of the linguistic variables were determined using the respondents' expert knowledge and trapezoidal and Gaussian membership functions. A fuzzy inference system model was created using the selected factors to determine the fuzzy collision risk index. In order to show the differences in risk assessment resulting from different ship relationships, it was necessary to define several sub-models. Additionally, considering the opinion of experts, it was necessary to create models in conditions of reduced visibility, thus modelling due regard as prescribed by the rules. By validating the model on an independent set of samples, those that used trapezoidal membership functions were selected as more favourable. The possibilities of the fuzzy inference system model are shown by verifying the results with case studies. The results indicate that the collision alerting has been improved. Among other things, timely alerting was achieved in some situations where the existing systems would not have alerted the navigation watch member in time. The additionally presented collision risk index is not binary and allows the comparison of ships with regard to the fuzzy collision risk index. In this way, the model provides a representation of the hierarchy of collision risk indices. The presented fuzzy system could be used as a decision support system for manned ships. The system can also be used by autonomous ships. The characteristics of fuzzy systems allow the introduction of additional input variables in future research. Additionally, it is possible to show in more detail the differences in risk assessment arising from ship relationships using a finer subdivision of submodels in future research |