자율운항선박 지원을 위한 실시간 관측 기반의 해양환경 인공지능 예측 기술 검증 |
엄대용1, 서의성2, 전형섭3, 이방희4 |
1㈜올포랜드 해양사업그룹 부장 2㈜올포랜드 해양사업그룹 수석 3㈜올포랜드 해양사업그룹 부사장 4㈜올포랜드 해양사업그룹 전무 |
Verification of Artificial Intelligence Prediction Technology of Marine Environment Based on Real-Time Observation for Maritime Autonomous Surface Ship |
Dae-Yong Eom1, Eui-Sung Seo2, Hyung-Seop Jeon3, Bang-Hee Lee4 |
1General Manager, Ocean Business Division, ALLforLAND.CO.LTD 2Senior Manager, Ocean Business Division, ALLforLAND.CO.LTD 3Vice President, Ocean Business Division, ALLforLAND.CO.LTD 4Senior Director, Ocean Business Division, ALLforLAND.CO.LTD |
Correspondence:
Bang-Hee Lee, Tel: 02)855-5724, Email: dyeatmos@all4land.com |
Received: 27 March 2024 • Revised: 23 April 2024 • Accepted: 24 September 2024 |
Abstract |
For smart ships, such as autonomous vessels, technology that can predict and observe the state of the marine environment along the route in real-time is essential. This requires observational data from ships. The capability to generate marine environmental information for effective decision-making is crucial. Typically, the information that can be collected in real-time under limited equipment conditions on ships includes wave and offshore wind data. In this study, we developed an algorithm that produces marine environment prediction information for subsequent time periods based solely on time series observation data from the route. This prediction information integrates four types of artificial intelligence algorithms (ANN, RNN, Conv-LSTM, and GAN) and two learning structure methods (ultra-short-term - discontinuous and long-term - continuous). The algorithm was compared and analyzed by season using one year of actual ship observation data from 'Meteorological No. 1,' operated by the Korea Meteorological Administration. Among the supervised learning models, RNN demonstrated the best performance in the ultra-short-term and discontinuous learning structure. While unsupervised learning models exhibited lower performance, all models achieved high accuracy in the long-term continuous learning structure across all seasons. In terms of prediction accuracy by algorithm based on learning structure, the long-term continuous learning structure outperformed the short-term continuous learning structure in seasonal prediction accuracy for both waves and sea surface winds. This study confirms the significance and potential of technology for providing data-based marine environment prediction information within autonomous ships. In the future, this technology is expected to facilitate the generation of more practical decision-making information, rather than simply linking it to next-generation waterway products, ship mobility, and route estimation. |
Key Words:
Maritime Autonomous Surface Ship(MASS), artificial intelligence, marine environment, prediction technology, validation |
|