| 학회 | 한국화학공학회 |
| 학술대회 | 2017년 봄 (04/26 ~ 04/28, ICC 제주) |
| 권호 | 23권 1호, p.1271 |
| 발표분야 | 화학공정안전 |
| 제목 | Automating of the HAZOP Analysis via Reinforcement Learning |
| 초록 | Hazard and Operability (HAZOP) analysis is one of most widely used methods of Process Hazard Analysis(PHA). Since process safety is becoming a critical issue due to numerous accidents occuring within plants, the necessity of a thorough and detailed HAZOP analysis is increasing evermore. Various efforts have been made to build HAZOP expert systems that could assist in this repetitive process. Many systems were over the last 20 years and have managed to aid the HAZOP analysis for the routine processes, such as defining the possible causes and consequences of a parameter deviation. Also, some tools applied the concept of ontology and case-based reasoning to speed up the non-routine processes where related database exist. However, these systems are not strictly speaking "automated", since human intervention is necessary to make decisions and select the appropriate add-ons and recommendations provided by the expert system. In this study, a novel framework to automate the HAZOP analysis process using reinforcement learning methods is proposed. A deep Q-learning network is used to formulate the process into a MDP, and optimized for different cases. |
| 저자 | 김창수, 이영근, 김경수, 박건희, 한종훈 |
| 소속 | 서울대 |
| 키워드 | 화학공정안전 |
| 원문파일 | 초록 보기 |
