Deep Reinforcement Learning-Based Observation Station Placement Optimization for Ocean Monitoring
文献类型：期刊论文
| 作者 | Zhao, Meihua1,3; Wang, Haoyu1,3; Wang, Jing1,2,3; Li, Xiaofeng1,2,3 |
| 刊名 | IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
 |
| 出版日期 | 2026 |
| 卷号 | 64页码:13 |
| 关键词 | Optimization Oceans Image reconstruction Vectors Sensor placement Accuracy Deep reinforcement learning Calibration Technological innovation Sea measurements Deep reinforcement learning (DRL) station placement optimization temperature-salinity (T-S) field reconstruction |
| ISSN号 | 0196-2892 |
| DOI | 10.1109/TGRS.2026.3671298 |
| 通讯作者 | Li, Xiaofeng(xiaofeng.li@ieee.org) |
| 英文摘要 | Ocean observation underpins our understanding of marine dynamics. Given the high cost of deploying and maintaining observation stations, optimizing their placement is essential for accurate ocean field reconstruction. Here, we propose RL-SPO, a deep reinforcement learning (DRL)-based framework for station placement optimization. The framework autonomously learns optimal placement strategies through closed-loop interactions with a reconstruction-driven simulation environment, enabling sequential decision-making in complex spatial domains. It integrates three key innovations: an image-based state representation to capture spatial correlations, dynamic action masking to strictly enforce feasibility constraints, and environment-specific reward calibration to balance heterogeneous objectives. Experiments using satellite and reanalysis datasets from the Northwest Pacific Ocean show that RL-SPO consistently outperforms state-of-the-art baselines in reconstructing temperature and salinity fields, and generalizes effectively across single- and multiobjective tasks in both 2-D and 3-D settings. Notably, RL-SPO reveals that optimal observation stations cluster in dynamically active, high-entropy regions such as coastal waters and the Kuroshio Extension, highlighting its ability to identify "informational hotspots" of the ocean system. Overall, RL-SPO provides a generalizable and data-driven paradigm for ocean observational network design, offering an efficient and cost-effective pathway toward intelligent ocean monitoring. |
| WOS关键词 | SENSOR PLACEMENT |
| 资助项目 | Shandong Province Postdoctoral Innovation Program[SDCX-ZG-202503098] ; National Natural Science Foundation of China[92258301] ; National Natural Science Foundation of China[42376175] ; National Natural Science Foundation of China[42221005] ; National Natural Science Foundation of China[42090044] ; China Postdoctoral Science Foundation[2025M780841] ; National Postdoctoral Research Program[GZC20250594] |
| WOS研究方向 | Geochemistry & Geophysics ; Engineering ; Remote Sensing ; Imaging Science & Photographic Technology |
| 语种 | 英语 |
| WOS记录号 | WOS:001717553500002 |
| 出版者 | IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC |
| 源URL | [http://ir.qdio.ac.cn/handle/337002/205127]  |
| 专题 | 中国科学院海洋研究所 |
| 通讯作者 | Li, Xiaofeng |
| 作者单位 | 1.Inst Oceanol, Chinese Acad Sci, Key Lab Ocean Observat & Forecasting, Qingdao 266071, Peoples R China 2.Qingdao Marine Sci & Technol Ctr, Lab Ocean Dynam & Climate, Qingdao 266200, Peoples R China 3.Lab Ocean Circulat & Waves, Qingdao 266071, Peoples R China |
| 推荐引用方式 GB/T 7714 | Zhao, Meihua,Wang, Haoyu,Wang, Jing,et al. Deep Reinforcement Learning-Based Observation Station Placement Optimization for Ocean Monitoring[J]. IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING,2026,64:13. |
| APA | Zhao, Meihua,Wang, Haoyu,Wang, Jing,&Li, Xiaofeng.(2026).Deep Reinforcement Learning-Based Observation Station Placement Optimization for Ocean Monitoring.IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING,64,13. |
| MLA | Zhao, Meihua,et al."Deep Reinforcement Learning-Based Observation Station Placement Optimization for Ocean Monitoring".IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING 64(2026):13. |
入库方式： OAI收割
来源：海洋研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。