中尺度涡观测任务中自主式水面水下航行器的进化神经网络控制器研究  

Evolving Neural Network Controllers for Autonomous Surface Underwater Vehicles in Mesoscale Eddies Observation Task

作　　者：王琨[1] 郭梦畅 干文浩 宋大雷[1] Wang Kun;Guo Mengchang;Gan Wenhao;Song Dalei(College of Engineering,Ocean University of China,Qingdao 266100,China)

机构地区：[1]中国海洋大学工程学院,山东青岛266100

出　　处：《中国海洋大学学报（自然科学版）》2023年第8期113-122,共10页Periodical of Ocean University of China

基　　金：中央高校基本科研业务费专项(201962010)资助。

摘　　要：针对现有观测手段难以实现中尺度涡目标性强、高机动性、高随机性以及时空同步的高精度观测这一问题,本文采用自主式水面水下航行器(ASUV)这一现场观测的新手段,通过进化神经网络控制器,实现对中尺度涡的高精度实时跟踪,从而进行水下垂直结构的采样观测。同领域内最新的控制算法相比,进化神经网络的方法对中尺度涡跟踪的精度更高,对卫星数据依赖更少。仿真实验结果表明,ASUV在整个跟踪过程中都能够达到1 km附近的跟踪精度。本文给出了中尺度涡跟踪精度的详细量化评价结果,并在仿真环境中控制ASUV对中尺度涡的水下垂直结构进行了采样,为实现中尺度涡现场三维精细化观测奠定了基础。To solve the problem that existing observation methods cannot track highly mobile and randomly moving mesoscale eddies with time-space synchronization and high precision,an autonomous surface underwater vehicle(ASUV)is adopted as a new means of in-situ observation to track the mesoscale eddies with high-precision in real time by evolving neural network controllers,so as to sample their underwater vertical structures.Compared with up-to-date control methods,the proposed evolutionary neural network method can track mesoscale eddies with higher precision and less satellite altimeter data.Experimental results show that ASUV can achieve tracking accuracy near 1 km during the entire tracking process.Extensive quantitative evaluation results are provided for tracking mesoscale eddies in this paper.Furthermore,simulation experiment is conducted for the ASUV to sample the three-dimensional underwater structure of mesocale eddies,which lays the foundation for the in-situ three-dimensional fine observation of mesoscale eddies.

关 键 词：中尺度涡 三维精细化观测自主式水面水下航行器 神经网络控制器 进化神经网络 跟踪精度 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]