SIA OpenIR  > 水下机器人研究室
AUV海洋动态特征自适应测绘方法研究
Alternative TitleResearch on AUV Adaptive Mapping for Dynamic Ocean Features
阎述学1,2
Department水下机器人研究室
Thesis Advisor封锡盛
ClassificationTP242.3
Keyword多auv协同 自适应采样 海洋特征测绘 队形控制
Call NumberTP242.3/Y17/2018
Pages124页
Degree Discipline模式识别与智能系统
Degree Name博士
2018-05-21
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract

针对HABs区域海洋特征的观测需求,本文主要完成以下几点工作:(1) 面向较小区域的观测应用,提出了一种基于高斯过程回归的小型自主水下机器人(AUV)自适应采样方法,该方法可以使AUV通过自身环境观测信息的更新,在线规划路径并完成对未知区域的快速观测。首先,确定使用AUV的合适采样间隔时间;在此基础上,根据AUV实时观测的数据进行高斯过程回归分析,预测未观测区域环境数据,并通过计算预测区域梯度极值和预测不确定度提出分解自适应采样过程为热点区搜寻和热点区脱离两个部分,并提出整合梯度下降法和平均方差引力线方法引导AUV进行在线路径规划。使用该方法,对具有不同特征分布的区域环境观测过程进行仿真,结果显示,本方法与常规观测方法相比,能够更高效的获得观测区域的低误差特征分布估计,更快地获得观测区域热点区特征,对不同特征分布的区域观测有更强的适应性。(2) 针对多AUV观测的需求,对多AUV协同的基本队形控制问题进行研究,在已有的基于领航者的队形控制方法上加入了跟随者状态反馈,并基于状态反馈设计有限状态自动机进行队形控制,通过仿真试验,验证了本方法相对于无反馈系统,队形保持的效果更好。在湖上使用两艘AUV进行了基于无线电和水声通信的多AUV队形控制试验,通过不同距离的编队试验,对基于状态反馈的队形控制算法进行了验证。(3) 针对较大范围的观测区域高效观测的需求,根据当前已有的AUV能力,提出了一种多AUV协同对同一海洋区域进行观测的方法,本方法在整个实现过程中无需人为的介入和监督,单纯依靠预定的海洋观测经验和AUV实际的实时在线海洋观测数据即可完成。在单AUV热点区自适应采样的路径规划方法基础上设计了基于数据交换后的全局估计边界点引力矩阵的AUV观测方向选择方法,通过此方法完成数据交换后的AUV自适应观测目标更新。在此基础上使用基于区块平均引力的观测方向更新替代边界点引力对方法进行改进。比较两种多AUV协同自适应采样和非协同多AUV独立自适应采样以及非自适应多AUV编队采样四种不同方法的采样仿真结果,采用多AUV协同自适应采样方法能够在相同时间获得更好的目标区域测绘效果。本文的创新性和研究亮点主要体现在以下几个方面:(1) 提出一套涵盖单AUV和多AUV的海洋特征自适应采样测绘方法。(2) 通过比较高斯过程回归中使用不同的回归推理方法的估计准确度和计算效率,确定使用真实AUV进行采样观测的合适采样间隔时间;根据回归模型估计的区域观测数据,分别通过计算预测区域梯度极值引导AUV在自适应采样中进行热点区搜寻,通过预测不确定度指导AUV进行热点区脱离,提出通过寻找最大的平均估计方差引力线方法引导AUV进行在线路径规划。 (3) 在基于领航者的队形控制方法上加入了跟随者状态反馈,设计了通用的跟随者状态判别计算矩阵和基于状态反馈的有限状态自动机,据此进行直接快速的状态判别与切换。(4) 使用搭载水声通信的AUV进行了多AUV的编队实航测试,对方法进行了验证。(5) 根据多AUV协同采样的仿真结果,提出基于区块平均引力的观测方向更新替代边界点引力,在保证协同自适应采样相对于其他非协同采样方式既有的低测绘误差的基础上进一步降低观测误差,具有更好的环境分布适应性。

Other Abstract

In view of the observational requirements of HABs regional ocean characteristics, this paper mainly completes the following tasks: (1) For small-area observation applications, an adaptive sampling method based on Gaussian process regression for portable autonomous underwater robots (AUV) is proposed. This method can make AUV use its own environmental observation information to update the online planning path and complete quick observations of unknown areas. First, the appropriate sampling interval time for using AUV is determined; based on this, according to the real-time observation data of AUV and Gaussian process regression analysis, unobserved area environmental data are predicted, and present the decomposition of the adaptive sampling process to hotspot search and hotspot escape by calculating the regional gradient extremes and predictive uncertainty. The integrated gradient descent method and average variance attraction line method are proposed to guide AUV for online path planning. Using this method, the regional environmental observation process with different feature distributions is simulated. The results show that compared with the conventional observation methods, this adaptive method can obtain the low-error feature distribution estimation of the observation area more efficiently and obtain the observation area hot spots more quickly. This method also has greater adaptability to regional observations with different feature distributions. (2) For the needs of multi-AUV observations, the basic formation control problem of multi-AUV coordination is studied. The follower state feedback is added to the existing leader-follower formation control method, and the finite state automaton is designed based on state feedback to keep the formation. It is verified that this method is better than the non-feedback system in maintaining formation through simulation tests. Two AUVs were used to conduct multi-AUV formation control experiments based on radio and acoustic communications in the lake. Formation control algorithms based on state feedback were verified by formation experiments at different distances. (3) To meet the needs of efficient observing in a wide range of observation areas, based on current AUV capabilities, a multi-AUV method for collaborative observation of the same ocean area is proposed. This method does not require human intervention during the entire implementation process. And supervision can be accomplished by simply relying on predetermined marine observation experience and actual real-time online ocean observation data of AUV. Based on the adaptive hotspot sampling path planning method in single AUV, the AUV observation direction selection method based on the globally estimated boundary point attraction matrix after data exchange is designed. Through this method, the AUV adaptive observation target updating after data exchange is completed. Based on this, using the method of updating the observation direction based on block average attraction to replace the boundary point attraction to improve the method. Compare the results of simulations between two kinds of multi-AUV cooperative adaptive sampling and non-cooperative multi-AUV independent adaptive sampling and non-adaptive multi-AUV formation sampling, the multi-AUV cooperative adaptive sampling method can achieve better target area mapping results at the same navigation time. The innovation and research highlights of this thesis are mainly reflected in the following aspects: (1) Propose a set of adaptive sampling and mapping methods for marine features covering single AUV and multiple AUV. (2) By comparing the estimation accuracy and computational efficiency of different regression reasoning methods in Gaussian process regression, determine the appropriate sampling interval time for sampling observations using real AUV, and estimate the regional prediction data by using the regional observation data estimated by the regression model. The gradient extremum guided AUV performs hotspot search in adaptive sampling, guides the AUV to break away from the hotspot by predicting uncertainty, and proposes to guide the AUV for on-line path planning by finding the maximum average estimated variance attraction line method. (3) Follower status feedback was added to the leader-follower formation control method, and a general follower state discriminant calculation matrix and finite state automata based on state feedback were designed. Based on this, direct and fast state discrimination and switching were performed. (4) A multi-AUV formation navigation test was conducted and the formation control method was verified by using two AUVs equipped with underwater acoustic communication. (5) According to the simulation results of multi-AUV sampling and mapping, an observation direction update based on block average attraction is used instead of the boundary point attraction to ensure that collaborative adaptive sampling is based on the low mapping errors existing in other non-cooperative sampling methods. The new method reduces observation errors and have better adaptability to environmental distribution.

Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/21802
Collection水下机器人研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院大学
Recommended Citation
GB/T 7714
阎述学. AUV海洋动态特征自适应测绘方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2018.
Files in This Item:
File Name/Size DocType Version Access License
AUV海洋动态特征自适应测绘方法研究.p(9912KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[阎述学]'s Articles
Baidu academic
Similar articles in Baidu academic
[阎述学]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[阎述学]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.