【作者】 迟冬南；

【导师】 严浙平；

【作者基本信息】 哈尔滨工程大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 UUV水下定位与控制是其进行未知环境探测的两个主要问题，本文主要针对其控制问题开展相关研究，而巡岸控制问题是面向环境轮廓构建任务的关键所在。UUV领域中通常采用多元测距声纳作为巡岸控制的信息采集手段，所获取数据的置信度较低。且UUV作业时“人不在环”，对环境的认知及理解要求更准确，尤其在未知环境中需要感知与巡岸控制同步进行，探测能力与控制要求之间存在一定矛盾。因此，需要结合测距声纳与UUV自身特性开展声纳数据的预处理、基于预处理数据的未知环境轮廓构建、局部环境特征约束下的巡岸路径平滑与决策、以及巡岸控制等问题研究，论文主要研究内容简述如下：一、针对多元测距声纳在复杂水下环境中获取数据的置信度较低，而无法满足UUV巡岸控制过程中对理想路径的实时需求问题，采用小波变换方法对声纳数据进行预处理。由于数据集合中异常数据相对于正常数据来说表现出高频特性，因此考虑利用小波变换对高频数据进行局部放大，以便于准确剔除异常数据。由于预处理数据用于构建环境轮廓线，而轮廓线作为UUV巡岸控制理想路径的参考信息，必须保证理想路径的连续性；因此对异常点出现时刻的小波系数进行估计，用于补充异常数据所表征的特征点。二、为了利用预处理后的数据集合作为数据源完成未知环境轮廓构建任务，需要将表征环境轮廓的数据聚集到某一数据类轮廓内，同时，将异常数据划归于轮廓线外。由于支持向量聚类算法能够在剔除异常点的同时完成数据类的聚合，且满足巡岸过程的实时性要求，符合轮廓线构建的前提条件并能够达到获取类轮廓的目的，由此引入该方法完成轮廓线构建过程。另外，为了对初次判定为异常数据的点进行确认，提出惰性算法改进支持向量聚类方法，结合数据的置信度和多普勒速度仪的信息，实现基于多元测距声纳数据的环境轮廓构建任务。三、根据UUV自身特性及控制器设计需求，构建UUV巡岸过程中的平滑理想路径，并结合环境的局部特征进行自主决策。一方面，为了满足UUV在水下环境中的惯性特点对理想路径平滑性的限制，利用Bézier曲线对支持向量聚类惰性算法所构建的、直线顺次相连的轮廓线进行平滑拟合。另一方面，面向未知环境的自主探测任务，UUV必须在构建轮廓线的同时完成巡岸跟踪，因此结合UUV自身状态进行局部环境约束下的自主决策方法研究。四、考虑到未知环境探测过程中的UUV巡岸控制问题，采用反馈线性化控制方法予以实现。依据UUV所面向的任务需求，将探测过程中的路径跟踪划分为两个阶段：对已知路径的精确路径跟踪和对未知路径的预测巡岸控制。反馈线性化方法在跟踪问题中的应用可划分为微分几何反馈线性化方法和稳定逆算法，恰好符合探测过程中两个跟踪过程的条件，因此，在两个跟踪阶段分别采用反馈线性化的两种方法予以实现。在精确跟踪过程中，利用滚动路径发生器导引以任意状态（包括航向和位置）布放的UUV平滑地驶入预设路径，并采用微分几何反馈线性化实现精确跟踪控制。在预测巡岸控制过程中，UUV将构建的环境轮廓作为跟踪的参考路径，利用稳定逆方法通过求解内动态的有界解得到稳定的逆输入，实现预测巡岸控制。为了进一步提高巡岸控制的实时性，利用有限时间窗内的理想路径作为算法的输入，并对时间窗参数进行优化，以保证探测任务与跟踪控制的同步性。为了验证文中所提出的轮廓构建方法及巡岸控制方法的有效性，设计仿真实验及基于海试实测数据的实验。通过设计算法验证流程，建立声纳模型及环境背景模型，提出评价标准，对算法的可行性进行分析。通过设计案例仿真及海试实测数据回演方法对所提出的算法予以验证及分析，说明轮廓构建与同步巡岸控制方法的可行性。更多还原

【Abstract】 Two main issues for unknown environment detection using Underwater UnmannedVehicle (UUV) is orientation underwater and the control method. In this paper, aSimultaneous Detection and Patrolling (SDAP) control algorithm is proposed. Multi rangesonars are effective equipment for collecting information underwater, but the confidencecannot satisfy the control requirement, therefore, weak reliability for range sonar data isrevealed obviously. Unmanned characteristic during UUV implementing mission requirecognition and understanding ability more accuracy, especially synchronism betweenperception and patrolling exists mutual exclusion. Therefore, considering weak reliabilitysonar data and UUV characteristic, pretreatment for sonar data, contour reconstruction forunknown environment, path smoothness, decision under local environment and patrollingcontrol are studied as follows.1. To solve the problem that low confidence of multi range sonar data collected inunderwater complex environment cannot satisfy path requirement of patrolling control,wavelet transform is introduced to pretreatment sonar data. Compared to normal data in theset, singular data is regards as high frequency. Local amplify ability of wavelet transform leadto its introduction for data pretreatment. Owing to environment contour constructed usingpretreatment data is treated to be the reference information for patrolling path; its successionmust be guaranteed. To complement data denoted by singular point, the estimation value atcorresponding time is calculated.2. To achieve contour through clustering data with same property on the basis ofpretreatment data and eliminating the singular data out of data bound. Considering supportvector algorithm can implement the function above and satisfy real time requirement forapplication to be reference for patrolling control, it is introduced to realize reconstructionprocess. In addition, to ensure the property of singular data detected at first time, inertiaalgorithm is proposed to improve SVC method. Combined data confidence and DopplerVelocity Lopper (DVL) information, unknown environment contour based on weak reliabilitymulti range sonar data can be reconstructed.3. According to the characteristic of UUV and the requirement for controller design,smoothness desired path for patrolling control is constructed and automatic decision withlocal environment characteristic is present. On the one side, different orders’ Bézier curves areintroduced to fit contour assembled by lines using SVC, to satisfy inertia character for vehicle underwater. On the other side, towards exploration automatically, contour reconstruction andpatrolling must evolve simultaneously, therefore, it is intuitive that automatic decision methodbase on vehicle state and local environment is studied.4. Feedback linearize algorithm is utilized to design controllers for unknownenvironment detection. According to mission requirements, it is separated to two stages:precise following with known path and patrolling control. Two different algorithms offeedback linearize, differential geometry feedback linearize and stable inversion, are justcorresponding to two control stage requirements. In this paper, on one side, a rolling pathgenerator is present to guide vehicle to follow preset path with arbitrarily initial state,including orientation and position, and differential geometry feedback linearize method isapplied to control vehicle following precisely. On the other side, stable inversion method withcontour as desired path is introduced to obtain stable input of vehicle in patrolling process.Considering patrolling in real time, it is proposed that desired path in finite time window isregards as input items of the algorithm, and the parameter is optimized to guarantee thesynchronism between detection mission and following control.To verify the effectiveness of contour reconstruction method and patrolling controlalgorithm, simulation and experiment using sea trial data are designed. Validation flows areclarified in detail. Sonar model and the background environment models are established.Evaluate criterions and analysis method for reconstruction and patrolling control algorithmsare present respectively to complete the analysis. The feasibilities of reconstruction methodand patrolling control algorithm are verified through simulation and experiment using realdata from sea trial.更多还原