【作者】 罗旭；

【导师】 柴利；

【作者基本信息】 武汉科技大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 污染监测与污染源定位对环境保护有重要意义。无线传感器网络具有节点分布密集、多节点协同工作、成本相对低廉、监测范围广、地理位置限制小等诸多优点，其在污染源监测与定位中的应用越来越广泛，是遥感探测、移动机器人探测、人工探测等常用方法的重要补充。本文研究了多污染对象监测网络覆盖问题、不同地形条件下的水污染源定位问题和移动的污染源追踪问题。主要内容如下：1.提出并解决了多种污染源监测情形下无线传感器网络多重覆盖问题。针对节点随机分布的无线传感器网络，提出一种平均子网寿命模型，在给定成本预算与各子对象的基本覆盖率需求下，给出一种基于整数向量规划的多目标多重覆盖算法。首先求取监测不同子对象的传感器数量，然后基于平均子网寿命模型，求取不同类型的异构节点数量。针对向量规划问题，文中给出两种不同次优解法。在仿真实验部分，将不同次优解法进行了对比，并分析了算法计算复杂度。仿真示例验证了所提出的多重覆盖算法的有效性。2.提出了具有较高稳健性的SL-n (Samples for Localization-n nodes)估计算法。在SL-n算法中，先将目标源周围所有观测节点以n个节点为一组进行分组，通过局部最小二乘方法求出每组观测值的估计样本，然后基于样本集数据估计源位置。运用该算法解决了无边界静态水体中的污染源定位问题。在仿真实验部分，对比了最小二乘算法，SL-n算法，最小一乘算法随观测点个数，观测误差变化的定位性能，验证了SL-n算法的优越性。3.分析了静态水体中边界影响下的近岸污染源扩散过程，提出了一种分段浓度模型，结合不同应用需求分别应用非线性最小二乘方法与无迹卡尔曼滤波方法进行参数估计问题建模，给出了通用模型法、近似函数法、基于无迹卡尔曼滤波(UKF)的估计法三种近岸污染源定位算法。在仿真实验部分，对近岸污染源扩散过程进行了水文模拟，根据MODFLOW模拟数据对比了不同算法的性能。结果表明三种算法各有优势，通用模型法可快速获取目标源相关信息，近似函数法有更稳健的参数估计性能，基于无迹卡尔曼滤波的估计方法可权衡数值计算复杂度与估计性能。4.分析了移动的污染扩散源的扩散过程，给出一种离散化浓度场模型。将连续线源目标追踪问题转化为离散点源目标追踪的次优问题，提出了一种离散化移动扩散源追踪算法。先采用约束最小二乘方法估计目标实时位置，到达时间等相关参数，并进一步采用仅针对位置序列的Sage-Husa卡尔曼滤波方法优化初始位置估计。该算法克服了一般基于动态序列的追踪方法无法直接应用于离散移动扩散源追踪问题的不足。在仿真实验中，分别在匀速率平滑曲线运动与变速非平滑曲线运动的情形下进行追踪实验，分析了追踪精度与采样间隔以及观测节点密度的关系。仿真结果说明了本文移动扩散源追踪方法的有效性。更多还原

【Abstract】 The reasearch of pollution monitoring and pollution source localization has importantsignificance to environmental protection. The sensor networks is based on collaborate effort of alarge number of low-cost nodes, which are densely deployed, and can monitor large-scale area.The deployment is not restricted by geographical environment. Because of these advantages,wireless sensor network has been widely used in pollution monitoring and pollution sourcelocalization, making up the shortcomings of the traditional methods, such as remote sensing,mobile robot detection and artificial detection. This disertation study some problems, includingthe coverage problem in the multiple pollution objects monitoring applications, the waterpollution source localization problems under different terrain conditions and the mobile pollutionsource tracking problem. The main research contents of this paper are as follows:1. The multi-coverage problem in the multi-objects source monitoring applications isproposed and solved. In sensor networks with random distributed nodes, an average subnetlifetime model is proposed. Given the constraints of cost budget and area coverage of differentobjects, a multi-objective multi-coverage algorithm based on integer vector programming isproposed. The first step is to compute the number of each type of sensors used to monitor onesubobject, and the second step is to determine the number of different kinds of heterogeneousnodes based on the average subnet life model. To solve the proposed vector programming issues,two suboptimal methods are given. In the simulation experiments, different suboptimal methodsare compared, and the computational complexity of the proposed algorithm is analysed.Simulation examples verify the effectiveness of the proposed algorithm.2. The SL-n (Samples for Localization-n nodes) algorithm is proposed to obtain robustposition estimation. In the algorithm, all samples using partial least square from everycombination of n observing nodes were obtained first, then the location of the source wereestimated based on the samples. To solve the localization problem of the diffusion source whichis in static water without boundaries, the SL-n algorithm is applied. In the simulation part, thelocalization errors varying with the number of observing nodes and different observation errorswere compared between least squares, SL-n and least absolute localization methods. Thesimulation results prove the advantages of the SL-n algorithm.3. Offshore plume source diffusion in static water is analysed and a piecewise concentrationmodel is put forward. Combining with different application requirements, nonlinear least squaresmethod and Unscented Kalman Filter(UKF) method were applied to the localization problemmodeling of the pollution source which nears the impervious boundary in static water. Threedifferent algorithms called the general localization model algorithm, the approximate functionalgorithm and the algorithm based on (UKF) are proposed. In the simulation part, thehydrological simulation of source diffusion exhibits the diffusion processes. Different parametersestimation algorithms are compared based on the MODFLOW simulation data. The results illustrate that, through the general localization model algorithm, source parameters can beacquired promptly, the approximate function algorithm is more robust in parameters estimatingcompared to the general localization model algorithm，and the algorithm based on UKF balancesthe computation complexity and the estimation accuracy well.4. The diffusion process of mobile diffusion source is analyzed, and a discretizationconcentration model is proposed. The continuous line diffusion source trajectory estimationproblem is transformed into the suboptimal problem which is tracking the moving diffusion pointsource. A mobile diffusion source tracking algorithm based on the discretization concentrationmodel has been proposed. In the algorithm, a constrained least square method is adopted toestimate the related parameters including initial positions and arrival times first. And then, theSage-Husa kalman filter method is used to obtain the optimal estimation of the target positions inreal time. The algorithm overcomes the shortcoming that the general tracking methods based ondynamic sequence can not be applied to the discrete mobile diffusion source tracking directly.The simulation experiments are carried out in two scenarios, in one of which the target movesalong a smooth curve with a constant rate, and the other a non-smooth curve with varyingvelocity. In the simulations, the tracking effects varying with sampling density and nodes densityare also studied. The results illustrate the effectiveness of the proposed algorithm.更多还原