【作者】 陈云翔；

【导师】 汤兵勇；

【作者基本信息】 东华大学 ， 管理科学与工程， 2008， 博士

【摘要】 在全球洪涝灾害日益加重,非工程防洪倍受重视的背景下,开展防洪决策中洪水灾情预测与评估系统的研究,具有重要的理论意义与广泛的实际价值。本论文结合当前我国灾情预测与评估工作的实际需求,对洪水灾情预测与评估系统的功能结构与模型进行了研究与设计,并运用计算机技术、人工智能技术、可视化仿真技术以及数学、管理科学、决策科学等多学科知识,建立相关的洪水灾情预测与评估模型,从而为实现洪水灾情预测与评估的科学化、系统化和定量化奠定了一定的基础。论文的主要贡献有:(1)提出了较为系统全面的防洪决策支持系统的三层体系架构,即接口层、应用层和基础层,并对各层分别进行了详细的设计。其中,基础层包括数据库(存放空间数据和属性数据)、模型库以及知识库,为系统提供数据和技术支持;接口层实现系统与决策分析人员的交互;应用层通过调用数据和模型,完成防洪决策过程各个环节的信息需求和分析功能,是系统的核心部分。然后对各功能子系统进行了设计。(2)建立了基于遗传算法(GA)的BP神经网络洪峰预测模型。针对BP神经网络在洪水预测中的不足之处,利用GA对BP网络预测模型进行改进,将GA计算步骤与网络的优化过程结合起来,形成了新的GA-BP网络预测模型。利用实际数据,采用BP网络模型和GA-BP网络模型分别进行了模拟预测和结果分析。结果表明,采用GA-BP网络的洪水灾情预测模型能够提高预测效率和准确度。(3)提出了洪水灾情预测系统中地形与洪水淹没模拟的算法与模型。为了完成淹没过程的可视化仿真,首先基于格网法进行三维地形模拟,然后提出了一种基于蝶形细分模式的改进的自适应算法,通过结合蝶形算法和多阶曲线插值的思想计算新的插值点,在细分的过程中有选择性地对三角形单元网格进行细分,来模拟三维地形,它可作为洪水实时呈现系统的基础结构。接着从DEM出发,采用改进的格网源点蔓延算法,模拟洪水演进过程,分别计算出给定洪水水位或洪量条件下淹没区的淹没范围和水深分布;最后运用Java 3D网络可视化工具绘制淹没范围、水深分布等洪水风险图,并对不同条件下的洪水淹没过程进行动画演示,从而为实现洪涝灾害损失评估提供准确的灾情数据。(4)研究了防洪决策中洪水灾情快速评估算法。针对洪水灾情不同阶段的评估需求,根据洪水灾情的评估指标,建立了灰色关联度分析、BP网络以及RBF网络三种洪水灾情快速评估模型;其中,RBF网络是在基本RBF网络上做了改进,将原来的线性层改为竞争层,使得模型能够更好地完成模式判断工作。经过实例分析,三种模型都能得到较好的评估结果,尤其是灰色关联度分析模型和RBF网络模型在评估速度以及稳定性方面有出色的表现,所以这两个模型可以应用于洪水灾情快速评估系统中,其运算结果可以为防洪决策提供科学的依据。(5)提出了基于免疫遗传算法(IGA)的神经网络洪水灾情评估模型。针对遗传算法暴露出的局部搜索能力不强、未成熟收敛等缺陷,提出了一种基于IGA的神经网络模型(IGA-NN)用于洪涝灾度评价,并给出了IGA-NN的设计方法和优化步骤。为了体现IGA-NN的优越性,还将IGA-NN与标准遗传算法(SGA)优化的神经网络模型进行了比较。实验结果表明,IGA具有更好的全局和局部寻优能力,且可以改善SGA未成熟收敛的缺陷。IGA-NN能够反映多个评价指标与灾度等级之间的相关关系,可用于洪涝灾度评价,且具有更快的应答速度和更高的评估精度。(6)进行了防洪DSS中洪水灾情预测与评估系统的设计与实现。通过洪水灾情预测与评估系统功能的需求分析,设计了防洪DSS体系结构下的洪水灾情预测与评估系统。根据洪水灾害发展的不同阶段,将灾情评估系统分成了灾前预测、灾中决策和灾后评估三大主要功能模块。探讨了灾情预测与评估系统在防洪DSS中的重要作用,并在此基础上确定了灾情预测与评估系统的功能结构。另外,还进行了各子系统之间的接口设计。最后分别对灾情预测与评估系统中的三个功能模块进行了实现。最后,对全文的工作进行了总结,指出了今后需要进一步研究的方向。更多还原

【Abstract】 Currently,with the aggravation of global floods disaster,non-structural measures against floods have drawn more attention.Under this background,it has practical sense to develop a flood disaster evaluation system in flood control decision-making.With the analysis of demands of current disaster evaluation we designed a function framework for the evaluation system.Also, computer technology,artificial intelligent technology,visual simulation technology,mathematics, management,decision-making,and other technologies have been used to build a complete set of evaluation models.Our work has established foundation for achieving scientific,systematic,and rational disaster evaluation.The main contributions of the paper are:(1)We put forward the three-layer architecture of the decision support system(DSS)for flood control,including interface layer,application layer,and foundation layer,and make detailed design separately.Hereinto,the foundation layer provides information and technology support for the system,it includes database(storing space data and attribute data),model-base and knowledge-base;the interface layer implements friendly interaction between system and analysers;the application layer uses data and models to implement the analysis of information requirements in every part,it is the hard-core of the system.After that,each function sub-system is designed,respectively.(2)A genetic algorithm(GA)-based BP neural network model for flood peak prediction is built.Aiming at the shortage of BP network used in the flood prediction,a novel GA-BP network prediction model is proposed through combined GA computing steps with network optimization process.With the practical data,simulation prediction and result analysis are done through adopting BP and GA-BP network model,respectively.The results demonstrate the flood prediction with the GA-BP network can improves the prection efficiency and preciseness.(3)We put forward an algorithm and model for terrain and flood simulation in the flood prediction system.In order to complete the visual simulation of submerge course,we do three-dimension landform simulation based on grid model method.Then we advance an improved adaption algorithm based on butterfly-subdivision model,and compute a new insert value point through combining with the butterfly-shape algorithm and multi-steps curve insert value,subdivide selectly the trangle unit network to simulate three-dimension terrain.It can be as a foundation architecture for the real-time flood system.And,from the DEM,we use improved grid spread algorithm to simulate flood-inundated process and estimate flood-inundated area and water-deep distribution under certain condition of flood water levels or flood volumes.Finally, we use Web-based 3D visualization tool(Java 3D)to draw flood hazard maps,and demonstrate flood-inundated process dynamically under different conditions.(4)The flood disaster fast evaluation algorithms during the flood control decision-making are studied.According to the evaluation requirements of the different flood phases and the evaluation indexes of the flood situation,we build the fast evaluation models of flood simulation using grey-relating analysis,BP network and RBF network.Thereinto,RBP network is improved based on the basic RBF network,that is,replacing the primary linearity layer with competion layer to make the model complete mode judgement.Through the instance analysis,three models can acquire the good evalutation results.Especially,the gray-relating analysis and RBF network model has the fast evaluation speed and high stability.Thus,these two models can be used into the fast flood disaster evaluation system and the operation results can provide scientific guidelines for the flood control decision-making.(5)We put forward a flood disaster evaluation model based on immune genetic algorithm (IGA).Aiming at the shortage of the weak local search capability and immaturity,we advance an IGA-NN model to evaluate the flood disaster,and present the IGA-NN design method and optimization steps.We compare the IGA-NN with the SGA-based optimized network model to show the superiority of the IGA-NN.The results shows that IGA has better global and locally optimization capability and improve the immaturity convergence of the SGA.IGA-NN can inflect the correlativity between the multi-indexes and disaster levels,evaluate the flood disaster, and have faster responsion speed and higher evaluation precision.(6)We design and implement flood precision and evaluation system for the flood control DSS.Through the requirement analysis of flood prediction and evaluation system,we design the system.According to the different course of the flood disaster development,the flood evaluation system is divided into three main functional modules,that is,flood prediction,flood decision-making,and flood evaluation.We discuss its important function in the flood contrl DSS and confirm its function architecture.In addition,we design the interface between the sub-systems.At last we implement the three function modules among flood prediction and evaluation system.Finally,we conclude the thesis,and provide the direction for further studies.更多还原