【作者】 李文娟；

【导师】 张莲；

【作者基本信息】 重庆理工大学 ， 测试计量技术及仪器， 2011， 硕士

【摘要】 水是万物生命之源,水环境管理的好坏直接影响到人类的生存与发展。而水环境质量评价是水环境管理所有工作的基础,传统的评价方法如单因子评价法和综合污染法,因其应用的局限性而备受质疑。因此,寻求一种客观、通用的水质评价方法显得尤为重要。近年来,BP神经网络在模式识别方面的突出表现为之带来了可能。将BP神经网络应用在水质评价中,可以克服传统评价方法的缺点,为各个河流水质类别的纵向对比提供了可能。但由于BP网络自身的缺陷与水质评价的特殊性,使得BP网络水质评价模型面临的两大问题——工作效率与识别精度问题,尚未得到很好的解决。本文围绕这两大问题进行了探索,对改进BP神经网络在水质评价中的应用进行了深入研究。本文的研究工作主要分为以下几个部分:(1)介绍了BP神经网络的基本理论知识,针对BP网络存在的三个缺陷以及其在水质评价中所遇到的问题,对现有的黄金分割算法进行了改进,将其用于寻找最优隐含层节点个数,达到了优化网络的目的。接着用LM算法对BP网络进行了改进,建立了基于LM-BP网络的水质评价模型,运用该模型对成都市新都区境内流域水质做出了评价,通过与综合污染指数法的评价结果比较,证明了该网络模型的可行性。(2)为了进一步增强网络的识别精度,将遗传算法与BP网络结合,运用遗传算法的全局寻优能力为BP网络寻找最优的权值与阈值,以此建立GA-BP网络水质评价模型。实验证明,该模型的网络性能(收敛速度与测试样本均方误差)均优于LM-BP网络模型。最后,利用GA-BP网络模型对同一实例进行检测,并分别与LM-BP网络评价模型及综合污染法的评价结果进行对比,证明了GA-BP网络水质评价模型更合理、实用。(3)为了探讨水质指标与类别之间蕴含的特殊关系,用线性插值取代随机插值生成样本,对已建立的GA-BP网络进行训练,通过实例检测结果对比,表明线性插值的评价结果不能反映水体污染的基本情况。由此证明了用随机插值生成训练样本最能体现水质指标与类别之间复杂的非线性关系。(4)以上研究表明,本文所建立的基于随机插值生成样本的GA-BP网络水质评价模型识别精度高,实用性和通用性强。最后,通过MATLAB R2009a实现了基于改进BP网络的水质评价模型人机交互界面(GUI),使BP网络水质评价模型从理论研究向实际应用又迈进了一步。更多还原

【Abstract】 Water is the source of life, and water environmental management has direct impact on humanity’s survival and development. Water environmental quality assessment is the basis of water environmental management. The traditional evaluation methods, such as single-factor evaluation and comprehensive pollution evaluation, are questioned because of their application limitations. Therefore, it’s very important for us to find an objective and universal water quality evaluation method. In recent years, the outstanding performance of BP neural network in pattern recognition makes it possible. The BP neural network used in water quality evaluation can overcome the shortcomings of traditional evaluation methods, and makes it possible for all kinds of rivers to compare water quality longitudinally. Because of the BP network’s defects and the particularity of water quality assessment, the problems of work efficiency and recognition accuracy have not well resolved to the water quality assessment model based on BP network. To solve these problems, this paper researched into water quality assessment model based on improved BP neural network. Main works are as follows:(1)The basic theory of BP neural network was introduced. Taking the BP network’s defects and its problems met in water quality assessment into account, the Golden Section Algorithm was improved to get the reasonable number of BP neural network’s hidden nodes. Then the BP network was improved by the LM algorithm, and a water quality assessment model based on LM-BP network was established. The optimal model was used to evaluate the water quality degree of Xindu area of Chengdu, which were compared with the results derived from comprehensive pollution evaluation method. The feasibility of the water quality assessment model, founded by LM-BP neural network, was proved.(2)In order to further improve the recognition accuracy of the network, the Genetic Algorithm and BP network were combined. By using genetic algorithm’s global search capability to find the optimal weights and threshold for the BP network, and the Water Quality Assessment based on GA-BP Network was established. Experiments indicate that the model’s network performance (convergence speed and the mean square error of test samples) are better than the LM-BP network model’s. Finally, the GA-BP network model was used to detect the same data as above, and the evaluation results were compared separately with the results from the LM-BP network model and the comprehensive pollution assessment method. The comparison showed that the GA-BP network water quality assessment model is more reasonable and practical than the model based on the LM-BP network.(3)In order to clarify the special relationship between the water quality indexes and water quality grades, linear interpolation was used to generate enough samples instead of random interpolation to train the founded GA-BP network. Comparing the test results, it was proved that the water evaluation grades can’t reflect the basic pollution condition of the river. This indicated that the training samples generated by random interpolation well revealed the complex nonlinear relationship between the water quality indexes and the water quality grades.(4)All studies above showed that the GA-BP network water quality assessment model trained by samples generated by random interpolation has high recognition accuracy, practicality and versatility. Finally, the Graphical User Interface (GUI) of the water quality assessment model based on improved BP network was set up by MATLAB R2009a, which accelerated the progress of the BP network water quality assessment model from the theory to the practical application.更多还原