【作者】 高鸿雁；

【导师】 崔宝侠；

【作者基本信息】 沈阳工业大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 神经网络自开创以来一直深受各国专家学者的重视，日渐成为一种重要的处理非线性问题的工具。预测是神经网络的又一个重要应用领域。经典的预测方法用于非线性系统预测有一定困难，而神经网络具有非常优良的非线性特性，特别适于高度非线性系统的处理。因此基于神经网络的智能预测是解决非线性预测问题的有效方法。 水资源是实现国家可持续发展的一项重要环境资源，在对其进行质量评价、影响预测时存在很多困难。由于水环境的复杂性，在利用常规方法来建立水质模型过程中，不可能把所有的影响因素都考虑进去，一般只把那些主要因素考虑进去而忽略那些次要因素。因此，不可避免地会给模型的结果带来不确定性。将神经网络引入水环境影响预测中，可以解决部分以上问题。 本文就神经网络在预测研究特别是在水环境影响预测评价中的应用作了以下工作：在基于GIS(地理信息系统)的水环境评价DSS(决策支持系统)中，完成了关于模型部分的设计与实现。其中包括：模型库子系统；水环境现状评价子系统；水环境预测分析子系统的设计与开发。在对原有水质数学模型进行分析、简化、具体实现的基础上，用神经网络黑箱模型代替水质模型进行水环境影响预测，用遗传算法对神经网络模型的参数及结构进行优化辨识。此外还对所有相关模型进行程序实现、界面开发及系统维护。 经实验结果表明，用神经网络预测模型代替原有水质数学模型具有结果精确、省时省力的优点，而且用遗传算法来选取神经网络的结构和初始权值，克服了结构选取时凭经验、靠凑试的缺点，较之单独使用神经网络模型运行速度更快、结果更加精确。更多还原

【Abstract】 Neural Network has been widely used in many fields since it came into being. Now ft has been developed to be a useful nonlinear processing tool. Prediction is one of important application field of neural network. Since most of the general predicting methods have difficult in processing nonlinear cases. While neural network is competent for nonlinear processing for its excellent nonlinear character, predicting methods based on neural network extend the space of predicting research.Water resource is an important environment resource in realizing the persistence development. In the course of establish water environment model, we commonly take the main factors into account and neglect the subordinate factors in virtual of the complexity of water environment. It brings on many problems in water environment evaluation management.The followings are what this paper has done in prediction research especially water environment influence evaluation. This paper designed and realized all models in the water environment evaluation DSS based on GIS. It includes model base management subsystem; water environment actuality evaluation subsystem; water environment forecast analysis subsystem. After predigesting and realizing the water quality mathematic models, the neural network model is used to forecast the water environment influence; the genetic arithmetic is used to optimize the parameters and structure of neural network. Then system designing, interface development and model programming have been done.The result shows that neural network model is more precisely and convenient than water mathematic model. Genetic algorithm can choose the exact structure of model. It makes the neural network more quickly and efficiently.更多还原