基于支持向量机的中长期径流预报研究-水利水电工程专业论文.docxVIP

华 华 中 科 技 大 学 硕 士 学 位 论 文 II II Abstract Runoff forecast plays an important part in not only water recourses management, but also economical operation of hydropower system, as well as the comprehensive impact on flood control, navigation, water environment and sustainable development. Moreover the forecast accuracy has a direct effect on the hydropower distribution. However the hydrologic system is a large and complicated system and the runoff element is influenced by climate, geographic environment and human activities, etc. Thereafter, its change characteristic and rule are complicated, random, gray and non-liner. This makes runoff forecast, especially mid-and-long-term runoff forecast has become complicated and difficult, the traditional methods based on linear theories are difficult to make a great progress in hydrologic forecast. It is necessary to develop more new ways. Artificial neural network has been applied to hydrologic forecast widely due to the capability of self-learning, self-organizing and self-adapting, but it has over-learning, local optimization and other issues. However, support vector machine(SVM) avoid over-learning problem effectively because it is based on Vapnik-Chervonenkis(VC) dimension theory of statistical learning theory and structural risk minimization theory; SVM adopt quadratic programming and Lagrange theory to get a global optimization. Furthermore it introduces kernel function to predigest the solution of non-liner question. In conclusion, SVM will be applied to runoff forecast. First, thinking about the difficulties of the mid-and-long-term runoff forecast and in order to improve the forecast accuracy, after physically reviewing the forecast factors, the right ones have been chosen by statistically analysis and fuzzy optimum method. Second, the forecast has been done bases on the artificial neural network and SVM method using the right factors and annually runoff flow. After contrasting and studying the result, the SVM is fit f