【作者】 潘磊；

【导师】 罗毅；

【作者基本信息】 华北电力大学（北京） ， 系统工程， 2010， 硕士

【摘要】 本文概括介绍了支持向量机算法及逆系统方的相关内容,并针对支持向量机参数难以选择的问题,通过分析参数变化对支持向量机学习结果的影响,提出了一种改进的粒子群优化参数的方法。即在粒子群优化算法的适应度函数中,引入了与支持向量机解的稀疏性相关性的优化方法。改进算法将支持向量个数在训练集中的比例与经验风险求和作为粒子群寻优的适应度函数,避免了可能存在的“过学习”问题。而后在采用参数优化保证支持向量机推广能力的基础上,将支持向量机算法引入逆系统建模,介绍了基于支持向量机的逆系统建模的一般方法；并针对多输入多输出非线性离散系统构造其以及α阶逆系统,通过仿真实验完成强耦合系统的解耦,证明了支持向量机建立逆系统模型的有效性。更多还原

【Abstract】 In this paper,the theory about Support Vector Machine (SVM) and inverse system method is firstly introced.And then,considering the problem of parameter selection of SVM. Particle Swarm Optimization(PSO) algorithm is applied to help to determine parameters of SVM.The sparseness in the training result of support vector machine is introduced as part of fitness function of PSO algorithm.Summation of empirical risk and support vector divided by training set’s size is employed as fitness function.In simulation experiment of a non-linear system, modeling result proved that the proposed method overcame overfitting problem in training process and the predict precision of SVM predict model is enhanced. As generalization ability of SVM’s trainning result is guaranteed,SVM algorithm was applied in inverse system modeling. The usual SVM inverse system method is detailedly introduced,and the modeling of MIMO nonlinear discrete system’sαth-order inverse was emulated and proved that SVM is just effective for modeling a th-order inverse and help a lot in decoupling the MIMO system.更多还原