【作者】 高异；

【导师】 刘军；

【作者基本信息】 西安理工大学 ， 模式识别与智能系统， 2006， 硕士

【摘要】 支持向量机(SVM)是九十年代中期发展起来的新的机器学习技术，SVM是以统计学习理论(SLT)为基础，SLT着重研究小样本条件下的统计规律和学习方法的，而传统统计学前提是有足够多样本，当样本数目有限时难以取得理想的效果。SVM较好的解决了小样本、非线性、局部极小值等实际问题。采用支持向量机回归进行非线性系统建模与控制的研究是最近两三年以来产生的智能控制的一个研究领域。这种建模与控制方法不仅模型简单，有完备的理论支持，更重要的是提供了一种实现复杂的非线性系统的建模与控制的新方法，拓宽了智能控制的研究领域。 本文首先系统地学习了支持向量机的基本理论和应用，分析比较了国际上近期出现的各种支持向量机优化算法，研究了支持向量回归进行非线性系统建模的方法。讨论了支持向量机核参数及惩罚因子等参数对回归估计性能的影响。为了能够自动获取最优的支持向量机参数，避免参数反复试凑的冗长过程，本文将模糊逻辑推理方法与遗传算法相结合，提出了基于模糊遗传算法的SVM参数选择方法，用模糊逻辑在线调整遗传算法的杂交概率p_c和变异概率p_m，提高了标准遗传算法的收敛速度和精度，为解决支持向量机参数选取问题提供了一条有效途径。并将该方法用于非线性系统辨识中，仿真结果验证了该方法的有效性。 其次研究了最小二乘支持向量机(LS-SVM)非线性系统建模原理，并和支持向量机回归进行了分析比较。为了能够自动获取最佳的LS-SVM参数，提出采用模糊遗传算法用于LS-SVM的参数选取。并将基于遗传优化的LS-SVM用于温度传感器非线性校正，实验结果表明在小样本情况下，该算法相对于小脑模型(CMAC)网络在铂热电阻温度传感器校正应用中有更高的精度，是一种非常有效的方法。 最后针对近几年来工业过程控制领域研究的热点问题——非线性系统预测控制，本文提出了一种基于最小二乘支持向量机的模型预测控制方法，该方法采用LS-SVM建立预测模型、模糊遗传更多还原

【Abstract】 Support vector machine (SVM) is a new machine learning technique which is development in the middle of 90’s. Statistical Learning Theory (SLT) is foundation of SVM. SLT study statistical regulation and learning methods under small samples condition. Tradition statistics premise has enough samples. SVM is powerful for the problem characterized by small sample, nonlinearity, high dimension and local minima, and has high generalization. Because of its advantage on modeling of nonlinear system, SVMR became a new and strong tool in intelligent control field in recent years. Supporting by mathematics theory, SVMR nonlinear modeling and SVMR nonlinear control theory not only has a simple model, but also provides a new control theory which is suitable for complex nonlinear system.Firstly, this paper systematically studied basic theories and applications of the support vector machine. Various support vector machine methods that appear in the word are compared. Paper studies method that using support vector regression to identify nonlinear system. The effect of SVM kernel parameter and punish gene are discussed. In order to get the optimal parameters of SVM automatically, avoiding costs lots of time to select parameters, a SVM parameter selection approach based on fuzzy genetic algorithms is proposed in this paper. The nonlinear system identification is studied using the crossover probability p_c and mutation probability p_m of the on-line adjustment genetic algorithm based on fuzzy logic. Fuzzy genetic algorithms have rapider evolvement speed and higher precision than genetic algorithms. An effective method to solve SVM parameter selection is provided. This method is used in nonlinear system identification. The simulation results show that this method is very effective.Secondly, this paper studied theories of nonlinear system identification based on least square support vector machine (LS-SVM). Paper compares LS-SVM and SVM difference. In order to get the optimal parameters of LS-SVM automatically, use Fuzzy Genetic Algorithm to select parameters of LS-SVM, and use FGA-Least Square SVM to nonlinear calibration of temperature sensor. Experimental results show that this method has more accurate than CMAC network for nonlinear calibration of temperature sensor. This method is very effective.Finally, a predictive control algorithm based on least squares support vector machines (LS-SVM) model is更多还原