Support Vector Machine Regression Algorithm and Its Application Research

【摘要】 基于数据的机器学习是现代智能技术中的重要方面。统计学习理论(SLT)是一种专门研究小样本情况下机器学习规律的理论，它建立在一套较坚实的理论基础之上的，为解决有限样本学习问题提供了一个统一的框架也发展了一种新的通用学习方法一支持向量机(SVM)，较好的解决小样本学习问题。与神经网络等其它学习方法相比，它的结构通过自动优化的方法计算出来，并且避免了局部最小点、过学习等缺陷。 以往大部分研究主要集中在支持向量机分类理论和应用上，近年来关于支持向量机回归(SVMR)的研究也显示出其优异的性能。作为一个新的理论和方法，支持向量机回归在训练算法和实际应用等方面有诸多值得深入探讨的课题。 本论文就以上主要内容进行了深入的研究并取得了以下结果： (1) 在深入了解支持向量机回归的基本原理和算法的基础上，提出一种用于在线训练的支持向量机回归(OSVR)算法。在线情况下采用批量训练方法对支持向量机回归(SVR)进行训练是非常低效的，因为训练集每次的变化都会导致对支持向量机的重新训练。OSVR训练样本采用序列输入代替了常规的批量输入。通过对两个标准集的测试表明：OSVR算法与SVMTorch算法相比具有可在线序列输入，生成支持向量机少和泛化性能强的优点。 (2) 在分析和了解工业过程软测量原理的基础上，将支持向量机方法引入蒸煮过程纸浆的Kappa值软测量技术中。针对纸浆蒸煮过程机理复杂、影响因素众多和数据不完备条件下纸浆Kappa值预报问题，探讨了支持向量机方法在纸浆Kappa值预报中的应用，经过与线性回归方法和人工神经网络方法预报结果比较，表明该方法具有精度高、速度快、泛化能力强的特点，取得了较传统软测量建模方法更好的预报效果。 (3) 利用LS-SVM为辨识器，提出了一种新的基于LS-SVM模型的预测控制结构。最小二乘支持向量机(LS-SVM)方法克服了经典二次规划方法求解支持向量机的维数灾问题，适合于大样本的学习。对一典型非线性系统—连续搅拌槽反应器(CSTR)的仿真表明，该控制方案表现出优良的控制品质并能适应被控对象参数的变化，具有较强的鲁棒性和自适应能力。在控制性能方面它优于神经网络预测控制和传统的PID控制。 还原 【Abstract】 Data based machine learning is an important topic of modern intelligent techniques. Statistical Learning Theory or SLT is a small-sample statistics, which concerns mainly the statistic principles when sample are limited. Especially the properties of learning procedure in such cases. SLT provides us a new framework for the general learning problem and a novel powerful learning method called Support Vector Machine or SVM, which can solve small- sample learning problems better. It has many advantages compared to Article Neural Networks or other learning methods, for example the automatic structure selecting, overcoming the local minimum and over-fitting etc.Most of the research works focuse on the Support Vector Machine classify theory and application, and the recently research works on Support Vector Machine Regression or SVMR also show its excellent performance. As a novel theory and method, the training algorithm, practical application and many other topics of SVMR are need to be discussed.This dissertation concentrated on the research work listed below and achieved some creative results.(1) Based on good understanding of the Support Vector Machine Regression (SVR) theory and algorithm, an Online Support Vector Machine Regression (OSVR) algorithm is proposed. Bach implementations of Support Vector Regression are inefficient when used in an online setting, because they must be retrained from scratch every time the training set is modified. This paper presents an online support vector regression for regression problems that have input data supplied in sequence rather than in batch. The OSVR has been applied to two benchmark problems shows that the OSVR algorithm has a much faster convergence and results in a smaller number of support vectors and a better generalization performance in comparison with the existing algorithms.(2) After the analysis and comprehension of industrial process soft sensing, we introduce the support vector machine method into the soft sensing of Kappa number of kraft pulping process. Aiming at the problem of predicting Kappa number of kraft pulping process under circumstances of complicated process kinetics and poor basic information, the support vector machine method was introduced. The basic theory and algorithm of the method were presented and 还原