【作者】 刘广利；

【导师】 邓乃扬；

【作者基本信息】 中国农业大学 ， 管理科学与工程， 2003， 博士

【摘要】 经济预警不仅是经济学的重要研究领域，而且倍受各国政府和公众的普遍关注。其研究结果将直接关系到对经济状况的正确认识和判断，从而影响宏观政策的合理制定。但是，传统预警方法往往囿于专家经验和简单的数学模型，难于处理高度非线性模型，无法满足宏观经济预警的客观要求。支持向量机(Support Vector Machine，SVM)是最近流行的一种数据挖掘技术。由于其坚实的理论基础和良好的推广性能，支持向量机已成为近几年的研究热点。 论文将支持向量机、模糊理论与宏观经济预警研究等相结合，尝试建立起基于支持向量机的宏观经济预警方法体系，并对支持向量机的理论和方法进行拓广。同时，结合实际数据进行经济预警的实证分析，以达到理论与实践的结合。 本文的主要研究成果如下： 1．对经济预警的理论、研究方法和发展历史进行了回顾和综述；详细讨论了传统预警系统的预警本质，包括经典预警理论和新预警理论及其预警系统，建立了预警系统框架范式；深入分析了统计学习理论和支持向量机的基本理论和特点。 2．分析模式分类、SVM和宏观经济预警的内在联系，指出经济预警可以看作一个模式分类过程。论文将SVM与经济预警相结合，首次提出了SVC(Support Vector Classification)智能经济预警模型，实现了模型参数的自动选择，给出了具体算法步骤和实证分析。 3．鉴于经济预警过程的不确定性，论文首次提出了带有不确定性的支持向量分类(USVC)预警方法。该方法将专家意见和不确定性信息融入预警系统，实现了预警方法与专家智能的有机结合，为宏观经济预警研究提供了新的思路和方法，把模糊理论和SVM引入了一个新的应用空间。 4．多类经济预警过程可以看作有序回归问题，论文首次将有序支持向量回归应用到经济预警系统，并结合USVC建立了不确定性有序支持向量回归(UOSVR)经济预警模型。 5．特征选择是模式分类中的一个重要步骤；基于经济预警指标之间存在相关性和冗余，提出一类新的经济预警系统的预警指标选择算法：SVM预警指标选择方法。 6．就SVM的核方法在经济预警系统中的应用进行了分析，包括支持向量回归、支持向量时间序列预测方法和核主成分分析(KPCA)多指标综合评价方法，并给出了数据实验。更多还原

【Abstract】 Economic early warning is one of the most important research fields of economics. It is widely concerned about by all governments and public for its significance in economic subjects. The results of research on early warning show direct relation to the correctness of the cognition and judgment, to the choice of macroeconomic policies. However, usual warning methods often based on experts’ experience or simple math models. And it is hard to deal with nonlinear problems so as not to meet the demand of macroeconomic early warning. As a popular arithmetic for data mining, Support vector machine or SVM has drawn much attention on this topic in recent years for its stabile basis in theory and good generalization.A macroeconomic early warning method is proposed by SVM combined with fuzzy theory and early warning research in this paper. Some new models are established to generalize and update SVM. Meanwhile, the methods are testified though data experiments in practice. Main results as following:1. Summarize the theories, research methods and developing history of economic early warning. Discuss the warning nature of usual early warning system including classical and new warning system and establish the frame pattern of warning system. Analyze the basic theory and character of Statistic Learning Theory (SLT) and SVM.2. Analyze the relations among pattern classification, SVM and macroeconomic early warning. Points out that early warning can be viewed as a process of pattern classification. For the first time a new intelligent warning system based on support vector classification is proposed which can auto-select the parameters in the model.3. It is required that every input must be exactly assigned to one of these two classes without any uncertainty in standard SVC. USVC early warning algorithm on expert advices is proposed for the first time, which is able to deal with the training data with uncertainty. Realize the effective combination between warning methods and expert intelligence.4. Ordinal support vector regression (OSVR) early warning algorithm is designed for multi-class early warning problem which label is associated with an integer from 1 to k. And fuzzy OSVR early warning Method is also designed, which can deal with the training data with uncertainty. A proper membership model named WBM is also proposed to fuzzify all the training data of every class.5. A new economic warning indices selection method is proposed for SVC early warning system based upon finding those indices which minimize bounds on the leaver-one-out error.6. Analyze the application of kernel function in SVM including support vector regression (SVR), kernel time series prediction and kernel principal component analysis (KPCA) comprehensive evaluation model.更多还原