基于Bagging的两阶段特征选择集成分类器研究

发布时间：2018-08-16 15:44

【摘要】：集成分类是利用分类器互补关系有效提高泛化能力的多分类器集成方法,日益成为机器学习的新的研究方向,在多个领域得到了有效应用。集成分类器性能指标关键在于两个方面因素的制约:一是个体分类器的精度,二是成员分类器相互间的差异化。目前大部分算法设计都是通过扰动训练集并以此来设计差异性较大的基分类器,如Bagging与Boosting算法等。如何获取成员分类器相互间较大差异化的同时提高成员分类器的精度,是当前集成学习研究的热点。本文提出了一种基于Bagging的两阶段特征选择集成分类器,从差异化成员分类器和提高成员分类器精度两个方面考虑,进一步提高集成效果。该集成分类器通过两阶段特征选择双重扰动训练集来提高个体分类器的精度和多样性,并选择性集成差异较大的个体分类器,最终运用加权投票思想输出分类结果。该集成分类器已运用于局部放电故障诊断的实际应用中,能有效地识别局部放电的类型,及时准确地对变压器的绝缘状况进行评估。主要工作如下:(1)提出了一种改进的集成分类器,通过融合Bagging算法和两阶段特征选择相结合方法来双重扰动训练集,来构建不同的输入空间,达到提高个体分类器多样性和精度的目的。(2)采用IAS算法对成员分类器进行差异化选择,并再次选用遗传算法择优,加权输出最终结果。实验表明,本集成分类器较其他集成分类器能获得较优的个体分类器精度和较大差异性的个体分类器组合。(3)在局部放电故障诊断实际问题中验证了本集成分类器。通过实验选择了熵值法、互信息、特征增减法作为两阶段方法的构成元素,选用SVM算法作为构建方法。实验表明,该集成分类器能较准确地识别出放电类别。同时,本集成分类器作为某商用变压器局放故障检测产品的内嵌挂块得到了应用。
[Abstract]:Ensemble classification is a multi-classifier ensemble method which can effectively improve generalization ability by using the complementary relationship of classifiers. It has become a new research direction of machine learning and has been effectively applied in many fields. The performance index of integrated classifier is restricted by two factors: the accuracy of individual classifier and the difference between member classifiers. At present, most algorithms are designed by perturbation training set to design the base classifier, such as Bagging and Boosting algorithms. How to obtain greater differences between member classifiers and improve the accuracy of member classifiers is a hot topic in the field of integrated learning. In this paper, a two-stage feature selection ensemble classifier based on Bagging is proposed. Considering the difference member classifier and improving the precision of the member classifier, the ensemble effect can be further improved. This ensemble classifier improves the accuracy and diversity of individual classifiers by selecting a dual disturbance training set for two-stage feature selection and selectively integrates individual classifiers with large differences. Finally the weighted voting idea is used to output the classification results. The integrated classifier has been applied to the practical application of partial discharge fault diagnosis. It can effectively identify the type of partial discharge and evaluate the insulation condition of transformer in time and accurately. The main work is as follows: (1) an improved ensemble classifier is proposed, which combines the Bagging algorithm and the two-stage feature selection method to construct different input space. To improve the diversity and accuracy of individual classifier. (2) IAS algorithm is used to select the member classifier and genetic algorithm is used again to select the best and the final result is weighted. The experimental results show that the ensemble classifier can obtain better individual classifier accuracy and more different individual classifier combination than other integrated classifiers. (3) this integrated classifier is verified in the practical problem of partial discharge fault diagnosis. The entropy method, mutual information method, feature addition and subtraction method are selected as the components of the two-stage method, and the SVM algorithm is chosen as the construction method. The experimental results show that the integrated classifier can identify the discharge types more accurately. At the same time, the integrated classifier is used as the embedded block of a commercial transformer PD fault detection product.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP181

【相似文献】