【作者】 刘艳芳；

【导师】 顾雪平；

【作者基本信息】 华北电力大学（河北） ， 电力系统及其自动化， 2003， 硕士

【摘要】 由于暂态稳定评估问题输入空间的复杂性以及神经网络本身的局限性，神经网络的分类结果中不可避免地会存在误分类。误分类大多是由处于两类样本分布的边界区内的样本造成的。如果能够应用某种方法把容易造成误分类的样本划分出来，势必提高神经网络暂态稳定评估的可靠性。本文提出了几种划分样本边界区的方法:提出了一种应用于半监督BP算法的实用结束判据，并根据粗糙集理论，研究了一种新的粗糙分类机制，取得良好的效果；应用支持向量机理论，构造分类器并划分样本边界区；最后研究多个分类器集成的方法寻找样本边界区，同样提高了暂态稳定评估的可靠性更多还原

【Abstract】 The transient stability assessment based on ANNs suffers from the unavoidable misclassifications in the boundary region between the two classes due to the complexity of TSA input dimension and the limitation of ANN. If these cases can be found out by some methods, then the reliability of the assessment results will be improved. This paper proposed some methods for finding out sure regions and ambiguous regions defined by lower and upper approximations in rough set theory. An applicable ending-criterion for semi-supervised back-propagation algorithm was proposed and a new rough classifier framework was studied, the assessment results show the effectiveness of the proposed criterion. A new classifier based on support vector machines was studied and applied. At last, an assembling classification schemes was studied by integrating different classifiers to improve the classification reliability. Using the assembled classifier, the boundary cases liable to be misclassified can be picked out, the misclassifications are hence reduced significantly. Liu Yanfang (power system and its automation) Directed by Prof. Gu Xueping更多还原