摘要
针对基于人工神经网络的暂态稳定评估数据预处理中的数据离散化进行了深入的研究,提出了一种基于信息熵和粗糙集理论的输入特征离散化新方法:通过对样本空间的聚类分析筛选出各条件属性在离散化过程中的可用断点;利用信息熵的相关概念,构建各条件属性的候选断点集;采用粗糙集理论中决策表不相容度的概念,检测出各条件属性间的最优断点组合。算例表明:该方法在保证暂态稳定评估精度的前提下,能有效地压缩训练样本集,减轻神经网络的训练负担,为基于神经网络的大系统暂态稳定评估提供了新思路。
As the kernel of data pretreatment, discretization of continuous attributes is investigated for transient stability assessment based on artificial neural networks. A new discretization scheme combined entropy with rough sets theory is proposed. The possible cut points for each conditional attribute are found out through clustering analysis of the input space. The initial cut point sets for each conditional attribute based on the entropy are established. The optimal combinations of initial cut point sets for different conditional attributes can be obtained through checking incompatibility degree of decision table based on rough sets theory. The simulation results indicate that the proposed scheme can compress the training sample set and reduce the training burden of artificial neural networks effectively. A new methodology is given for neural- network-based transient stability assessment of large-scale power systems.
出处
《中国电机工程学报》
EI
CSCD
北大核心
2005年第15期56-61,共6页
Proceedings of the CSEE
