摘要
本文建议用一种新型的决策树（ＤＴ）－人工神经网络（ＡＮＮ）混合结构形式，把简单的ＡＮＮ装于ＤＴ的叶子上来拟合电力系统动态安全域，并且提出了一种改进的神经网络训练学习方法。同时，依据近似安全域的知识改进了样本的选取方法。验证表明，在采取了这三种新的方法之后所得的结果同传统的决策树和神经网络相比，不仅可使训练速度提高近一个数量级，而且在边界上具有很高的精度。
This paper suggests a new kind of composite structure for Descision Tree (DT) and Artifical Neural Network (ANN),which is used to simulate the power system Dynamic Security Region (DSR). Furthermore, it proposes an improved training method for the ANN. At the mean time, by the knowledge of approximated DSR, a measure to improve the method of the preparing patterns is presented. The test on an illustrative example power system shows that after adopting the above three methods, the training speed of DT-ANN is raised higher by nearly one quantity grade than that of traditional DT and ANN,and there is very high precision near to the boundary of DSR.
出处
《中国电机工程学报》
EI
CSCD
北大核心
1996年第6期378-383,共6页
Proceedings of the CSEE
基金
国家自然科学基金
关键词
动态安全域
决策树
神经网络
电力系统
dynamic security region
decision tree
artificial neural network
power system
