摘要
针对电力系统多因素负荷预测问题的复杂性,融合粗糙集方法与神经网络方法各自的优势,提出一种新型的负荷预测模型——粗糙集径向基函数神经网络模型(RSRBFN).运用粗糙集方法和信息熵概念,在不改变样本分类质量的条件下约简负荷影响因素,简化了网络输入变量.通过消去冗余信息,提炼学习样本,获得典型样本.用典型样本约简隐含层神经元和训练网络,并将网络连接权值学习的非线性极值问题转化为线性规划问题,使网络结构得到优化,提高径向基神经网络的计算效率和预测精度,增强实用性.数值实验结果说明RSRBFN模型是可行、有效、实用的.
For a multifactor power load prediction problem, this paper attempts to propose a new power load forecasting model-Rough Set Radial Basis Function Networks(abbreviated to RSRBFN), by combining rough set and artificial neural network. Rough set approaches and the conception of information entropy are employed to reduce factors of loads and input variables of the input layer with no changing classification quality of samples. Typical samples can been gotten by expurgating redundant information. These typical samples will been used to reduce neurons of the hidden layer and train neural network. The non-linear optimal problem about learning connection weights could been transformed to linear programming so as to optimize structure of neural network, improves the computing efficiency, the prediction accuracy and the potential practicability of Radial Basis Function Networks(abbreviated to RBFN). The feasibility, effectiveness and practicability \$f\$ RSRBFN was verified by experiments comparing RBAN with the proposed approach.
出处
《系统工程理论与实践》
EI
CSCD
北大核心
2004年第6期113-119,共7页
Systems Engineering-Theory & Practice
基金
重庆市科学技术计划项目(7117)
关键词
负荷预测
粗糙集
神经网络
信息熵
load forecasting
rough set
neural network
information entropy
