论文部分内容阅读
提出了基于模糊粗糙集理论的样本归一化方法,用于解决因神经网络分类器在不同类样本间距离较近时训练速度较慢的问题。将神经网络的输入作为粗糙集信息系统的条件属性,神经网络的输出作为决策属性,构建决策表。利用粗糙集理论对训练样本离散化,根据离散化样本与两类不同样本间的距离差和两类样本的能量差,利用模糊集理论对该原始样本进行伸缩处理,然后,对伸缩预处理后的样本进行归一化,最后,用归一化处理后的样本对神经网络进行训练。以配电网故障选线为例,对该方法进行了分析和验证。仿真实验结果表明,经模糊粗糙集理论样本归一化处理后的神经网络训练时间明显缩短。因此,该方法正确、有效。
A sample normalization method based on fuzzy rough set theory is proposed to solve the problem that the training speed is slow when the neural network classifier is in a short distance between different types of samples. The input of neural network is used as the condition attribute of rough set information system and the output of neural network as decision attribute to construct decision table. The training samples were discretized by rough set theory. According to the distance difference between the discretized samples and two different samples and the energy difference between the two samples, the original samples were scaled by fuzzy set theory. Then, Of the sample normalized. Finally, the neural network is trained with the normalized samples. Taking fault line selection in distribution network as an example, this method is analyzed and verified. The simulation results show that the neural network training time normalized by the fuzzy rough set theory sample is obviously shortened. Therefore, the method is correct and effective.