基于改进的自组织特征映射网络的电能质量检测与识别

Detection and identification of power quality based on improved SOFM network

针对电能质量扰动信号频谱广(从0～数兆赫兹)、不同扰动信号之间相互叠加的特点,采用小波变换和神经网络(ANN)相结合的方法对电能质量扰动信号进行识别。利用db4小波对IEEE定义的9种电能质量扰动信号进行粗略分类,提取扰动特征信号;与其他文献中不同的是,这里利用一些少量的已知样本对权向量进行初始化,对网络进行非强制性的修正,确定收敛准则,自适应调节学习速率等,从而对自组织特征映射(SOFM)网络进行改进,利用有限的学习样本对神经网络进行训练,提高神经网络分类的精度。用改进的自组织特征映射网络对电能质量扰动信号进行Matlab仿真,结果表明达到了较好的分类效果。

According to its features in spectrum and superposition, a method combining wavelet transform and ANN（Artificial Neural Networks） to identify power quality disturbance is introduced. The eigenvalues of nine power quality disturbances defined by IEEE are extracted using db4. This method, different from other methods, uses less sample vectors to initialize weight vectors, revises the network non-enforcedly, determines the convergence criterion and regulates the learning rate adaptively to improve SOFM（Self Organization-Feature- Map） net and increase identification accuracy with limited samples for neural network training. Matlab simulative results show its effectiveness.