改进的互信息最小化非线性盲源分离算法  

Improved mutual information minimization algorithm for nonlinear blind source separation

作　　者：杨杰明[1] 齐厚颖 

机构地区：[1]东北电力大学信息工程学院,吉林吉林132012

出　　处：《电测与仪表》2015年第9期66-69,共4页Electrical Measurement & Instrumentation

基　　金：吉林省科技发展计划项目(2014020471GX)

摘　　要：提出了一种改进的互信息量最小化非线性盲源分离算法,改善了优化算法在串音误差方面大等的不足。该方法利用自然梯度优化算法来优化目标函数,避免了对矩阵的求逆计算,减少了计算时间。此外,在网络参数优化的过程中引入了扰动信号,提高了非线性盲源分离算法的寻优能力。实验结果表明改进的非线性盲源分离算法是有效的,而且相对传统的非线性盲源分离方法具有较小的误差。To reduce the error between the separated signal and the original signal , an improved mutual information minimization nonlinear blind source separation algorithm is proposed .To avoid computing matrix inversion and reduce the computing time , natural gradient method is used to optimize the cost function of blind source separation algorithm . Especially , disturbance signal is introduced to the process of parameter optimization to improve the searching ability of mutual information minimization nonlinear blind source separation algorithm .Experimental results show that the pro-posed nonlinear blind source separation optimization algorithm is validity and has smaller error comparing with the tra-ditional nonlinear blind sources separation .

关 键 词：非线性盲源分离 互信息 自然梯度 扰动信号 

分 类 号：TM933[电气工程—电力电子与电力传动]