基于改进遗传算法的独立分量分析算法

针对现有独立分量分析算法的分离效果依赖于非线性对比函数的选择,并且无法有效地分离超高斯和亚高斯混合信号这一现象,提出了一种基于遗传算法的独立分量分析算法,该算法采用直方图法根据信号的样本序列来估计信号的概率分布,解决了信号问互信息的计算问题,然后通过遗传算法最小化信号间的互信息,实现了对线性混叠信号的分离;同时,针对标准遗传算法存在的一些缺点如局部搜索能力差、容易出现早熟收敛等,提出了一种改进遗传算法,提高了遗传算法的寻优能力.对模拟信号的分离结果表明,基于改进遗传算法的独立分量分析算法的性能优于FastICA算法,对亚高斯和超高斯信号的混合信号具有优异的分离能力.模拟仿真实验结果同时也证实了改进遗传算法的寻优能力.

New Independent Component Analysis Method Based on Improved Genetic Algorithm

The performance of existing independent component analysis methods is highly affected by the non-linear contrast functions that are selected according to the distribution of original signals,the separation results are not always ideal,especially for the mixture of super-Gaussian signal and sub-Gaussian signal.To solve this problem,a new independent component analysis method based on improved genetic algorithm was proposed,where the probability of separated signals was estimated by histogram method,so the mutual entropy could be easily evaluated,then genetic algorithm was applied to find the separation matrix to minimize the mutual entropy.At the same time,an improved genetic algorithm was proposed to overcome some shortcomings of standard genetic algorithm,such as poor local searching ability and premature convergence.Simulation results show that the proposed independent component analysis method is superior to FastICA in separating the mixture of super-Gaussian signal and sub-Gaussian signal.Simulation results also prove the optimization ability of improved genetic algorithm.