基于入侵性杂草优化算法的有噪独立分量分析方法

Noisy independent component analysis based on invasive weed optimization algorithm

针对传统独立分量分析(ICA)算法在含噪情况下分离效果不好,容易陷入局部收敛的问题,提出基于入侵性杂草优化(IWO)算法的有噪独立分量分析方法.以分离信号负熵和为目标函数,选用高斯密度函数估计负熵,消除目标函数中的不稳定项,提高算法的稳定性和准确性;采用入侵性杂草优化算法估计混合矩阵,提高算法的全局寻优性能.仿真结果表明:与传统Fast ICA和Fast NoisyICA算法相比,文中算法的分离信号和源信号的相似因数更大,随着信噪比增加,相似因数趋向1,可以更好地估计源信号;PI指标明显小于其他两种算法的,可以更为精确地估计混合矩阵.研究结果对有噪ICA信号处理有一定参考意义.

A noisy independent component analysis algorithm based on invasive weed optimization was proposed,aiming at solving the problems of poor separation performance and falling into local convergence in noisy condition of traditional ICA algorithms.It regarded the sum of negentropy of all separated signals as the objective function.The proposed algorithm chose the Gaussian density function to estimate negentropy.It succeeded in eliminating the unstable term and enhancing the stability of the algorithm.Mixing matrix could be estimated by invasive weed optimization algorithm which has good global optimal performance.Simulation results illustrate that,in contrast to Fast ICA algorithm and Fast NoisyICA algorithm,similarities obtained from the proposed algorithm between separated signals and source signals are higher.With the increase of the signal-to-noise ratio,similarities tend to 1,illustrating that it can estimate the source signals better.PI calculated according to the proposed algorithm are smaller than other two algorithms,which illustrates that it can estimate the mixing matrix more accurately.