Abstract: It offers a fast algorithm for blind extraction based on the improved adaptive genetic algorithm in order to overcome shortcomings in the blind source separation，including slow convergence，not high accuracy and easy to fall into the local optimum.Marginal entropy minimize criterion of independent signal are established on the basis of negentropy criterion.Some of key technologies of the genetic algorithm are improved on the basis of optimization function of blind extraction，so it proposes a fitness function fitting for blind signal extraction and monitoring strategy preventing algorithm from local convergence so that the algorithm can automatically jump out of the best local，and rapidly converges in the global optimum point.Thus blind extraction of transient mixed signals are obtained fast when the improved adaptive genetic algorithm are used as optimization approach.Simulation experiments show that performance of the algorithm is stable，convergence rate high，and global optimum answer achieved，so that blind extraction is realized.

Key words: blind source separation, blind extraction, genetic algorithm, negentropy, marginal entropy, probability density estimation

摘要： 对盲分离问题中存在收敛速度慢、精度不高和容易陷入局部最优等缺点进行了研究，提出了一种基于改进自适应遗传算法的快速盲提取算法。在负熵判据的基础上，建立了最小化独立信号边缘熵准则。以盲提取目标优化函数为基础，对遗传算法的关键技术进行了改进，同时提出一种适合盲信号提取的适应度函数和防止算法局部收敛的监测策略，使算法能够自动跳出局部最优，快速地收敛于全局最优解。以改进的自适应遗传算法作为寻优算法，快速地实现了瞬时混合信号的盲提取。仿真实验表明，该算法性能稳定、收敛速度快，得到了全局最优解，有效地实现了信号盲提取。

关键词: 盲源分离, 盲提取, 遗传算法, 负熵, 边缘熵, 概率密度估计