自适应步长神经网络盲分离算法的研究与应用  

Research and application of blind source separation algorithm of adaptative-step-size neural network

作　　者：杨硕[1] 刘小斌[2] 杨建青[1] 

机构地区：[1]甘肃农业大学工学院,甘肃兰州730070 [2]兰州工业学院汽车工程学院,甘肃兰州730050

出　　处：《甘肃农业大学学报》2016年第2期155-160,共6页Journal of Gansu Agricultural University

摘　　要：【目的】利用神经网络盲分离算法来解决含噪声音信号盲分离问题.【方法】提出了一种改进的自适应步长混合神经网络算法,该算法可以很好的掌控步长因子函数的形状,且在接近零点处步长变化缓慢,使性能更加优越;同时针对神经网络结构的不足引入了递归结构,通过与改进的自适应步长算法的结合达到更好的分离效果.【结果】通过在汽车发动机含噪声音信号中的应用,表明出该算法稳定性、收敛性好,可灵活的控制了步长因子分离出发动机噪声信号,达到较好的分离效果.[Objective] Blind source separation algorithm based on neural network was used to solve the problem of blind source separation of sound signal including noise. [Method] An improved adaptive-step- size compound neural network algorithm was proposed, which can well control the shape of the step size factor function, the step size changed slowly near the zero making the performance more superior. At the same time, recursive structure was introduced aiming at the shortcomings o{ the neural network structure, which was combined with the improved adaptive step size algorithm to achieve better separating effect. [Result] Application of the algorithm in automobile engine including noise signal showed good stability and convergence and it could control flexibly the step factor and separate engine noise signal achieving good separating effect.

关 键 词：盲分离 神经网络 自适应步长 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]