针对经典自然梯度算法收敛速度慢的问题,文章提出了一种梯度搜索方向优化的自然梯度盲分离算法。首先设计了非线性函数,采用非线性函数对梯度矩阵的各个元素进行加权处理,优化算法迭代的搜索方向,使其更加贴近目标函数实际的变化方向。与分级迭代变步长的自然梯度算法和梯度优化的EASI算法相比,该算法进一步提高了收敛速度和分离精度。 Aiming at the problem of slow convergence of classical natural gradient algorithm, this paper proposes a natural gradient blind separation algorithm with gradient search direction optimization. First of all, the nonlinear function is designed. The non-linear function is used to weight each element of the gradient matrix and optimize the searching direction of the iterative algorithm so that it is closer to the actual changing direction of the objective function. Compared with the natural gradient algorithm and the gradient optimized EASI algorithm, the proposed algorithm can further improve the convergence speed and separation accuracy.