提出了一种新颖的有导-无导联合学习神经网络盲源分离方法.该方法用经典的有导学习误差反传规则训练多层感知器,对信号的概率分布函数进行自适应逼近,从而给出分离矩阵无导学习规则中非线性激励函数的恰当估计.将该方法与纯粹无导学习方法进行了对比,说明了其性能上的优越性. A novel blind source separation method based on guided-leadless learning neural networks is proposed, which uses the classical guided learning error backpropagation rule to train multi-layer perceptrons and adaptively approximates the probability distribution function of signals The appropriate estimation of the nonlinear excitation function in the non-guided learning rule of the separation matrix is ​​given.Compared with pure non-guided learning method, this method shows its superiority in performance.