通常的神经网络逆元自适应控制器存在两个缺陷 ,一是训练算法收敛过慢 ;二是无法控制非最小相位系统 ,因而限制了其使用范围。利用Davidon最小二乘法训练多层前馈网络 ,用于逼近被控对象的逆模型 ,并利用构造伪系统的方法 ,构成一种对非最小相位系统仍然有效的神经网络逆元自适应控制器。仿真结果表明了该方法的有效性 The common neural network inverse element adaptive controller has two defects. One is that the training algorithm converges too slowly. The other is that it can not control the non-minimum phase system, which limits the scope of its application. The multi-layer feedforward network is trained by David’s least-squares method, which is used to approximate the inverse model of the controlled object. By using the method of constructing pseudo-system, a neural network inverse element adaptive controller which is still effective for non-minimum phase system is constructed. Simulation results show the effectiveness of the method