前馈神经网络在非线性系统的建模及控制中有着广阔的应用前景，但是该网络的学习算法—向后传播算法（Ｂａｃｋｐｒｏｐａｇａｔｉｏｎ（ＢＰ）Ａｌｇｏｒｉｔｈｍ）算法存在一些不足。为了提高多层前馈神经网络的学习效率及稳定性，引入了非线性最小二乘法。通过与其他学习算法的比较，得出结论：其中用差商近似代替导数的Ｐｏｗｅｌ法是一种高效、快速的学习方法，其学习速率比带动量项的学习率自适应的ＢＰ算法高一个量级，而比ＤａｖｉｄｅｎＦｌｅｔｃｈｅｒＰｏｗｅｌ（ＤＦＰ）、ＢｒｏｙｄｅｎＦｌｅｔｃｈｅｒＧｏｌｄｆａｒｌＳｈａｎｎｏ（ＢＦＧＳ）等变尺度方法以及其他非线性最小二乘法的稳定性要好得多。 Feedforward neural networks have broad application prospects in the modeling and control of nonlinear systems, but there are some shortcomings in this algorithm, Backpropagation (BP) Algorithm. In order to improve the learning efficiency and stability of multi-layer feedforward neural network, a nonlinear least square method is introduced. Through the comparison with other learning algorithms, the conclusion is drawn that the Powel method, in which the difference quotient is approximately replaced by the difference quotient, is an efficient and fast learning method whose learning rate is higher than the BP algorithm with learning rate-adaptive driven algorithm Level, but much better than the variable-scale methods such as DavidenFletcherPowel (DFP), BroydenFletcherGoldfarlShanno (BFGS), and other nonlinear least-squares methods.