基于输出层函数为线性函数的三层前馈神经网络，结合自适应步长和动量解耦的伪牛顿算法及 迭代最小二乘法导出了一种混合算法．仿真证明该混合算法自适应能力强、计算量小、收敛速度快，是一 种有效的工程实用算法。 Based on the three-layer feedforward neural network whose output layer function is a linear function, a hybrid algorithm is derived by combining pseudo-Newton algorithm with adaptive step and momentum decoupling and iterative least square method. Simulation shows that the hybrid algorithm has strong adaptability, low computational complexity and fast convergence speed. It is an effective engineering practical algorithm.