本文讨论了采用多层前馈神经网络作自适应均衡器问题，提出了一种改进的递推最小二乘ＬＲＬＳ）新算法，即带有阈值处理的采用分段函数的速推最小二乘（ＲＬＳ）快速学习算法。它可克服传统ＢＰ算法收敛速度慢、易陷入局部最小类的缺点，经计算机仿真证明用该学习算法训练自适应均衡器时，大大缩短了训练时间，提高了收敛率。 This paper discusses the problem of using multi-layer feedforward neural network as an adaptive equalizer, and proposes an improved recursive Least Squares (LLSS) algorithm, that is, the push-push least squares with piecewise function RLS) fast learning algorithm. It can overcome the shortcomings of the traditional BP algorithm such as slow convergence rate and easy falling into local minimum class. The computer simulation proves that when training the adaptive equalizer by using the learning algorithm, the training time is greatly shortened and the convergence rate is improved.