针对递归神经网络BP学习算法收敛慢的缺陷 ,通过引入递推预报误差 (RPE)学习算法 ,提出一种新的递归神经网络快速学习算法·该算法的基本原理是沿着指标函数的Gauss Newton搜索方向修正权值 ,同时不必计算二阶偏导数和进行矩阵求逆运算·仿真结果表明 ,该算法比传统的递归BP学习算法具有更快的收敛速度 ,用于非线性动态系统建模是有效的 Aiming at the shortcoming of slow convergence of BP learning algorithm in recurrent neural network, a new recurrent neural network fast learning algorithm is proposed by introducing Recursive Predictive Error (RPE) learning algorithm. The basic principle of this algorithm is to search along the Gauss Newton Direction correction weights without calculating second-order partial derivatives and matrix inversion. Simulation results show that the proposed algorithm has faster convergence rate than the traditional recursive BP learning algorithm and is effective for nonlinear dynamic system modeling