利用多层前馈神经网络的非线性建模特性，基于动态ＢＰ网络的串并联和并联模型，提出一种高鲁棒性ＢＰ算法．与传统的ＢＰ算法相比，鲁棒ＢＰ算法有５个优点：（１）适合于非线性动态系统辨识；（２）辨识精度高；（３）不必内插所有训练样本；（４）具有高鲁棒性，能抵制过失误差和量测误差；（５）收敛速度得到了改进，因为错误样本的影响得到了适度的抑制．把该算法用于非线性动态系统辨识，仿真结果表明此方法是有效的 Based on the nonlinear modeling of multi-layer feedforward neural network and based on the series-parallel and parallel models of dynamic BP network, a highly robust BP algorithm is proposed. Compared with the traditional BP algorithm, the robust BP algorithm has five advantages: (1) suitable for nonlinear dynamic system identification; (2) high recognition accuracy; (3) no need to interpolate all training samples; (4) Robustness, can resist errors of error and measurement error; (5) The convergence rate has been improved, because the influence of the wrong sample has been moderately suppressed. The algorithm is applied to nonlinear dynamic system identification. Simulation results show that this method is effective