【摘要】 针对BP算法在进行系统辨识时存在着速度慢、非平滑内插、受噪声影响很大、逼近精度不高，特别是对强干扰不具有鲁棒性等问题，提出一种BP网络的鲁棒算法，该算法直接利用样本点对样本点的分布特征进行估计，并采用带有损失因子的与误差分布有关的二次型能量函数，并用于动态系统辨识，仿真结果表明了算法不仅对白噪声具有鲁棒性，而且对强干扰也具有鲁棒性．更多还原

【Abstract】 A robust back propagation algorithm is proposed,which is aimed at solving problems that BPalgorithm has some confronted when applied to the dynamic identification,such as slow convergence, nonsmoothing interpolation,noise sensitivity and poor accuracy of identification,especially lack of robustness forstrong interference. The algorithm uses samples directly to estimate distribution characteristics of samplepoints in which a quadratic energy function with lose factors relevant to error of distributions is used. The algorithm can be used in identification of dynamical systems,the simulation results showed the robustness forwhite noise as well as for gross errors.更多还原