中文摘要：

针对训练模式对的小幅摄动可能对模糊神经网络的性能产生不利影响，提出了单体模糊神经网络对训练模式对摄动的鲁棒性概念，并就训练模式对的最大保序摄动的情形对单体模糊神经网络（MFNN）进行了具体分析，一般的模糊神经网络对训练模式对摄动的鲁棒性概念可类似定义。理论研究表明：当训练模式对发生最大γ保序摄动时，在h=5的条件下，单体模糊神经网络对训练模式对的摄动全局拥有好的鲁棒性，这将有助于MFNN系统的性能分析、学习算法的选择和模式对获取。

英文摘要：

Small perturbations of training pattern pairs may cause some disadvantages to performance of a fuzzy neural network （FNN） , and a new concept is established for the robustness of a monolithic fuzzy neural network （MFNN） to perturbations of training pattern pairs. When the training pattern pairs come into the keep-order perturbations, the robustness of MFNN model is analyzed. The concept for the robustness of a FNN to perturbations of training pattern pairs can be defined similarly. The theoretical studies show that the MFNN has good robustness with h = 5 when the training pattern pairs come into the γ keep-order perturbations, which is beneficial to the performance analyses, choice of learning algorithms, and the acquisition of pattern pairs of MFNN.