中文摘要：

当前，将MBR大规模推广并加以应用的过程中最大的阻碍是膜污染问题。运用模糊推理方法将MBR系统的混合液污泥浓度（MLSS）、操作压力、温度等输入参数转换成模糊集，并用改进的基于梯度的实时学习算法进行模糊计算；将通过计算得到的膜通量的模糊值反模糊化，得出系统膜通量的明确评价值。这样就可以找出在膜通量最大情况下的输入参数，以便改进输入参数，达到更好的出水水质。该方法能够使传统的梯度下降方法中所存在的收敛速度和震荡之间的冲突问题从根本上得到解决，并且仿真计算结果与实际实验情况相符。

英文摘要：

At present,the biggest obstacle in the process of applying MBR in the process of large-scale promotion and application is membrane fouling.Fuzzy inference method is used to convert the mixed liquid sludge concentration（MLSS）,operating pressure,temperature and other input parameters into fuzzy set,and the improved gradient-based real time learning method is used to calculate the MBR system.The fuzzy value of membrane flux obtained by calculation is defuzzified to get the clear value of system membrane flux.Thus,the input parameters under the maximum membrane flux can be found to improve the input parameters and realize the better effluent quality.This method can make the conflict between the convergence rate and the shock in the traditional gradient descent method,and the results of simulating calculation are consistent with the actual experimental results.