基于神经网络的MBR仿真预测的研究

Research of MBR Simulation Predictions Based on Neural Network

该文在分析MBR膜污染形成机制、膜污染影响因素的基础上，首先利用主元分析法对影响膜污染的各种参数实现输入变量的降维和去相关，求出影响膜污染最为明显的三大因素：混合液悬浮固体（MLSS）、总阻力和操作压力（TMP），进而运用BP和RBF人工神经网络建立了这三大参数与表征膜污染程度大小的膜通量之间关系的MBR智能仿真系统模型，并分析了两种神经网络模型MBR污水处理膜污染过程的适应能力。实验结果表明：1）神经网络模型可以准确地反映出实际情况，具有很好的自适应能力；2）PCA—RBF神经网络模型的收敛精度高而且速度快于PCA-BP神经网络模型。

In this paper, based on the analysis of the MBR membrane fouling formation mechanism and the impact factors of membrane fouling, first of all,authors used principal component analysis on various parameters affecting the membrane fouling to achieve dimensionality reduction and decorrelation of the input variables and obtain the most obvious three factors affecting membrane fouling： the mixed liquor suspended solids （MLSS）, the total resistance and the operating pressure （TMP）, second, established the MBR intelligent simulation system model reflecting the relationship between the three parameters and the membrane flux characterizing the size of the extent of membrane fouling with BP and RBF neural network, and then analyzed the adaptability of the two neural network models with MBR wastewater treatment membrane fouling process. The experimental results show： 1） the neural network model can accurately reflect the actual situa- tion, with a good adaptive capacity; 2） the convergence precision and the approximation rate of the R_BF neural network model are better than BP neural network model.