作　　者： ; ; ; ; ; ; ; ;

机构地区： 湖南大学环境科学与工程学院环境生物与控制教育部重点实验室

出　　处： 《中国环境科学》 2010年第11期1453-1458,共6页

摘　　要： 采用序批式生物膜反应器(SBBR)与人工快速渗滤系统(CRI)工艺结合对模拟生活污水进行处理,由于该工艺影响因素与出水参数的复杂非线性关系,利用人工神经网络(ANN)对SBBR-CRI处理生活污水的过程进行仿真模拟.在MATLAB语言环境下,以DO、淹没时间/落干时间、曝气时间/停曝时间、进水COD、进水NH4+-N、进水TP为输入因素,出水COD、NH4+-N、TN和TP为输出因素,构建具有自适应学习规则的人工神经网络.结合最优网络运行参数:隐含层节点数6,初始学习率0.13,动量因子0.6,训练次数6000次,对样本仿真学习,预测值与实际值拟合度较好,样本的绝对平均误差率在7.5%之内,均方根误差均在0.085之内.结果表明,当DO为2mg/L,曝气时间/停曝时间为2/1,淹没时间/落干时间为1/3时,NH4+-N去除率能达到98%以上,TN和TP去除率85%以上,COD去除率94%以上.通过权重分析,进水NH4+-N、DO和进水TP对出水参数影响较大. The water technology of SBBR combined with CRI was employed to treat simulative domestic wastewater.SBBR-CRI process was simulated using the artificial neural network which adapted to the complicated nonlinear relation between the influence factors and the effluent parameters.The artificial neural network with adaptive study algorithm was built with the inputs of DO,wetting time/drying time,aeration time/nonaeration time,the influent COD,NH4＋-N,TP and outputs of the effluent COD,NH4＋-N,TN,TP using MATLAB software.Combining with the parameter optimization of SI 6,lr 0.13,mc 0.6,studying time 6000,the numerical outputs and the experimental values matched well,and the MARE of the sample were within 7.5% and the RSM were within 0.085.NH4＋-N removal efficiency was over 98%,TN and TP removal efficiency were both over 85% and COD removal efficiency was over 94% under the conditions of DO concentration 2 mg/L,aeration time/nonaeration time 2/1 and wetting time/drying time 1/3.Through the weight analysis,it indicated that the influent DO,NH4＋-N and TP had a strong impact on the effluent parameters.

关 键 词： 人工神经网络 权重 人工快渗 生活污水

领　　域： [环境科学与工程]