基于模糊神经网络的A2/O工艺出水氨氮在线预测模型

采用厌氧/缺氧/好氧污水处理系统(A2/O)对人工合成污水进行处理,并利用人工神经网络(ANN)模型和自适应模糊人工神经网络(ANFIS)模型对A2/O处理污水的过程进行仿真模拟.在MATLAB环境下,选取可在线监测的水力停留时间(HRT)、进水pH值(pH)、好氧池溶解氧(DO)和混合液回流比(r)作为输入参量,系统出水氨氮浓度(NH4+eff)为输出量,建立在线预测模型.结合自适应模糊C均值聚类算法,确定ANFIS模型的模糊规则数及最优运行参数,对实验数据进行仿真预测.结果表明,与ANN模型相比,ANFIS模型的仿真输出值与实际值的拟合程度更高,相对误差在6.45%之内,平均绝对百分比误差(MAPE)为2.8%,均方根误差(RMSE)为0.1209,相关系数(R)达0.9956.模型训练过程中所得到的三维曲面图,可直观的反映各因素与出水氨氮浓度之间的非线性函数关系,为A2/O系统的高效稳定运行提供指导.

Online prediction model based on fuzzy neural network for the effluent ammonia concentration of A~2/O system

Based on the prototype experiment of treating synthetic wastewater in anaerobic/anoxic/oxic(A2/O) wastewater treatment system,an artificial neural network(ANN) model and an adaptive network based fuzzy inference system(ANFIS) model were employed to simulate the treatment process.When constructing the online prediction model in MATLAB,the online monitoring parameters,namely hydraulic retention time(HRT),influent pH(pH),dissolved oxygen(DO),and mixed-liquid return ratio(r),were adopted as the input variables,and effluent ammonia concentration(NH4+eff) was used as output variable.A self-adapted fuzzy c-means clustering algorithm was used to identify the fuzzy rules and optimize the model’s operational parameters.The simulation results shown that,compared with the ANN model,the ANFIS model’s predicted effluent ammonia concentrations fitted the observed ones better,which was supported by the maximum relative error of 6.45%,mean absolute percentage error(MAPE) of 2.8%,root mean square error(RMSE) of 0.1209,and correlation coefficient(R) value of 0.9956.Furthermore,3D surfaces obtained during the model training,which directly reflected the non-linear function between the factors and the effluent ammonia concentration,can guide the efficient and stable operation of the A2/O system.