用神经网络和ANFIS模拟污水生物处理过程

Simulation of Wastewater Biotreatment Process by Using Neural Network and ANFIS

为了对污水生物处理过程进行有效的控制,首先要对该过程进行模拟以分析其动态特性。神经网络和ANFIS同样具有以任意精度逼近任何线性或非线性函数的功能,可以作为污水生物处理过程建模的工具。通过对深圳盐田污水处理厂的模拟发现:当采用实际运行数据作为模型的训练样本时,对样本进行适当的筛选处理是非常必要的;训练样本相同时,用ANFIS进行模拟则对出水COD和NH3-N的预测误差比用BP神经网络进行模拟的误差分别低79.7和86.8;在同样的预测精度下,用ANFIS模拟所需的训练样本数可比用神经网络的少很多。

For the effective control of wastewater biotreatment process, the process simulation should be firstly carried out to analyze the dynamic characteristics of the process. Neural network and ANFIS can both be used to approach linear or non-linear functions with any precision and can be used as tools of simulating wastewater biotreatment process. Simulation of Yantian Wastewater Treatment Plant indicates that it is very important to screen the sample data when using practical running data of wastewater treatment plant as training data. With the same sample data, the forecast errors for effluent COD and NH3 -N by using ANFIS are less than that by using neural network by 79.7% and 86.8% respectively. In order to get same forecast precision, less sample data are required by ANFIS than by neural network.