CAST工艺的BP和RBF人工神经网络仿真模型  

BP and RBF Neural Network Simulation Models for CAST Process

作　　者：刘俊萍[1] 严敏[1] 胡坚 

机构地区：[1]浙江工业大学建筑工程学院,浙江杭州310032 [2]镇江水工业公司排水管理处,江苏镇江212003

出　　处：《中国给水排水》2009年第21期98-101,共4页China Water & Wastewater

基　　金：国家水体污染控制与治理科技重大专项(2008ZX07421-002)

摘　　要：将误差反向传播前馈(BP)神经网络模型和径向基函数(RBF)神经网络模型应用到CAST工艺中,并采用多输入、双输出神经网络模拟处理过程中各变量之间的关系和预测出水水质。误差分析结果表明,训练阶段RBF神经网络模型的拟合精度比BP神经网络模型的高,但两者的预测精度相差不大;测试阶段BP神经网络模型和RBF神经网络模型预测出水COD的平均相对误差分别为6.35%、6.80%,预测出水TN的平均相对误差分别为7.19%、5.49%,均在8%以下,这说明两种神经网络模型均可用于模拟CAST污水处理工艺各变量之间的关系和预测出水水质,为污水厂的运行管理提供了理论依据。Cyclic activated sludge technology （CAST）was simulated using back propagation （BP） and radial basis function （RBF） neural network models, and the multiple input and dual-output neural networks were used to simulate the relationship between every variable in the treatment process and to predict the effluent quality. The error analysis shows that the fitting precision of RBF neural network model is higher than that of BP neural network model in the training phase, but their prediction accuracy is more or less the same. In the test phase, the mean relative errors of predicting COD using BP and RBF are 6.35% and 6.80% respectively, the mean relative errors of predicting TN using BP and RBF are 7.19% and 5.49% respectively, all of the mean relative errors are lower than 8%. This indicates that the use of two neural network models to evaluate the performance of CAST is an efficient way to determine the complex dependencies among process variables, providing a theoretical basis for the operation and management of wastewater treatment plant.

关 键 词：CAST工艺 BP人工神经网络 RBF人工神经网络 出水水质 预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]