基于多神经网络的污水氨氮预测模型

Multiple Neural Network-Based Model to Predict Ammonia Nitrogen Content in Wastewater

针对污水生化处理过程的非线性、大滞后等特点,建立了一种基于多神经网络的出水水质预测模型.通过减聚类方法将输入空间划分为一些小的局部空间,在每个局部空间中用神经网络建立子模型;各个子模型的预测输出通过主元递归(PCR)方法连接以解决子模型相互之间的严重相关问题,从而提高了模型的精度和鲁棒性;同时,应用改进目标函数以提高对偏高值的建模精度,采用加权反馈校正以提高模型的泛化能力.将该方法应用于某污水处理厂出水氨氮指标的预测,结果验证了模型的有效性.

In order to overcome the nonlinearity and large time delay in the biochemical process of wastewater,this paper proposes a prediction model of effluent quality based on the multiple neural network.In this model,the input space is divided into several subspaces via the subtractive clustering,and the corresponding submodels are established based on neural network in the subspaces.Then,in order to eliminate the severe correlation among the submodels and to improve the accuracy and robustness of the model,the submodels are combined via the principal component regression（PCR）.Moreover,the prediction accuracy of the high measured value is improved by using a modified objective function and the generalization ability of the model is strengthened via the weighted feedback correction.Finally,the proposed method is applied to the prediction of ammonia nitrogen content of the effluent from a wastewater treatment plant,and the results verify the effectiveness of the proposed method.