Abstract:

To achieve efficient operation of the wastewater treatment plant(WWTP), it is necessary to establish a model that accurately describes the behavior of the plan. In this paper, the radial basis function neural network (RBFNN) is utilized in the modeling of the WWTP basing on the available influent and effluent data. Considering the bounded modeling error, linear-in-parameters set membership identification algorithm is used to describe an uncertain set of each vector representing the weights of the links between all the hidden neurons and one output neuron. Comparing with the existing methods which are all proposed for a single effluent variable, the method here builds a predictor model which can compute confidence intervals for multiple effluent variables simultaneously according to the values of the influent variables. The confidence intervals can characterize the existence ranges of the effluent variables, such that reliable estimates of them are obtained. By the estimates, the effluent quality or the WWTP performance can be evaluated. Besides, the interval predictor model is also applied to the fault detection and isolation of the WWTP to realize reliable operation. The experiment results show the satisfying performance of the proposed method.

Key words: wastewater treatment, model, parameter estimation, prediction, fault detection and isolation, set membership identification, radial basis function neural network

摘要：

为了实现污水处理厂的有效运行，需要建立能够精确描述水厂行为的模型。根据水厂入水和出水数据，采用径向基函数神经网络建立污水处理过程模型。考虑到建模误差有界，使用参数线性集员辨识算法分别得到隐含层到输出层各神经元连接权值向量的不确定集合描述。与现有的单输出区间预测模型相比，该模型能够根据水厂入水数据同时给出多个出水水质变量的置信区间。这些区间能表征出水变量的存在范围，从而实现水质变量的可靠估计，进而评估出水水质或水厂性能。此外，还将此出水区间预测模型用于污水处理厂的故障检测与隔离，以提高水厂运行的可靠性。实验结果表明文中所提方法的有效性。

关键词: 污水处理, 模型, 参数估值, 预测, 故障检测与隔离, 集员辨识, 径向基函数神经网络