污水处理过程的递阶神经网络建模

Modeling wastewater treatment plant via hierarchical neural networks

针对污水处理过程的多变量和多非线性子系统的串级结构特点,提出了一种基于活性污泥过程机理的递阶神经网络建模方法.该方法将神经网络与过程机理模型以串级方式连接,以神经网络辨识活性污泥过程模型中的非线性组分反应速率.分析各子过程建模误差的关系,给出了模型的稳定学习算法和稳定性理论分析.最后通过某污水处理厂生化脱氮过程实际运行数据的实验表明所提出的建模方法是有效的.

A hierarchical neural networks based on the mechanism of activated sludge process is introduced for modeling the wastewater treatment plant （WWTP） which includes multivariable and multi-nonlinear subsystems with serial structure. This approach combines the neural network and the mechanism model in a serial configuration; and the nonlinear uncertainties of the activated sludge process are estimated by neural networks. A stable learning algorithm and the theoretical analysis are given for this model based on the relations of various modeling errors among sub-processes. Operational data of a wastewater treatment plant illustrate the efficacy of this modeling approach.