中文摘要：

针对污水处理过程出水氨氮难以在线测量的问题，文中提出了一种基于递归RBF神经网络的软测量方法来预测氨氮。首先，提取与出水氨氮相关的主元变量，剔除主元变量的异常数据。其次，利用递归RBF神经网络建立主元变量与出水氨氮的蕴含关系，完成出水氨氮软测量模型的设计。最后，将提出的出水氨氮软测量方法应用于污水处理实际运行过程，结果表明，基于递归RBF神经网络的软测量方法能够实现出水氨氮的在线预测；同时，与其他方法的比较结果显示基于递归RBF神经网络的软测量方法具有较好的预测精度。

英文摘要：

Due to the difficulties of effluent ammonia nitrogen network measured online in the waste water treatment process, a soft-computing method, based on the recurrent RBF neural network, is developed in this paper. Firstly, the principal component variables of the effluent ammonia nitrogen are extracted. And the abnormal data of the principal component variables are excluded. Secondly, the proposed recurrent RBF neural network is used to establish the relations between the principal component variables and the effluent ammonia nitrogen to complete the design of the soft-computing method. Finally, the proposed method is applied to measure the effluent ammonia nitrogen concentration in a real waste water treatment process. The results show that the recurrent RBF neural network soft-computing method can predict the effluent ammonia nitrogen concentration online. In addition, the comparisons with other methods show that this proposed soft-computing method has better predicting accuracy.