化学-生物絮凝工艺类神经网络模型比较研究

Comparison of chemical-biological flocculation process models based on artificial neural network

在化学-生物絮凝工艺中试研究的基础上,分别建立了基于BP类神经网络的多输入多输出(M IMO)模型与多输入单输出(M ISO)模型。应用化学-生物絮凝工艺中试6个不同工况的实测数据对2个模型进行训练,均表现出很好的收敛性。通过另外2个中试工况的实测数据对模型预测性能进行测试,M ISO模型对化学-生物絮凝反应器出水的COD、TP和SS的预测相对误差均低于M IMO模型,其预测相对误差均在9%以下。研究表明,M ISO模型是一个很易使用的建模工具,能很好地预测化学-生物絮凝工艺出水水质。

Finishing a pilot experiment of the chemical-biological flocculation process,the multi-input multi-output（MIMO） model and the multi-input single-output（MISO） model were made based on the back-propagation（BP） artificial networks.Trained by the data of the six different operating modes of the processes,two models achieved convergence well individually.The data of another two operating modes were used to test the ability of the model prediction.The percent errors of the MISO model were lower than that of the MIMO model and its percent errors were less than 9%.This study suggests that the MISO model is an easy-to-use modelling tool to obtain a quick preliminary assessment of effluent quality of the chemical-biological flocculation process.