神经网络在造纸废水中zeta电位建模上的应用

On the Application of Neural Networks in Zeta Potential Modeling in Paper-Making Waste Water

针对造纸中段废水絮凝过程中zeta电位变化复杂,难以建立准确的数学模型的问题,提出利用神经网络对其中的zeta电位进行建模预测,研究了基于BP、Elman、RBF神经网络模型的多输入单输出建模方法。该建模法通过建立待测量zeta电位与絮凝剂投放量之间的非线性函数关系,间接得到待测变量的估计值。仿真结果表明RBF神经网络预测模型有较好的实时性和良好的泛化能力。与BP网络相比,RBF网络具有误差小、计算量小的优点,与Elman网络相比,具有周期短的优点,是一种有效的建模方法,能够快速、准确地在线预测废水中的zeta电位。

Due to the problem of complicated change of zeta potential in paper-making waste water floceulation process, ac- curate mathematical model is difficult to be established. Therefore, this paper proposes the use of neural network to establish zeta potential model and predicts its modeling values. It studies the muhiple input single output modeling method based on BP, El- man, RBF neurv] network. The modeling method can indirectly get unmeasured estimation values by establishing the nonlinear re- lationship between the measured amount of flocculant and zeta potential. The simulation results show that the RBF neural network forecast model has a good real-time performance and good generalization ability. Compared with BP neural network, RBF network has the advantages of less errors and simple calculation. While compared with Elman neural network, it has the advantage of short cycle. Therefore, RBF neural network is an effective modeling method, which can rapidly and accurately predict waste water zeta potential on-line.