基于AK-RBF神经网络的出水氨氮软测量方法研究

Soft Measurement of Ammonia Nitrogen in Effluent Based on AK-RBF Neural Network

为解决污水处理过程出水氨氮难以精确测量问题,提出一种基于自适应核函数RBF神经网络的出水氨氮软测量方法。由于隐层激活函数对神经网络性能影响较大,AK-RBF神经网络将基于欧几里得的高斯核与余弦核通过线性组合形成新的隐层神经元激活函数。网络参数学习采用梯度下降算法推导的迭代公式更新以提高网络预测精度。仿真实验表明,基于AK-RBF神经网络的出水氨氮软测量方法能够在线预测出水氨氮,比RBF神经网络具有更高的预测精度和更好的自适应能力。

In order to solve the problem that it is difficult to measure ammonia nitrogen accurately,soft measurement of ammonia nitrogen in effluent based on AK-RBF is proposed in this paper.The performance of neural network is influenced by the hidden layer activation function,AK-RBF which is the basis function utilizes a linear combination of Euclidean distance based Gaussian kernel and cosine kernel.To improve the prediction accuracy of the network,the gradient descent algorithm is used to network parameter learning of AK-RBF network.The on-line prediction of ammonia nitrogen in effluent can be realized by the method mentioned in the paper,the method has higher prediction accuracy and better adaptive ability than RBF,which has shown in the ammonia nitrogen simulation in sewage treatment.