摘要
针对硫回收装置中硫化氢和二氧化硫浓度的实时监控预测问题,提出一种基于改进的深度回声状态网络(DeepESN)软测量建模方法,给出了其离线学习算法。改进的DeepESN网络能够通过多层回声状态网络的结构,可以对具有强非线性特性的化工过程进行有效的深度学习和预测。离线学习算法在求输出权值时加入了岭回归算法,有效地提高了网络学习的稳定性。将该方法在同等条件下与现有的软测量建模方法进行了比较,基于改进的DeepESN软测量建模方法具有更好的学习能力、更高的学习效率和预测精度。
In order to solve the problem of real-time monitoring and prediction of hydrogen sulfide and sulfur dioxide concentrations in sulfur recovery units,a soft sensor modeling method based on improved deep echo state network(DeepESN)is proposed,and its offline learning algorithm is given.For the chemical processes with strong nonlinear characteristics,the improved DeepESN can be used for effective deep learning and prediction through the structure of multilayer echo state network.A ridge regression algorithm is added to the offline learning algorithm to calculate the output weights,which effectively improves the stability of the network learning.Compared with existing soft sensor modeling,under the same condition,the improved DeepESN can achieve better learning capability,higher learning efficiency and prediction accuracy.
作者
岳文琦
YUE Wen-qi(Gansu Lanjing Optoelectronic Technology Co.,Ltd.,Lanzhou 730030,China)
出处
《测控技术》
2021年第10期63-68,共6页
Measurement & Control Technology
关键词
深度回声状态网络
软测量
预测
算法
化工过程
deep echo state networks(DeepESN)
soft sensor
prediction
algorithm
chemical processes
