基于RLS在线学习算法的DeepESN动态软测量建模方法及应用  

DeepESN Dynamic Soft-sensing Modeling Method and Its Application Based on RLS Online Learning Algorithm

作　　者：岳文琦 YUE Wen-qi(Gansu Lanjing Optoelectronic Technology Co.,Ltd.)

机构地区：[1]甘肃兰晶光电科技有限公司

出　　处：《化工自动化及仪表》2021年第5期491-496,共6页Control and Instruments in Chemical Industry

摘　　要：提出了一种非线性自回归时间序列模型与深度回声状态网络(DeepESN)相结合的动态软测量建模方法,给出了深度回声状态网络的递推最小二乘(RLS)在线学习算法。将该方法应用于预测脱丁烷塔塔底丁烷含量,实现对相关产品质量的实时在线监控。在同等条件下,与回声状态网络(ESN)、泄漏积分回声状态网络(LiESN)的在线软测量建模结果进行比较,该方法的仿真结果显示出更高的预测性能和预测精度。A dynamic soft-sensing modeling method which combining a nonlinear autoregressive(NARX)time sequence model with a deep echo state network(DeepESN)was proposed and a recursive least squares(RLS)online learning algorithm for DeepESN was presented.Applying this method to predict butane content at debutanizer bottom can realize real-time online monitoring of the quality of relevant products.Under the same conditions,comparing it with the online soft-sensing modeling methods of basic echo state network(ESN)and leakage integral echo state network(LiESN)showed that,this method has higher prediction performance and accuracy.

关 键 词：软测量 丁烷含量预测 脱丁烷塔 深度回声状态网络 时间序列 递推最小二乘 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]