基于近似最小一乘准则的Hammerstein-Wiener模型随机梯度辨识  

Hammerstein-Wiener Model Stochastic Gradient Identification Based on Approximate Least Absolute Criterion

作　　者：徐宝昌[1] 荣志超 王雅欣 董秀娟 许立伟 XU Bao-chang;RONG Zhi-chao;WANG Ya-xin;DONG Xiu-juan;XU Li-wei(College of Information Science and Engineering,China University of Petroleum(Beijing;PipeChina Beijing Natural Gas Pipeline Co.,Ltd.)

机构地区：[1]中国石油大学(北京)信息科学与工程学院 [2]国家管网集团北京管道有限公司

出　　处：《化工自动化及仪表》2022年第1期1-6,19,共7页Control and Instruments in Chemical Industry

基　　金：国家重点研发计划项目——复杂油气智能钻井理论与方法(2019YFA0708304);中国石油天然气集团有限公司-中国石油大学(北京)战略合作科技专项(ZLZX2020-03)。

摘　　要：基于最小一乘准则和随机梯度算法原理,针对受尖峰噪声影响的非线性Hammerstein-Wiener模型,提出基于最小一乘准则的随机梯度算法。为解决最小一乘准则函数不可微的问题,该算法引入确定性可导函数近似代替残差绝对值。仿真实验表明:基于最小一乘准则的随机梯度算法对于非线性Hammerstein-Wiener模型可以有效地辨识出模型参数,同时抑制尖峰噪声对辨识结果的影响,具有更高的辨识精度、更快的收敛速度和良好的鲁棒性。In this paper,a stochastic gradient algorithm based on the least absolute criterion was proposed for the nonlinear Hammerstein-Wiener model that the spike noise affected.For purpose of solving the non-differentiable problem of the least absolute-criterion function,an approximate least absolute deviation objective function was established by introducing a deterministic differentiable function to replace the absolute residual.Simulation experiments show that,the proposed stochastic gradient method based on the least absolute criterion can effectively identify the model parameters for the nonlinear Hammerstein-Wiener model,and suppress the influence of the spike noise on the identification results and it has higher identification accuracy and stronger stability.

关 键 词：参数辨识 HAMMERSTEIN-WIENER模型 最小一乘准则 随机梯度 尖峰噪声 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]