基于过采样结构的贝叶斯辨识方法  

Variational Bayesian Approach Based on Over-sampling Structure

作　　者：白振轩 徐宝昌[1] 陈贺龙 马超 BAI Zhen-xuan;XU Bao-chang;CHEN He-long;MA Chao(College of Information Science and Technology, China University of Petroleum;Kuche Oil and Gas Development Department, PetroChina Tarim Oilfield Company)

机构地区：[1]中国石油大学(北京)信息科学与技术学院 [2]中国石油塔里木油田分公司库车油气开发部

出　　处：《化工自动化及仪表》2019年第6期469-473,502,共6页Control and Instruments in Chemical Industry

基　　金：国家重点研发计划项目(2016YFC0303700)

摘　　要：将过采样闭环结构与贝叶斯变分法相结合,推导出基于过采样闭环结构的递推贝叶斯变分法,并且通过分析过采样闭环结构估计模型的渐近方差表达式,得出过采样结构可以利用超出模型频带之外的高频信息减小辨识模型的误差。仿真结果表明:基于过采样结构的贝叶斯变分法在输出噪声仅为白噪声情况下,相较于传统辨识方法具有更高的辨识精度。当输出噪声受到尖峰噪声或脉冲噪声污染时,笔者方法能够利用外加噪声中含有的高频信息提高辨识精度。In this paper, having Variational Bayesian Approach combined with over-sampled closed-loop structure to deduce the over-sampling closed-loop structure-based Variational Bayesian Approach was proposed, including having the the asymptotic variance expression of the over-sampled closed-loop structure estimation model analyzed to obtain that over-sampled structure can reduce the error of the identification model by means of utilizing the frequency band information beyond the model. Simulation results show that, the proposed approach has higher identification accuracy than the traditional identification method when white noise exiting only. When the system output is contaminated by spike noise or impulse noise, the proposed method can improve the identification accuracy by using the high-frequency information contained in spike noise or impulse noise.

关 键 词：过采样闭环结构 贝叶斯变分法 渐近方差 尖峰噪声 

分 类 号：TP13[自动化与计算机技术—控制理论与控制工程]