2.765

2022影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

不确定性环境下维纳模型的随机变分贝叶斯学习

刘切 李俊豪 王浩 曾建学 柴毅

刘切, 李俊豪, 王浩, 曾建学, 柴毅. 不确定性环境下维纳模型的随机变分贝叶斯学习. 自动化学报, 2024, 50(6): 1185−1198 doi: 10.16383/j.aas.c210925
引用本文: 刘切, 李俊豪, 王浩, 曾建学, 柴毅. 不确定性环境下维纳模型的随机变分贝叶斯学习. 自动化学报, 2024, 50(6): 1185−1198 doi: 10.16383/j.aas.c210925
Liu Qie, Li Jun-Hao, Wang Hao, Zeng Jian-Xue, Chai Yi. Stochastic variational Bayesian learning of Wiener model in the presence of uncertainty. Acta Automatica Sinica, 2024, 50(6): 1185−1198 doi: 10.16383/j.aas.c210925
Citation: Liu Qie, Li Jun-Hao, Wang Hao, Zeng Jian-Xue, Chai Yi. Stochastic variational Bayesian learning of Wiener model in the presence of uncertainty. Acta Automatica Sinica, 2024, 50(6): 1185−1198 doi: 10.16383/j.aas.c210925

不确定性环境下维纳模型的随机变分贝叶斯学习

doi: 10.16383/j.aas.c210925
基金项目: 国家重点研发计划(2021YFB1715000), 国家自然科学基金(61903051, U2034209)资助
详细信息
    作者简介:

    刘切:重庆大学自动化学院副教授. 2016年获得北京化工大学控制科学与工程专业博士学位. 主要研究方向为人工智能及其在复杂过程的控制和优化中的应用. 本文通信作者. E-mail: qieliu@cqu.edu.cn

    李俊豪:重庆大学自动化学院硕士研究生. 2020年获得西安理工大学自动化与信息工程学院学士学位. 主要研究方向为系统辨识与人工智能. E-mail: 202013021042@cqu.edu.cn

    王浩:重庆大学自动化学院硕士研究生. 2021年获得安徽师范大学物理与电子信息学院学士学位. 主要研究方向为模型预测与人工智能. E-mail: 202113021007@cqu.edu.cn

    曾建学:重庆大学自动化学院硕士研究生. 2018年获得华北科技学院电子信息工程学院学士学位. 主要研究方向为容器技术与人工智能. E-mail: lc9zjx@126.com

    柴毅:重庆大学自动化学院教授. 2001年获得重庆大学博士学位. 主要研究方向为信息融合, 故障诊断, 智能控制系统. E-mail: chaiyi@cqu.edu.cn

Stochastic Variational Bayesian Learning of Wiener Model in the Presence of Uncertainty

Funds: Supported by National Key Research and Development Program of China (2021YFB1715000) and National Natural Science Foundation of China (61903051, U2034209)
More Information
    Author Bio:

    LIU Qie Associate professor at the School of Automation, Chongqing University. He received his Ph.D. degree from Beijing University of Chemical Technology in 2016. His research interest covers artificial intelligence and its applications on the control and optimization of complex processes. Corresponding author of this paper

    LI Jun-Hao Master student at the School of Automation, Chongqing University. He received his bachelor degree from the School of Automation and Information Engineering, Xi'an University of Technology in 2020. His research interest covers system identification and artificial intelligence

    WANG Hao Master student at the School of Automation, Chongqing University. He received his bachelor degree from the School of Physics and Electronic Information, Anhui Normal University in 2021. His research interest covers model prediction and artificial intelligence

    ZENG Jian-Xue Master student at the School of Automation, Chongqing University. He received his bachelor degree from the School of Electronic Information Engineering, North China Institute of Science and Technology in 2018. His research interest covers container and artificial intelligence

    CHAI Yi Professor at the School of Automation, Chongqing University. He received his Ph.D. degree from Chongqing University in 2001. His research interest covers information fusion, fault diagnosis, and intelligent control system

  • 摘要: 多重不确定性环境下的非线性系统辨识是一个开放问题. 贝叶斯学习在描述、处理不确定性方面具有显著优势, 已在线性系统辨识方面得到广泛应用, 但在非线性系统辨识的应用较少, 且面临概率估计复杂、计算量大等难题. 针对上述问题, 以典型维纳(Wiener)非线性过程为对象, 提出基于随机变分贝叶斯的非线性系统辨识方法. 首先对过程噪声、测量噪声以及参数不确定性进行概率描述; 然后利用随机变分贝叶斯方法对模型参数进行后验估计. 在估计过程中, 利用随机优化思想, 仅利用部分中间变量概率信息估计模型参数分布的自然梯度期望, 与利用所有中间变量概率信息估计模型参数比较, 显著降低了计算复杂性. 该方法是首次在系统辨识领域中的应用. 最后, 利用一个仿真实例和一个维纳模型的Benchmark问题, 证明了该方法在对大规模数据下非线性系统辨识的有效性.
  • 图  1  维纳模型结构示意图

    Fig.  1  The structure of Wiener model

    图  2  SVBI中目标函数的更新示意图

    Fig.  2  The update process of the objective function in SVBI

    图  3  辨识参数的收敛状况

    Fig.  3  Convergence of identified parameters

    图  4  下界函数的收敛过程

    Fig.  4  Convergence process of the lower bound function

    图  5  预测输出与实际输出比较

    Fig.  5  Comparison of predicted output with actual output

    图  6  系统预测输出与实际输出

    Fig.  6  Predicted output and actual output of the system

    表  1  不同子采样数据点对应的参数辨识情况

    Table  1  Identification of parameters corresponding to different sub-sampling data points

    $ \langle \theta _0 \rangle $ $ \langle \theta _1 \rangle $ $ \langle \theta _2 \rangle $ $ \langle \theta _3 \rangle $ $ \langle \theta _4 \rangle $ $ \langle \lambda _0 \rangle $ $ \langle \lambda _1 \rangle $ $ \langle \lambda _2 \rangle $ 时间(s)
    真实值 1 −0.5000 0.2500 −0.1250 0.0625 0 1 1
    采样1个点 1±0 −0.5463±0.3604 0.2507±0.2471 −0.2446±0.2655 0.0358±0.2882 0.5434±0.4180 0.6625±0.2907 0.3803±0.2185 0.6005
    采样5% 1±0 −0.5060±0.0330 0.2693±0.0497 −0.1252±0.0323 0.0633±0.0323 0.0908±0.2707 0.9871±0.1480 0.9103±0.1246 3.1829
    采样10% 1±0 −0.5055±0.0248 0.2571±0.0257 −0.1341±0.0255 0.0594±0.0256 0.0631±0.0504 0.9684±0.0498 0.9499±0.0459 7.7402
    采样20% 1±0 −0.5077±0.0204 0.2544±0.0202 −0.1287±0.0289 0.0659±0.0291 0.0575±0.0540 0.9813±0.0518 0.9574±0.0451 11.4620
    采样全部 1±0 −0.5078±0.0278 0.2541±0.0283 −0.1299±0.0271 0.0685±0.0246 0.0777±0.0726 0.9439±0.1183 0.9252±0.1326 9.0772
    下载: 导出CSV

    表  2  不同异常值存在时的参数辨识情况

    Table  2  Parameter identification when different outliers exist

    $ \langle \theta _0 \rangle $ $ \langle \theta _1 \rangle $ $ \langle \theta _2 \rangle $ $ \langle \theta _3 \rangle $ $ \langle \theta _4 \rangle $ $ \langle \theta _5 \rangle $ 时间 (s)
    真实值 1 −0.5000 0.2500 −0.1250 0.0625 −0.03125
    无异常值 1±0 −0.4989±0.0292 0.2495±0.0293 −0.1254±0.0223 0.0611±0.0257 −0.0338±0.0262 2.9369
    2% 异常值 1±0 −0.5097±0.0389 0.2672±0.0497 −0.1305±0.0426 0.0652±0.0452 −0.0291±0.0494 2.9480
    5% 异常值 1±0 −0.5060±0.0330 0.2693±0.0497 −0.1252±0.0323 0.0633±0.0323 −0.0314±0.0523 3.1829
    10% 异常值 1±0 −0.5349±0.0325 0.2627±0.0323 −0.1314±0.0330 0.0685±0.0389 −0.0377±0.0355 2.9057
    下载: 导出CSV

    表  3  不同辨识方法的性能比较

    Table  3  Performance comparison of different recognition methods

    $ b_0 $ $ a_1 $ $ \langle \lambda _0 \rangle(\lambda_0) $ $ \langle \lambda _1 \rangle(\lambda_1) $ $ \langle \lambda _2 \rangle(\lambda_2) $ 均方误差 时间(s)
    真实值 1 0.5 0 1 1
    无异常值 SVBI 0.0648±0.0620 0.9633±0.0509 0.9766±0.0626 0.9136 2.936 9
    VBEM 0.0503±0.0346 0.9411±0.0393 0.9655±0.0459 0.8978 9.7046
    MLE 1±0 0.5102±0.0136 0.1054±0.0405 1.0154±0.0464 0.9490±0.0411 0.9130 9.0350
    PEM 1±0 0.4948±0.0172 0.0828±0.0524 0.9905±0.0373 1.0072±0.0449 0.9132 0.6474
    5% 异常值 SVBI 0.0575±0.0540 0.9813±0.0520 0.9573±0.0450 5.4540 2.9352
    VBEM 0.0503±0.0411 0.9770±0.0532 0.9748±0.0518 3.8695 9.7709
    MLE 1±0 0.4150±0.0711 −0.9407±0.1253 1.0019±0.1839 1.3715±0.1895 3.9574 9.6693
    PEM 1±0 0.4999±0.0549 0.1072±0.1871 0.9646±0.1926 0.9878±0.1558 3.8374 0.6580
    10% 异常值 SVBI 0.1439±0.1065 0.9163±0.0924 0.8416±0.0924 7.5364 2.9057
    VBEM 0.0556±0.0468 0.9711±0.0538 0.9568±0.0553 5.5110 9.9245
    MLE
    PEM 1±0 0.4723±0.2004 0.1458±0.5211 0.9746±0.3091 1.0030±0.3253 5.4992 0.6620
    下载: 导出CSV

    表  4  式(52)部分参数辨识结果

    Table  4  The identification results of the part parameters of the process (52)

    参数 $\theta_0$ $\theta_1$ $\theta_2$ $\theta_3$ $\theta_4$ $\theta_5$ $\theta_6$ $\theta_7$ $\theta_8$ $\theta_9$ $c_0$ $c_1$ $c_2$ $Q$ $R$
    结果值 −0.0390 0.0648 −0.0547 0.0856 −0.0462 0.2613 0.0501 0.2041 0.3396 0.4154 −0.0188 0.1035 −0.0030 0.0034 0.0014
    下载: 导出CSV

    表  5  不同方法的性能比较

    Table  5  Performance comparison of different methods

    采样点数 方法 均方误差(V) 参数个数 时间(s)
    2 000 SVBI 0.056 95 25 256.12
    VBEM 0.062 83 25 1 211.27
    SVBI 0.034 07 40 264.27
    VBEM 0.034 25 40 1 214.55
    10 000 SVBI 0.061 79 25 1 299.99
    VBEM 0.093 34 25 6 347.28
    SVBI 0.033 85 40 1 332.31
    VBEM 0.034 04 40 6 442.98
    下载: 导出CSV
  • [1] 王乐一, 赵文虓. 系统辨识: 新的模式、挑战及机遇. 自动化学报, 2013, 39(7): 933−942 doi: 10.1016/S1874-1029(13)60062-2

    Wang Le-Yi, Zhao Wen-Xiao. System identification: New paradigms, challenges, and opportunities. Acta Automatica Sinica, 2013, 39(7): 933−942 doi: 10.1016/S1874-1029(13)60062-2
    [2] 刘鑫. 时滞取值概率未知下的线性时滞系统辨识方法. 自动化学报, 2023, 49(10): 2136−2144

    Liu Xin. Identification of linear time-delay systems with unknown delay distributions in its value range. Acta Automatica Sinica, 2023, 49(10): 2136−2144
    [3] Stoica P. On the convergence of an iterative algorithm used for Hammerstein system identification. IEEE Transactions on Automatic Control, 1981, 26(4): 967−969 doi: 10.1109/TAC.1981.1102761
    [4] 张亚军, 柴天佑, 杨杰. 一类非线性离散时间动态系统的交替辨识算法及应用. 自动化学报, 2017, 43(1): 101−113

    Zhang Ya-Jun, Chai Tian-You, Yang Jie. Alternating identification algorithm and its application to a class of nonlinear discrete-time dynamical systems. Acta Automatica Sinica, 2017, 43(1): 101−113
    [5] 黄玉龙, 张勇刚, 李宁, 赵琳. 一种带有色量测噪声的非线性系统辨识方法. 自动化学报, 2015, 41(11): 1877−1892

    Huang Yu-Long, Zhang Yong-Gang, Li Ning, Zhao Lin. An identification method for nonlinear systems with colored measurement noise. Acta Automatica Sinica, 2015, 41(11): 1877−1892
    [6] Ljung L. Perspectives on system identification. Annual Reviews in Control, 2008, 34(1): 1−12
    [7] Schön T B, Wills A, Ninness B. System identification of nonlinear state-space models. Automatica, 2011, 47(1): 39−49 doi: 10.1016/j.automatica.2010.10.013
    [8] Billings S A. Nonlinear System Identification: NARMAX Methods in the Time, Frequency, and Spatio-Temporal Domains. Chichester: John Wiley & Sons, 2013.
    [9] Carini A, Orcioni S, Terenzi A, Cecchi S. Nonlinear system identification using Wiener basis functions and multiple-variance perfect sequences. Signal Processing, 2019, 160: 137−149 doi: 10.1016/j.sigpro.2019.02.017
    [10] Schoukens M, Tiels K. Identification of block-oriented nonlinear systems starting from linear approximations: A survey. Automatica, 2017, 85: 272−292 doi: 10.1016/j.automatica.2017.06.044
    [11] Bershad N J, Celka P, McLaughlin S. Analysis of stochastic gradient identification of Wiener-Hammerstein systems for nonlinearities with Hermite polynomial expansions. IEEE Transactions on Signal Processing, 2001, 49(5): 1060−1072 doi: 10.1109/78.917809
    [12] Valarmathi K, Devaraj D, Radhakrishnan T K. Intelligent techniques for system identification and controller tuning in pH process. Brazilian Journal of Chemical Engineering, 2009, 26(1): 99−111 doi: 10.1590/S0104-66322009000100010
    [13] Zhu Y. Distillation column identification for control using Wiener model. In: Proceedings of the American Control Conference. San Diego, USA: IEEE, 1999. 3462−3466
    [14] Schoukens J, Suykens J, Ljung L. Wiener-Hammerstein benchmark. In: Proceedings of the 15th IFAC Symposium on System Identification (SYSID 2009). Malo, France: 2009.
    [15] Hagenblad A, Ljung L, Wills A. Maximum likelihood identification of Wiener models. Automatica, 2008, 44(11): 2697−2705 doi: 10.1016/j.automatica.2008.02.016
    [16] Xu W Y, Bai E W, Cho M. System identification in the presence of outliers and random noises: A compressed sensing approach. Automatica, 2014, 50(11): 2905−2911 doi: 10.1016/j.automatica.2014.10.017
    [17] Bottegal G, Castro-Garcia R, Suykens J A K. A two-experiment approach to Wiener system identification. Automatica, 2018, 93: 282−289 doi: 10.1016/j.automatica.2018.03.069
    [18] Westwick D T, Schoukens J. Initial estimates of the linear subsystems of Wiener-Hammerstein models. Automatica, 2012, 48(11): 2931−2936 doi: 10.1016/j.automatica.2012.06.091
    [19] Giordano G, Gros S, Sjöberg J. An improved method for Wiener-Hammerstein system identification based on the fractional approach. Automatica, 2018, 94: 349−360 doi: 10.1016/j.automatica.2018.04.046
    [20] Liu Q, Lin W Y, Jiang S L, Chai Y, Sun L. Robust estimation of Wiener models in the presence of outliers using the VB approach. IEEE Transactions on Industrial Electronics, 2021, 68(11): 11390−11399 doi: 10.1109/TIE.2020.3028806
    [21] Xie L, Yang H Z, Huang B. FIR model identification of multirate processes with random delays using EM algorithm. AIChE Journal, 2013, 59(11): 4124−4132 doi: 10.1002/aic.14147
    [22] Agamennoni G, Nieto J I, Nebot E M. Approximate inference in state-space models with heavy-tailed noise. IEEE Transactions on Signal Processing, 2012, 60(10): 5024−5037 doi: 10.1109/TSP.2012.2208106
    [23] Bishop C M. Pattern Recognition and Machine Learning. New York: Springer, 2006.
    [24] Amari S I. Natural gradient works efficiently in learning. Neural Computation, 1998, 10(2): 251−276 doi: 10.1162/089976698300017746
    [25] Amari S I. Differential geometry of curved exponential families-curvatures and information loss. The Annals of Statistics, 1982, 10(2): 357−385
    [26] Bottou L. On-line learning and stochastic approximations. On-line Learning in Neural Networks. New York: Cambridge University Press, 1999.
    [27] Robbins H, Monro S. A stochastic approximation method. The Annals of Mathematical Statistics, 1951, 22(3): 400−407 doi: 10.1214/aoms/1177729586
    [28] Hoffman M D, Blei D M, Wang C, Paisley J. Stochastic variational inference. The Journal of Machine Learning Research, 2013, 14(1): 1303−1347
  • 加载中
图(6) / 表(5)
计量
  • 文章访问数:  632
  • HTML全文浏览量:  216
  • PDF下载量:  96
  • 被引次数: 0
出版历程
  • 收稿日期:  2021-09-27
  • 录用日期:  2022-03-01
  • 网络出版日期:  2022-05-08
  • 刊出日期:  2024-06-20

目录

    /

    返回文章
    返回