高级搜索

时域流信号的多任务稀疏贝叶斯动态重构方法研究

董道广 芮国胜 田文飚

引用本文: 董道广, 芮国胜, 田文飚. 时域流信号的多任务稀疏贝叶斯动态重构方法研究[J]. 电子与信息学报, 2020, 42(7): 1758-1765. doi: 10.11999/JEIT190558 shu
Citation:  Daoguang DONG, Guosheng RUI, Wenbiao TIAN. Research on the Dynamic Sparse Bayesian Recovery of Multi-task Observed Streaming Signals in Time Domain[J]. Journal of Electronics and Information Technology, 2020, 42(7): 1758-1765. doi: 10.11999/JEIT190558 shu

时域流信号的多任务稀疏贝叶斯动态重构方法研究

    作者简介: 董道广: 男,1990年生,博士生,研究方向为贝叶斯统计学习、压缩感知和蒸发波导反演;
    芮国胜: 男,1968年生,教授,博士生导师,主要研究方向为混沌通信系统及现代滤波理论;
    田文飚: 男,1987年生,副教授,主要研究方向为压缩感知及蒸发波导反演
    通讯作者: 董道广,sikongyu@yeah.net
  • 基金项目: 国家自然科学基金(41606117, 41476089, 61671016)

摘要: 为了解决多任务观测条件下时域流信号动态重构面临的块效应问题,该文基于重叠正交变换(LOT)和稀疏贝叶斯学习的贪婪重构框架先后提出了一种流信号多任务稀疏贝叶斯学习算法及其鲁棒增强型的改进算法,前者将LOT时域滑窗推广到多任务条件下,通过贝叶斯概率建模将未知的噪声精度的估计任务从信号重构中解耦并省略,后者进一步引入了重构不确定性的度量,提高了算法的鲁棒性和抑制误差积累的能力。基于浮标实测数据的实验结果表明,相比多任务重构领域代表性较强的时间多稀疏贝叶斯学习(TMSBL)和多任务压缩感知(MT-CS)算法,本文算法在不同信噪比、观测数目和任务数目条件下具有显著更高的重构精度、成功率和效率。

English

    1. [1]

      LEINONEN M, CODREANU M, and JUNTTI M. Sequential compressed sensing with progressive signal reconstruction in wireless sensor networks[J]. IEEE Transactions on Wireless Communications, 2015, 14(3): 1622–1635. doi: 10.1109/TWC.2014.2371017

    2. [2]

      ASIF M S and ROMBERG J. Sparse recovery of streaming signals using 1-homotopy[J]. IEEE Transactions on Signal Processing, 2014, 62(16): 4209–4223. doi: 10.1109/TSP.2014.2328981

    3. [3]

      周超杰, 张杰, 杨俊钢, 等. 基于ROMS模式的南海SST与SSH四维变分同化研究[J]. 海洋学报, 2019, 41(1): 32–40. doi: 10.3969/j.issn.0253-4193.2019.01.005
      ZHOU Chaojie, ZHANG Jie, YANG Jungang, et al. 4DVAR assimilation of SST and SSH data in South China Sea based on ROMS[J]. Acta Oceanologica Sinica, 2019, 41(1): 32–40. doi: 10.3969/j.issn.0253-4193.2019.01.005

    4. [4]

      ZHANG Yonggang, ZHANG Jianxue, JIAO Lin, et al. Algorithms of wave reflective critical angle on interface[C]. SPIE 10250, International Conference on Optical and Photonics Engineering, Chengdu, China, 2017: 8–13. doi: 10.1117/12.2266713.

    5. [5]

      AO Dongyang, WANG Rui, HU Cheng, et al. A sparse SAR imaging method based on multiple measurement vectors model[J]. Remote Sensing, 2017, 9(3): 297. doi: 10.3390/rs9030297

    6. [6]

      TIPPING M E and FAUL A. Fast marginal likelihood maximisation for sparse Bayesian models[C]. The 9th International Workshop on Artificial Intelligence and Statistics, Key West, USA, 2003: 3–6.

    7. [7]

      WIPF D P and RAO B D. An empirical bayesian strategy for solving the simultaneous sparse approximation problem[J]. IEEE Transactions on Signal Processing, 2007, 55(7): 3704–3716. doi: 10.1109/tsp.2007.894265

    8. [8]

      ZHANG Zhilin and RAO B D. Sparse signal recovery in the presence of correlated multiple measurement vectors[C]. 2010 IEEE International Conference on Acoustics, Speech and Signal Processing, Dallas, USA, 2010: 3986–3989. doi: 10.1109/ICASSP.2010.5495780.

    9. [9]

      ZHANG Zhilin and RAO B D. Iterative reweighted algorithms for sparse signal recovery with temporally correlated source vectors[C]. 2011 IEEE International Conference on Acoustics, Speech and Signal Processing, Prague, Czech Republic, 2011: 3932–3935. doi: 10.1109/ICASSP.2011.5947212.

    10. [10]

      ZHANG Zhilin and RAO B D. Sparse signal recovery with temporally correlated source vectors using sparse Bayesian learning[J]. IEEE Journal of Selected Topics in Signal Processing, 2011, 5(5): 912–926. doi: 10.1109/JSTSP.2011.2159773

    11. [11]

      JI Shihao, DUNSON D, and CARIN L. Multitask compressive sensing[J]. IEEE Transactions on Signal Processing, 2009, 57(1): 92–106. doi: 10.1109/tsp.2008.2005866

    12. [12]

      ZHANG Zhilin and RAO B D. Extension of SBL algorithms for the recovery of block sparse signals with intra-block correlation[J]. IEEE Transactions on Signal Processing, 2013, 61(8): 2009–2015. doi: 10.1109/TSP.2013.2241055

    13. [13]

      BERNAL E A and LI Qun. Tensorial compressive sensing of jointly sparse matrices with applications to color imaging[C]. 2017 IEEE International Conference on Image Processing, Beijing, China, 2018: 2781–2785. doi: 10.1109/ICIP.2017.8296789.

    14. [14]

      HAN Ningning and SONG Zhanjie. Bayesian multiple measurement vector problem with spatial structured sparsity patterns[J]. Digital Signal Processing, 2018, 75: 184–201. doi: 10.1016/j.dsp.2018.01.015

    15. [15]

      QIN Yanhua, LIU Yumin, and YU Zhongyuan. Underdetermined DOA estimation using coprime array via multiple measurement sparse Bayesian learning[J]. Signal, Image and Video Processing, 2019, 13(7): 1311–1318. doi: 10.1007/s11760-019-01480-x

    16. [16]

      DU Yang, DONG Binhong, ZHU Wuyong, et al. Joint channel estimation and multiuser detection for uplink Grant-Free NOMA[J]. IEEE Wireless Communications Letters, 2018, 7(4): 682–685. doi: 10.1109/LWC.2018.2810278

    17. [17]

      SHAHIN S, SHAYEGH F, MORTAHEB S, et al. Improvement of flexible design matrix in sparse Bayesian learning for multi task fMRI data analysis[C]. The 23rd Iranian Conference on Biomedical Engineering and 20161st International Iranian Conference on Biomedical Engineering, Tehran, Iran, 2017: 3823–3826. doi: 10.1109/ICBME.2016.7890927.

    18. [18]

      FENG Weike, GUO Yiduo, ZHANG Yongshun, et al. Airborne radar space time adaptive processing based on atomic norm minimization[J]. Signal Processing, 2018, 148: 31–40. doi: 10.1016/j.sigpro.2018.02.008

    19. [19]

      WU Jingjing, LI Siwei, ZHANG Saiwen, et al. Fast analysis method for stochastic optical reconstruction microscopy using multiple measurement vector model sparse Bayesian learning[J]. Optics Letters, 2018, 43(16): 3977–3980. doi: 10.1364/OL.43.003977

    20. [20]

      WIJEWARDHANA U L and CODREANU M. A Bayesian approach for online recovery of streaming signals from compressive measurements[J]. IEEE Transactions on Signal Processing, 2017, 65(1): 184–199. doi: 10.1109/TSP.2016.2614489

    21. [21]

      MALVAR H S and STAELIN D H. The LOT: Transform coding without blocking effects[J]. IEEE Transactions on Acoustics, Speech, and Signal Processing, 1989, 37(4): 553–559. doi: 10.1109/29.17536

    22. [22]

      BABIN S M, YOUNG G S, and CARTON J A. A new model of the oceanic evaporation duct[J]. Journal of Applied Meteorology, 1997, 36(3): 193–204. doi: 10.1175/1520-0450(1997)036<0193:ANMOTO>2.0.CO;2

    1. [1]

      张天骐, 范聪聪, 葛宛营, 张天. 基于ICA和特征提取的MIMO信号调制识别算法. 电子与信息学报, 2020, 41(0): 1-8.

    2. [2]

      刘坤, 吴建新, 甄杰, 王彤. 基于阵列天线和稀疏贝叶斯学习的室内定位方法. 电子与信息学报, 2020, 42(5): 1158-1164.

    3. [3]

      刘汝卿, 蒋衍, 姜成昊, 李锋, 朱精果. 应用于激光雷达信号处理系统的放大电路接口设计. 电子与信息学报, 2020, 42(7): 1636-1642.

    4. [4]

      褚征, 于炯. 基于随机森林的流处理检查点性能预测. 电子与信息学报, 2020, 42(6): 1452-1459.

    5. [5]

      胡召鹏, 李实锋, 向渝. 一种新型调频广播授时信号体制研究. 电子与信息学报, 2020, 42(7): 1661-1665.

    6. [6]

      卢丹, 白天霖. 利用信号重构的全球导航卫星系统欺骗干扰抑制方法. 电子与信息学报, 2020, 42(5): 1268-1273.

    7. [7]

      陈根华, 陈伯孝. 复杂多径信号下基于空域变换的米波雷达稳健测高算法. 电子与信息学报, 2020, 42(5): 1297-1302.

    8. [8]

      周义明, 李英顺, 田小平. 基于瑞利多径衰落信道的信号包络频谱感知. 电子与信息学报, 2020, 42(5): 1231-1236.

    9. [9]

      石慧慧, 王萌, 饶永南, 卢晓春, 王雪. 北斗系统GEO-3卫星临退役期B1信号质量研究. 电子与信息学报, 2020, 42(7): 1573-1580.

    10. [10]

      周杨, 张天骐. 多径环境下异步长码DS-CDMA信号伪码序列及信息序列盲估计. 电子与信息学报, 2020, 41(0): 1-8.

    11. [11]

      雷大江, 张策, 李智星, 吴渝. 基于多流融合生成对抗网络的遥感图像融合方法. 电子与信息学报, 2020, 41(0): 1-8.

    12. [12]

      夏士超, 姚枝秀, 鲜永菊, 李云. 移动边缘计算中分布式异构任务卸载算法. 电子与信息学报, 2020, 41(0): 1-8.

    13. [13]

      雷维嘉, 杨苗苗. 时间反转多用户系统中保密和速率优化的预处理滤波器设计. 电子与信息学报, 2020, 42(5): 1253-1260.

    14. [14]

      刘通, 唐伦, 何小强, 陈前斌. 融合区块链与雾计算系统中基于网络时延和资源管理的优化任务卸载方案. 电子与信息学报, 2020, 42(0): 1-6.

    15. [15]

      刘明, 孟宪辉, 熊鹏, 刘秀玲. 基于核稀疏编码的阵发性房颤检测. 电子与信息学报, 2020, 42(7): 1743-1749.

    16. [16]

      刘焕淋, 杜理想, 陈勇, 胡会霞. 串扰感知的空分弹性光网络频谱转换器稀疏配置和资源分配方法. 电子与信息学报, 2020, 42(7): 1718-1725.

    17. [17]

      申滨, 吴和彪, 赵书锋, 崔太平. 基于稀疏感知有序干扰消除的大规模机器类通信系统多用户检测. 电子与信息学报, 2020, 41(0): 1-9.

    18. [18]

      王茜竹, 方冬, 吴广富. 基于改进稀疏度自适应匹配算法的免授权非正交多址接入上行传输多用户检测. 电子与信息学报, 2020, 42(0): 1-7.

  • 图 1  时域流信号的多任务在线滑窗观测

    图 2  不同算法的SER随观测数目变化的比较结果

    图 3  不同算法的SER随信噪比变化的比较结果

    图 4  不同算法的SR随观测数目变化的比较结果

    图 5  不同算法的SR随SNR变化的比较结果

    图 6  不同算法的运行时间随观测数目变化的比较结果

    图 7  不同算法的运行时间随SNR变化的比较结果

    表 1  目标函数、中间变量及超参数估计公式列表

    目标函数及其分解形式(其中${\tilde a_l} = 2{a_l} + M\left( {2d + 2} \right)$):
    $L\left( { { { {\bar{ \alpha } } }_t} } \right) = { { - \displaystyle\sum\limits_{l = 1}^L {\left\{ { { {\tilde a}_l}\ln \left( { {\bar{ y} }_{t,l}^{\rm{T} }{\bar{ C} }_l^{ - 1}{ { {\bar{ y} } }_{t,l} } + 2{b_l} } \right) + \ln \left| { { { {\bar{ C} } }_l} } \right|} \right\} } } / 2} \;\;\quad (29)$
    $L\left( { { { {\bar{ \alpha } } }_t} } \right) = { { - \displaystyle\sum\limits_{l = 1}^L {\left\{ { { {\tilde a}_l}\ln \left( {1 - \frac{ { { {q_{j,l}^2} / { {g_{j,l} } } } } }{ { {\alpha _j} + {s_{j,l} } } }} \right) + \ln \left( {1 - \alpha _j^{ - 1}{s_{j,l} } } \right)} \right\} } } / 2}\;\; (30)$
    中间变量:
    ${{\bar{ C}}_l} = {{I}} + {{{\varPsi }}_l}{\bar{ A}}_t^{ - 1}{{\varPsi }}_l^{\rm{T}},_{}^{}{{\bar{ C}}_{ - j,l}} = {{I}} + \displaystyle\sum\limits_{k \ne j} {\alpha _k^{ - 1}{{{\psi }}_{k,l}}{{\psi }}_{k,l}^{\rm{T}}} \left( {{\rm{SMT - SBL}}} \right)\;\; (31)$
    ${{\bar{ C}}_l} = {{\hat{ \varOmega }}_{t,l}} + {{{\varPsi }}_l}{\bar{ A}}_t^{ - 1}{{\varPsi }}_l^{\rm{T}},{{\bar{ C}}_{ - j,l}} = {{\hat{ \varOmega }}_{t,l}} + \displaystyle\sum\limits_{k \ne j} {\alpha _k^{ - 1}{{{\psi }}_{k,l}}{{\psi }}_{k,l}^{\rm{T}}} \left( {{\rm{SMT - RSBL}}} \right)\;\;\;(32)$
    ${s_{j,l}} = {{\psi }}_{j,l}^{\rm{T}}{\bar{ C}}_{ - j,l}^{ - 1}{{{\psi }}_{j,l}},_{}^{}{q_{j,l}} = {{\psi }}_{j,l}^{\rm{T}}{\bar{ C}}_{ - j,l}^{ - 1}{{\bar{ y}}_{t,l}},_{}^{}{g_{j,l}} = {\bar{ y}}_{t,l}^{\rm{T}}{\bar{ C}}_{ - j,l}^{ - 1}{{\bar{ y}}_{t,l}} + 2{b_l}\;\;\; (33)$
    ${S_{j,l}} = {{\psi }}_{j,l}^{\rm{T}}{\bar{ C}}_l^{ - 1}{{{\psi }}_{j,l}},_{}^{}{Q_{j,l}} = {{\psi }}_{j,l}^{\rm{T}}{\bar{ C}}_l^{ - 1}{{\bar{ y}}_{t,l}},_{}^{}{G_l} = {\bar{ y}}_{t,l}^{\rm{T}}{\bar{ C}}_l^{ - 1}{{\bar{ y}}_{t,l}} + 2{b_l}\;\;\; (34)$
    ${\alpha _j}$更新公式:
    ${\alpha _j} = \left\{ \begin{aligned} & {L / { {\theta _j} } },{\theta _j} > 0\\ & \infty ,\quad {\simfont\text{其他} } \end{aligned} \right.\;\;\;\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\quad (35)$
    ${\theta _j} = \displaystyle\sum\limits_{l = 1}^L {\frac{ { { {\tilde a}_l}({ {q_{j,l}^2} / { {g_{j,l} } } }) - {s_{j,l} } } }{ { {s_{j,l} }({s_{j,l} } - { {q_{j,l}^2} / { {g_{j,l} } } })} } } \;\;\;\;\qquad\qquad\qquad\qquad\qquad\qquad\quad (36)$
    下载: 导出CSV

    表 2  相关的快速更新公式列表

    添加原子${{{\psi }}_{j,l}}$删除原子${{{\psi }}_{j,l}}$维持原子${{{\psi }}_{j,l}}$说明
    $ {\rm{SMT - SBL}}:$$ {\rm{SMT - SBL}}:$$ {\rm{SMT - SBL}}:$添加情形中:
    $ \begin{array}{l} {{\tilde{\hat{ \varSigma }}}}_{t,l}^{{\bar{ w}}} = [{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}} + {{\hat \varSigma }_{jj,l}}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}{{\varPsi }}_l^{\rm{T}}{{{\psi }}_{j,l}}{{\psi }}_{j,l}^{\rm{T}}{{{\varPsi }}_l}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}},\\ - {{\hat \varSigma }_{jj,l}}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}{{\varPsi }}_l^{\rm{T}}{{{\psi }}_{j,l}}; - {{\hat \varSigma }_{jj,l}}{{\psi }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}},{{\hat \varSigma }_{jj,l}}]\\ {\tilde{ \mu }}_{t,l}^{{\bar{ w}}} = [{{\mu }}_{t,l}^{{\bar{ w}}} + {\mu _{j,l}}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}{{\varPsi }}_l^{\rm{T}}{{{\psi }}_{j,l}};{\mu _{j,l}}]\\ {{{e}}_{j,l}} = {{{\psi }}_{j,l}} - {{{\varPsi }}_l}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}{{\varPsi }}_l^{\rm{T}}{{{\psi }}_{j,l}} \end{array}$$ \begin{array}{l} {{\tilde S}_{k,l}} = {S_{k,l}} + {{{{({\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{{{\psi }}_{k,l}})}^2}} / {{{\hat \varSigma }_{jj,l}}}}\\ {{\tilde Q}_{k,l}} = {Q_{k,l}} + {{{\mu _{j,l}}{\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{{{\psi }}_{k,l}}} / {{{\hat \varSigma }_{jj,l}}}}\\ {{\tilde G}_l} = {G_l} + {{{{({\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{{{\bar{ y}}}_{t,l}})}^2}} / {{{\hat \varSigma }_{jj,l}}}} \end{array}$$ \begin{array}{l} {{\tilde S}_{k,l}} = {S_{k,l}} + {\gamma _{j,l}}{({\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{{{\psi }}_{k,l}})^2}\\ {{\tilde Q}_{k,l}} = {Q_{k,l}} + {\gamma _{j,l}}{\mu _{j,l}}{\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{{{\psi }}_{k,l}}\\ {{\tilde G}_l} = {G_l} + {\gamma _{j,l}}{({\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{{{\bar{ y}}}_{t,l}})^2} \end{array}$$ \begin{array}{l} {{\hat \varSigma }_{jj,l}} = {\left( {{{\tilde \alpha }_j} + {S_{j,l}}} \right)^{ - 1}},\\ {\mu _{j,l}} = {{\hat \varSigma }_{jj,l}}{Q_{j,l}} \end{array}$
    $ {\rm{SMT - RSBL}}:$$ {\rm{SMT - RSBL}}:$$ {\rm{SMT - RSBL}}:$删除情形中:
    $ \begin{array}{l} {{\tilde{\hat{ \varSigma }}}}_{t,l}^{{\bar{ w}}} = [{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}} + {{\hat \varSigma }_{jj,l}}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\psi }}_{j,l}}{{\psi }}_{j,l}^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\varPsi }}_l}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}},\\ - {{\hat \varSigma }_{jj,l}}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\psi }}_{j,l}}; - {{\hat \varSigma }_{jj,l}}{{\psi }}_{j,l}^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{\varPsi }}_l^{}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}},{{\hat \varSigma }_{jj,l}}]\\ {\tilde{ \mu }}_{t,l}^{{\bar{ w}}} = [{{\mu }}_{t,l}^{{\bar{ w}}} + {\mu _{j,l}}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\psi }}_{j,l}};{\mu _{j,l}}]\\ {{{e}}_{j,l}} = ({\hat{ \varOmega }}_{t,l}^{ - 1} - {\hat{ \varOmega }}_{t,l}^{ - 1}{{{\varPsi }}_l}{\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}){{{\psi }}_{j,l}} \end{array}$$ \begin{array}{l} {{\tilde S}_{k,l}} = {S_{k,l}} + {{{{({\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\psi }}_{k,l}})}^2}} / {{{\hat \varSigma }_{jj,l}}}}\\ {{\tilde Q}_{k,l}} = {Q_{k,l}} + {{{\mu _{j,l}}{\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\psi }}_{k,l}}} / {{{\hat \varSigma }_{jj,l}}}}\\ {{\tilde G}_l} = {G_l} + {{{{({\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\bar{ y}}}_{t,l}})}^2}} / {{{\hat \varSigma }_{jj,l}}}} \end{array}$$ \begin{array}{l} {{\tilde S}_{k,l}} = {S_{k,l}} + {\gamma _{j,l}}{({\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\psi }}_{k,l}})^2}\\ {{\tilde Q}_{k,l}} = {Q_{k,l}} + {\gamma _{j,l}}{\mu _{j,l}}{\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\psi }}_{k,l}}\\ {{\tilde G}_l} = {G_l} + {\gamma _{j,l}}{({\hat{ \varSigma }}_{j,l}^{\rm{T}}{{\varPsi }}_l^{\rm{T}}{\hat{ \varOmega }}_{t,l}^{ - 1}{{{\bar{ y}}}_{t,l}})^2} \end{array}$${\hat \varSigma _{jj,l}}$是${\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}$的第$j$个对角元素,${{\hat{ \varSigma }}_{j,l}}$是${\hat{ \varSigma }}_{t,l}^{{\bar{ w}}}$的第$j$列,${\mu _{j,l}}$是${{\mu }}_{t,l}^{{\bar{ w}}}$的第$j$个元素。
    通用公式:通用公式:通用公式:维持情形中:
    $ \begin{array}{l} {{\tilde S}_{k,l}} = {S_{k,l}} - {{\hat \varSigma }_{jj,l}}{({{\psi }}_{k,l}^{\rm{T}}{{{e}}_{j,l}})^2}\\ {{\tilde Q}_{k,l}} = {Q_{k,l}} - {\mu _{j,l}}{{\psi }}_{k,l}^{\rm{T}}{{{e}}_{j,l}}\\ {{\tilde G}_l} = {G_l} - {{\hat \varSigma }_{jj,l}}{({\bar{ y}}_{t,l}^{\rm{T}}{{{e}}_{j,l}})^2}\\ 2\Delta L = \sum\nolimits_{l = 1}^L {\ln \left[ {{{{{\tilde \alpha }_j}} / {\left( {{{\tilde \alpha }_j} + {s_{j,l}}} \right)}}} \right]} \\ \mathop {}\nolimits - \sum\nolimits_{l = 1}^L {{{\tilde a}_l}\ln \left[ {1 - {{\left( {{{q_{j,l}^2} / {{g_{j,l}}}}} \right)} / {\left( {{{\tilde \alpha }_j} + {s_{j,l}}} \right)}}} \right]} \end{array}$$ \begin{array}{l} 2\Delta L = - \sum\nolimits_{l = 1}^L {\ln \left( {1 - {{{S_{j,l}}} / {{\alpha _j}}}} \right)} \\ \mathop {}\nolimits \mathop {}\nolimits - \sum\nolimits_{l = 1}^L {{{\tilde a}_l}} \ln \left[ {1 + \frac{{{{Q_{j,l}^2} / {{G_l}}}}}{{{\alpha _j} - {S_{j,l}}}}} \right]\\ {{\tilde{\hat{ \varSigma }}}}_{t,l}^{{\bar{ w}}} = {\hat{ \varSigma }}_{t,l}^{{\bar{ w}}} - {{{{{\hat{ \varSigma }}}_{j,l}}{\hat{ \varSigma }}_{j,l}^{\rm{T}}} / {{{\hat \varSigma }_{jj,l}}}}\\ {\tilde{ \mu }}_{t,l}^{{\bar{ w}}} = {{\mu }}_{t,l}^{{\bar{ w}}} - {\mu _{j,l}}{{{{{\hat{ \varSigma }}}_{j,l}}} / {{{\hat \varSigma }_{jj,l}}}} \end{array}$$ \begin{array}{l} 2\Delta L = \sum\nolimits_{l = 1}^L {\left( {{{\tilde a}_l} - 1} \right)\ln \left( {1 + \frac{{{\alpha _j} - {{\tilde \alpha }_j}}}{{{\alpha _j}{{\tilde \alpha }_j}}}} \right)} \\ + \sum\nolimits_{l = 1}^L {{{\tilde a}_l}} \ln \frac{{\left[ {\left( {{\alpha _j} + {s_{j,l}}} \right){g_{j,l}} - q_{j,l}^2} \right]{{\tilde \alpha }_j}}}{{\left[ {\left( {{{\tilde \alpha }_j} + {s_{j,l}}} \right){g_{j,l}} - q_{j,l}^2} \right]{\alpha _j}}}\\ {{\tilde{\hat{ \varSigma }}}}_{t,l}^{{\bar{ w}}} = {\hat{ \varSigma }}_{t,l}^{{\bar{ w}}} - {\gamma _{j,l}}{{{\hat{ \varSigma }}}_{j,l}}{\hat{ \varSigma }}_{j,l}^{\rm{T}}\\ {\tilde{ \mu }}_{t,l}^{{\bar{ w}}} = {{\mu }}_{t,l}^{{\bar{ w}}} - {\gamma _{j,l}}{\mu _{j,l}}{{{\hat{ \varSigma }}}_{j,l}} \end{array}$${\hat \varSigma _{jj,l}},{{\hat{ \varSigma }}_{j,l}},{\mu _{j,l}}$与前述相同,${\gamma _{j,l}} = {\left[ {{{\hat \varSigma }_{jj,l}} + {{\left( {{{\tilde \alpha }_j} - {\alpha _j}} \right)}^{ - 1}}} \right]^{ - 1}}$。
    下载: 导出CSV
  • 加载中
图(7)表(2)
计量
  • PDF下载量:  14
  • 文章访问数:  676
  • HTML全文浏览量:  292
文章相关
  • 通讯作者:  董道广, sikongyu@yeah.net
  • 收稿日期:  2019-07-25
  • 录用日期:  2020-03-26
  • 网络出版日期:  2020-04-24
  • 刊出日期:  2020-07-01
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

/

返回文章