基于径向基函数的非线性系统近似偏最小一乘准则辨识算法

doi:10.11949/j.issn.0438-1157.20180857

摘要/Abstract

摘要：

针对输入信号非线性相关的非线性系统，提出了基于径向基函数的近似偏最小一乘准则辨识算法。首先对观测数据矩阵进行列扩展，以径向基函数（radial basis function，RBF）网络的输出作为观测数据矩阵的扩展项，然后利用近似偏最小一乘算法对扩展的观测矩阵和输出矩阵进行线性回归。近似偏最小一乘算法用确定性可导函数近似代替残差绝对值，可以抑制对称α稳定（symmetrical alpha stable，SαS）分布的尖峰噪声。同时，通过主成分分析去除非线性系统数据向量矩阵之间的非线性相关，得出模型参数的唯一解。仿真实验表明，本文算法可以对输入信号存在非线性相关的非线性系统进行直接辨识，抑制了尖峰噪声对辨识结果的影响，具有优良的稳健性。

关键词: 非线性系统, 径向基函数, 主成分分析, 近似偏最小一乘, 尖峰噪声

Abstract:

For nonlinear system with nonlinear correlation of input signals, a partial approximate least absolute deviation based on radial basis function is proposed. In this paper, the observation data matrix is first extended by columns, and the output of the RBF network is used as an extension of the observed data matrix. Then, the extended observation data matrix and the output matrix are linearly regressed by using the partial approximate least absolute deviation. An approximate least absolute deviation objective function is established by introducing a deterministic function to replace the absolute value under certain situations. The proposed method can overcome the disadvantage of large square residual of least square criterion when the identification data is disturbed by the impulse noise which obeys symmetrical alpha stable (SαS) distribution. By adopting principal component analysis to eliminate the nonlinear correlation among the elements of data vector of nonlinear systems, the unique solution of model parameters can be easily acquired by the proposed method. The simulation experiments show that the proposed algorithm can directly recognize the nonlinear system with nonlinear correlation of input signals and suppress the influence of impulse noise.

Key words: nonlinear system, radial basis function, principal component analysis, partial approximate least absolute deviation, impulse noise

中图分类号:

TP 273

徐宝昌, 张华, 王金山. 基于径向基函数的非线性系统近似偏最小一乘准则辨识算法[J]. 化工学报, 2019, 70(2): 653-660.

Baochang XU, Hua ZHANG, Jinshan WANG. Partial approximate least absolute deviation for nonlinear system identification based on radial basis function[J]. CIESC Journal, 2019, 70(2): 653-660.

图/表 10

图1 RBF模型结构

Fig.1 RBF structure

图2 RBF-PALAD算法流程

Fig.2 Flow diagram of RBF-PALAD algorithm

图3 仅存在白噪声干扰时的相对误差曲线

Fig.3 Relative error curve when white noise exists only

表1 只有白噪声时的辨识结果

Table 1 Identified results when white noise exists only

Algorithm	a₁₁	a₂₂	b₁₁	b₂₂	c₁₂	c₂₂	$δ$ /%
RBF-PALAD	0.9625	0.9029	0.0613	-0.1387	0.3342	2.5840	1.65
RBF-PLS	0.9610	0.9036	0.0638	-0.1740	0.2464	2.0926	0.59
RBF-ALAD	NaN	NaN	NaN	NaN	NaN	NaN	NaN
true value	0.9608	0.9048	0.0699	-0.1748	0.2735	2.0381	0

表1 只有白噪声时的辨识结果

Table 1 Identified results when white noise exists only

Algorithm	a₁₁	a₂₂	b₁₁	b₂₂	c₁₂	c₂₂	$δ$ /%
RBF-PALAD	0.9625	0.9029	0.0613	-0.1387	0.3342	2.5840	1.65
RBF-PLS	0.9610	0.9036	0.0638	-0.1740	0.2464	2.0926	0.59
RBF-ALAD	NaN	NaN	NaN	NaN	NaN	NaN	NaN
true value	0.9608	0.9048	0.0699	-0.1748	0.2735	2.0381	0

图4 α=1.5时的相对误差曲线

Fig.4 Relative error curve when α=1.5

图5 α=1.2时的相对误差曲线

Fig.5 Relative error curve when α=1.2

图6 α=0.9时的相对误差曲线

Fig.6 Relative error curve when α=0.9

表2 α=1.5时的辨识结果

Table 2 Identified results when α=1.5

Algorithm	a₁₁	a₂₂	b₁₁	b₂₂	c₁₂	c₂₂	δ/%
RBF-PALAD	0.9589	0.9088	0.0477	-0.1188	0.3742	3.1502	5.37
RBF-PLS	0.9541	0.9173	0.0076	-0.2887	2.0352	1.2742	6.39
true value	0.9608	0.9048	0.0699	-0.1748	0.2735	2.0381	0

表3 α=1.2时的辨识结果

Table 3 Identified results when α=1.2

Algorithm	a₁₁	a₂₂	b₁₁	b₂₂	c₁₂	c₂₂	δ/%
RBF-PALAD	0.9605	0.9058	0.0482	-0.1950	0.3903	1.5724	2.52
RBF-PLS	0.9543	0.9132	-0.0345	-0.4892	-0.10835	0.6921	17.47
true value	0.9608	0.9048	0.0699	-0.1748	0.2735	2.0381	0

表4 α=0.9时的辨识结果

Table 4 Identified results when α=0.9

Algorithm	a₁₁	a₂₂	b₁₁	b₂₂	c₁₂	c₂₂	δ/%
RBF-PALAD	0.9195	0.0596	1.0655	-0.1284	1.7075	3.0427	7.55
RBF-PLS	0.9571	0.9023	-0.2999	-1.3921	-0.3546	0.1617	70.00
true value	0.9608	0.9048	0.0699	-0.1748	0.2735	2.0381	0

参考文献 30

1	李幼凤, 苏宏业, 褚健. 子空间模型辨识方法综述[J]. 化工学报, 2006, 57(3): 473-479.
	LiY F, SuH Y, ChuJ. Overview on subspace model identification methods[J]. Journal of Chemical Industry and Engineering(China), 2006, 57(3): 473-479.
2	吕恩辉, 王雪松, 程玉虎. 基于反卷积特征提取的深度卷积神经网络学习[J]. 控制与决策, 2018, 33(3): 447-454.
	LyuE H, WangX S, ChengY H. Deep convolution neural network learning based on deconvolution feature extraction[J]. Control and Decision, 2018, 33(3): 447-454.
3	BekirogluK, LagoaC, LanzaS T, et al. System identification algorithm for non-uniformly sample data[J]. IFAC-PaperOnLine, 2017, 50(1): 7296-7301.
4	FeldmanM, BraunS. Nonlinear vibrating system identification via Hilbert decomposition[J]. Mechanical Systems and Signal Processing, 2017, 84(B): 65-96.
5	ZhangX Y, HoaggJ B. Subsystem identification of multivariable feedback and feedforward systems[J]. Automatica, 2016, 72: 131-137.
6	张伟, 许爱强, 平殿发. 基于稀疏化核方法的非线性动态系统在线辨识[J]. 系统工程与电子技术, 2017, 39(1): 223-230.
	ZhangW, XuA Q, PingD F. Nonlinear system online identification based on kernel sparse learning algorithm with adaptive regulation factor[J]. Systems Engineering and Electronics, 2017, 39(1): 223-230.
7	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[J]. 自动化学报, 2017, 43(1): 101-113.
	ZhangY J, ChaiT Y, YangJ. Alternating identification algorithm and its application to a class of nonlinear discrete-time dynamical systems[J]. Acta Automatica Sinica, 2017, 43(1): 101-113.
8	SchoukensM, TieldsK. Identification of block-oriented nonlinear systems starting from linear approximations: a survey[J]. Automatica, 2017, 85: 272-292.
9	ChamC L, TanA H, TanW H. Identification of a multivariable nonlinear and time-varying mist reactor system[J]. Control Engineering Practice, 2017, 63: 13-23.
10	李寒霜, 赵忠盖, 刘飞. 基于变分贝叶斯算法的线性变参数系统辨识[J]. 化工学报, 2018, 69(7): 3125-3134.
	LiH S, ZhaoZ G, LiuF. Identification of linear parameter varying systems with variational Bayesian algorithm[J]. CIESC Journal, 2018, 69(7): 3125-3134.
11	罗健旭, 邵惠鹤. 应用多神经网络建立动态软测量模型[J]. 化工学报, 2003, 54(12): 1770-1773.
	LuoJ X, ShaoH J. Developing dynamic soft sensors using multiple neural networks[J]. Journal of Chemical Industry and Engineering（China）, 2003, 54(12): 1770-1773.
12	白晶, 毛志忠, 浦铁成. 多变量Hammerstein-Wiener模型的参数辨识[J]. 东北大学学报(自然科学版), 2018, (1): 6-10.
	BaiJ, MaoZ Z, PuT C. Parameter identification of multivariate Hammerstein-Wiener model[J]. Journal of Northeastern University(Natural Science), 2018， (1): 6-10.
13	叶健, 葛临东, 吴月娴. 一种优化的RBF神经网络在调制识别中的应用[J]. 自动化学报, 2007, 33(6): 652-654.
	YeJ, GeL D, WuY X. An application of improved RBF neural network in modulation recognition[J]. Acta Automatica Sinica, 2007, 33(6): 652-654.
14	韩红桂, 乔俊飞, 薄迎春. 基于信息强度的RBF神经网络结构设计研究[J]. 自动化学报, 2012, 38(7): 1083-1090.
	HanH G, QiaoJ F, BoY C. On structure design for RBF neural network based on information strength[J]. Acta Automatica Sinica, 2012, 38(7): 1083-1090.
15	黄晓峰, 潘立登. 基于增广RBF神经网络的混沌系统辨识[J]. 北京化工大学学报, 1999, 26(1): 45-48.
	HuangX F, PanL D. Identifying chaotic system using extended RBF neural net work[J]. Journal of Beijing University of Chemical Technology(Natural Science Edition), 1999, 26(1): 45-48.
16	董均华, 徐向东. 基于RBF神经网络的热力系统煤种辨识方法[J]. 清华大学学报, 2006, 46(8): 1430-1433.
	DongJ H, XuX D. Coal-information online identification based on RBF neural network[J]. Journal of Tsinghua University(Science and Technology), 2006, 46(8): 1430-1433.
17	樊兆峰, 马小平, 邵晓根. 非线性系统RBF神经网络多步预测控制[J]. 控制与决策, 2014, 29(7): 1274-1278.
	FanZ F, MaX P, ShaoX G. RBF neural network multi-step predictive control for nonlinear systems[J]. Control and Decision, 2014, 29(7): 1274-1278.
18	WalczakB, MassartD L. The radial basis function-partial least squares approach as a flexible non-linear regression technique[J]. Analytical Chemical Acta, 1996, 331(3): 177-185.
19	方甜莲, 贾力. 含有色噪声的神经模糊Hammerstein模型分离辨识[J]. 控制理论与应用, 2016, 33(1): 23-31.
	FangT L, JiaL. Separation identification of neuro-fuzzy Hammerstein model with colored noise[J]. Control Theory and Application, 2016, 33(1): 23-31.
20	DingF, ChenT. Identification of Hammerstein nonlinear ARMAX systems[J]. Automatica, 2005, 41(9): 1479-1489.
21	GrbicR, ScitovskiK, SaboK, et al. Approximating surfaces by the moving least absolute deviations method[J]. Applied Mathematics & Computation, 2013, 219(9): 4387-4399.
22	KelkinnamaM, TaheriS M. Fuzzy least-absolutes regression using shape preserving operations[J]. Information Sciences, 2012, 214(10): 105-120.
23	DongJ, XieK G. Research of the non-linear regress models based on the least absolute criteria[J]. Journal of Chongqing Normal University (Natural Science Edition), 2001, 18(4): 71-74.
24	徐宝昌, 林忠华, 肖玉月. 基于近似偏小一乘的闭环系统辨识新方法[J]. 控制理论与应用, 2016, 33(11): 1543-1551.
	XuB C, LinZ H, XiaoY Y. Identification of closed-loop system by partial least absolute deviation[J]. Control Theory & Applications, 2016, 33(11): 1543-1551.
25	徐宝昌, 张华, 王学敏. 基于近似偏最小一乘准则的多变量非线性系统辨识方法[J]. 化工学报, 2018, 69(3): 1129-1135.
	XuB C, ZhangH, WangX M. Partial approximate least absolute deviation for multivariable nonlinear system identification[J]. CIESC Journal, 2018, 69(3): 1129-1135.
26	王桂增, 叶昊. 主元分析与偏最小二乘法[M]. 北京: 清华大学出版社, 2012.
	WangG Z, YeH. Principal Component Analysis and Partial Least Squares Method[M]. Beijing: Tsinghua University Press, 2012.
27	KangS H, SandbergB, YipA M. A regularized k-means and multiphase scale segmentation[J]. Inverse Problems & Imaging, 2017, 5(2): 407-429.
28	KamperH, LivescuK, GoldwaterS. An embedded segmental k-means model for unsupervised segmentation and clustering of speech[C]//IEEE Automatic Speech Recognition and Understanding Workshop. IEEE, 2017: 719-726.
29	HlaingY M, ChiuM S, LakshminarayananS. Modeling and control of multivariable process using generalized Hammerstein model[J]. Chemical Engineering Research & Design, 2007, 85(4): 445-454.
30	CoatesM J, KuruogluE E. Time-frequency based detection in impulsive noise environments using alpha-stable noise model[J]. Digital Signal Processing, 2002, 82(3): 1917-1925.

[1]	罗顺桦, 王振雷, 王昕. 基于二子空间协同训练算法的半监督软测量建模[J]. 化工学报, 2022, 73(3): 1270-1279.
[2]	周红标, 张钰, 柏小颖, 刘保连, 赵环宇. 基于自适应模糊神经网络的非线性系统模型预测控制[J]. 化工学报, 2020, 71(7): 3201-3212.
[3]	于仙毅, 巫江虹, 高云辉. 基于主成分分析与支持向量机的热泵系统制冷剂泄漏识别研究[J]. 化工学报, 2020, 71(7): 3151-3164.
[4]	柴伟, 郭龙航, 池彬彬. 污水处理厂出水水质变量区间预测建模[J]. 化工学报, 2019, 70(9): 3449-3457.
[5]	廉小亲, 王俐伟, 安飒, 魏伟, 刘载文. 基于SOM-RBF神经网络的COD软测量方法[J]. 化工学报, 2019, 70(9): 3465-3472.
[6]	唐俊苗, 俞海珍, 史旭华, 童楚东. 基于潜变量自回归算法的化工过程动态监测方法[J]. 化工学报, 2019, 70(3): 987-994.
[7]	王志甄, 邹志云. 基于神经网络的pH中和过程非线性预测控制[J]. 化工学报, 2019, 70(2): 678-686.
[8]	徐宝昌, 张华, 王学敏. 基于近似偏最小一乘准则的多变量非线性系统辨识方法[J]. 化工学报, 2018, 69(3): 1129-1135.
[9]	周丽春, 靳鑫, 刘毅, 高增梁, 金福江. 即时局部建模在填料塔液泛气速预测的应用[J]. 化工学报, 2016, 67(3): 1070-1075.
[10]	韩敏, 张占奎. 基于改进核主成分分析的故障检测与诊断方法[J]. 化工学报, 2015, 66(6): 2139-2149.
[11]	周丽春, 刘毅, 金福江. 一种非线性系统在线辨识的选择性递推方法[J]. 化工学报, 2015, 66(1): 272-277.
[12]	冯凯, 卢建刚, 陈金水. 基于最小二乘支持向量机的MIMO线性参数变化模型辨识及预测控制[J]. 化工学报, 2015, 66(1): 197-205.
[13]	许锋, 汪晔晔, 罗雄麟. 基于广义状态观测器的催化裂化装置软仪表[J]. 化工学报, 2013, 64(5): 1683-1695.
[14]	贾润达, 毛志忠, 王福利. 基于KPLS模型的间歇过程产品质量控制[J]. 化工学报, 2013, 64(4): 1332-1339.
[15]	李东娟. 连续搅拌反应釜的自适应神经网络控制[J]. 化工学报, 2013, 64(12): 4674-4680.