CN108153147B - 一种无源性的分析方法 - Google Patents

一种无源性的分析方法 Download PDF

Info

Publication number
CN108153147B
CN108153147B CN201711444587.5A CN201711444587A CN108153147B CN 108153147 B CN108153147 B CN 108153147B CN 201711444587 A CN201711444587 A CN 201711444587A CN 108153147 B CN108153147 B CN 108153147B
Authority
CN
China
Prior art keywords
time lag
passivity
follows
neural network
network system
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201711444587.5A
Other languages
English (en)
Other versions
CN108153147A (zh
Inventor
董宏丽
李佳慧
王子栋
韩非
高宏宇
张勇
路阳
宋金波
王婷婷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northeast Petroleum University
Original Assignee
Northeast Petroleum University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Northeast Petroleum University filed Critical Northeast Petroleum University
Priority to CN201711444587.5A priority Critical patent/CN108153147B/zh
Publication of CN108153147A publication Critical patent/CN108153147A/zh
Application granted granted Critical
Publication of CN108153147B publication Critical patent/CN108153147B/zh
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G05CONTROLLING; REGULATING
    • G05BCONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/0265Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
    • G05B13/027Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion using neural networks only
    • GPHYSICS
    • G05CONTROLLING; REGULATING
    • G05BCONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance

Landscapes

  • Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Complex Calculations (AREA)

Abstract

本发明提出一种无源性的分析方法,在其中考虑了随机发生混合时滞、随机干扰和参数不确定对无源性分析的影响,构造李亚普诺夫函数完整利用了时滞的有效信息,相比于已有的神经网络动态***的无源性分析方法,本发明的分析方法能够一并考虑随机发生的混合时滞、随机干扰和参数不确定,得出了依赖于线性矩阵不等式解的无源性分析方法,达到抗非线性扰动的目的,本发明适用于非线性复杂动态***的无源性分析。

Description

一种无源性的分析方法
技术领域
本发明属于控制领域,涉及一种具有随机发生混合时滞的随机神经网络***的无源性/鲁棒无源性的分析方法,本发明适用于非线性复杂动态***的无源性分析。
背景技术
人工神经网络作为一门交叉学科,在许多领域得到应用,研究它的发展过程和前沿问题具有重要的理论意义。由于神经网络具有信息的分布存储,并行处理和自学习能力等优点,所以在信息处理,模式识别智能控制等领域有着广泛的应用前景。一个神经网络***若想在工程中发挥作用就必须具备稳定性,因此在研究神经网络的核心问题中,稳定性分析是极为重要的且必不可少的一个环节。在稳定性分析过程中,需要构造合适的李雅普诺夫函数,但其实际可行的构造方法并不多。而无源***的存储函数在一定条件下便可以作为李雅普诺夫函数,所以神经网络***的无源性分析问题具有广阔的应用前景。
由于有限的信息处理速度,在实际控制***,尤其在一些生态***和神经网络中,时滞是常见的,而时其发生的不确定性更是令人困扰的问题。由于时滞的产生可能会降低***的质量,甚至导致振荡甚至不稳定。因此,对时滞***的无源性研究具有理论和实际意义。
发明内容
本发明为了解决现有无源性分析方法不能同时处理随机发生的混合时滞、随机干扰和参数不确定,进而影响***无源性分析的问题,提出一种无源性的分析方法。
本发明的目的通过以下技术方案实现:一种无源性的分析方法,包括以下步骤:
步骤一、建立具有随机发生混合时滞和随机干扰的神经网络***动态模型;
步骤二、根据所述动态模型,建立综合考虑随机发生混合时滞与随机干扰的无源性性能指标;
步骤三、根据所述无源性性能指标构建李雅普诺夫函数,进而求出李雅普诺夫函数差分的期望;
步骤四、根据步骤三中结果及神经网络***中相关约束条件,建立不等式关系;
步骤五、根据无源性判据,对具有随机发生混合时滞与随机干扰的神经网络***进行无源性分析;
步骤六、建立具有参数不确定、随机发生混合时滞和随机干扰的神经网络***的动态模型;
步骤七:根据步骤六建立的动态模型,重复步骤二至步骤四,根据鲁棒无源性判据,对具有参数不确定、随机发生混合时滞和随机干扰的神经网络***进行鲁棒无源性分析。
进一步地,所述步骤一具体为:
建立具有随机发生混合时滞和随机干扰的神经网络***动态模型,其***状态空间形式为:
Figure BDA0001527294430000021
式中,
Figure BDA0001527294430000022
表示神经网络***的状态向量,u(k)为***的输入向量,
Figure BDA0001527294430000023
Figure BDA0001527294430000024
Figure BDA0001527294430000025
皆为非线性激励函数,y(k)为神经网络的测量输出,ω(k)为标准高斯噪声序列,
Figure BDA0001527294430000026
为扩散系数,τ(k)为时变时滞,满足
Figure BDA0001527294430000027
d为无穷分布时滞,τm(k)为有限分布时滞,满足dmin≤τm(k)≤dmax,A=diag{a1,a2,…,an}为实对角常矩阵,B=(bij)n×n为连接权值矩阵,C=(cij)n×n、D=(dij)n×n及E=(eij)n×n皆为时滞连接权值矩阵;
Figure BDA0001527294430000028
为相互独立的伯努利分布白序列,概率分布为
Figure BDA0001527294430000029
Figure BDA00015272944300000210
其中
Figure BDA00015272944300000211
Figure BDA00015272944300000212
为已知非负常数,q为已知正整数,常数μd满足下列条件:
Figure BDA00015272944300000213
非线性激励函数连续有界且满足以下约束条件:
Figure BDA00015272944300000214
Figure BDA00015272944300000215
Figure BDA00015272944300000216
式中,Gj +,Gj -,Fj +,Fj -,Hj +,Hj -为常数,gj(·),fj(·),hj(·)为第j个神经元的激励函数。
进一步地,所述步骤二具体为:
建立无源性性能指标如下:
Figure BDA0001527294430000031
式中
Figure BDA0001527294430000032
γ>0为待求标量。
进一步地,所述步骤三具体为:
构建李雅普诺夫函数如下:
Figure BDA0001527294430000033
其中:V1(k,x(k))=xT(k)Px(k),
Figure BDA0001527294430000034
Figure BDA0001527294430000035
Figure BDA0001527294430000036
Figure BDA0001527294430000037
式中P,Q1,Q2,Q3,Z1,Z2m为待求正定矩阵;
所述李雅普诺夫函数的差分的期望为:
Figure BDA0001527294430000038
式中,η(k)=x(k+1)-x(k),ΔV(k,x(k))=V(k+1,x(k+1))-V(k,x(k)),V(k,x(k))为k时刻的李雅普诺夫函数,ΔV(k,x(k))为k时刻的李雅普诺夫函数的差分,xT(k)为x(k)的转置,ηT(j)为η(j)的转置,fT(x(j))为f(x(j))的转置,hT(x(i))为h(x(i))的转置。
进一步地,所述步骤四具体为:
建立不等式关系如下:
Figure BDA0001527294430000041
式中:
Figure BDA0001527294430000042
Figure BDA0001527294430000043
Figure BDA0001527294430000044
Figure BDA0001527294430000045
Figure BDA0001527294430000046
其中:
Figure BDA0001527294430000047
Figure BDA0001527294430000048
Figure BDA0001527294430000049
Figure BDA00015272944300000410
Π22=BTPB+τBTZ1B-M,
Figure BDA00015272944300000411
Figure BDA0001527294430000051
Π26=BTP+τBTZ1-I,Π29=R2-S2,
Figure BDA0001527294430000052
Figure BDA0001527294430000053
Π39=R3-S3+G2N,
Figure BDA0001527294430000054
Figure BDA0001527294430000055
Π49=R4-S4,
Figure BDA0001527294430000056
Figure BDA0001527294430000057
Π59=R5-S566=P+τZ1-γI,Π69=R6-S6,
Figure BDA0001527294430000058
Π79=R7-S7,
Figure BDA0001527294430000059
Π89=R8-S8,
Figure BDA00015272944300000510
Figure BDA00015272944300000511
Figure BDA00015272944300000512
Figure BDA00015272944300000513
Figure BDA00015272944300000514
Ri,Si,Wi(i=1,…,9)为待求矩阵,M,N,U,V为待求正定矩阵,λ为待求标量,ρ1,ρ2为标量,
Figure BDA00015272944300000516
为矩阵Z1的逆矩阵。
进一步地,所述步骤五具体为:
所述无源性判据如下:
τZ1+P-λI≤0, (6)
Figure BDA00015272944300000515
式中:
Figure BDA0001527294430000061
Figure BDA0001527294430000062
Π3=diag{-Q1,-Q3,-Z1,-Z1,-Z1}
对具有随机发生混合时滞与随机干扰的神经网络***进行无源性分析:应用舒尔补引理于(7)式,可得:
Figure BDA0001527294430000063
综合(5)式可得
Figure BDA0001527294430000064
根据V(k,x(k))的定义,下式成立:
Figure BDA0001527294430000065
所以
Figure BDA0001527294430000066
成立,满足(2)式无源性指标,即若(6)-(7)式有可行解,证明***无源,否则,说明***并不是无源的。
进一步地,所述步骤六具体为:
建立具有参数不确定、随机发生混合时滞和随机干扰的神经网络***的动态模型如下:
Figure BDA0001527294430000071
式中:K,A,B,C,D,E和Ni(i=1,2,3,4,5)为已知常矩阵,ΔA,ΔB,ΔC,ΔD,ΔE为时变矩阵且满足:[ΔA ΔB ΔC ΔD ΔE]=KF(k)[N1 N2 N3 N4 N5],F(k)为未知时变值函数,满足FT(k)F(k)≤I。
进一步地,所述步骤七具体为:
所述鲁棒无源性判据如下:
τZ1+P-λI≤0, (9)
Figure BDA0001527294430000072
其中,ε>0为待求标量,
Figure BDA0001527294430000073
Figure BDA0001527294430000074
Figure BDA0001527294430000075
Figure BDA0001527294430000081
Figure BDA0001527294430000082
Figure BDA0001527294430000083
Σ22=diag{-Q1,-Q3,-Z1,-Z1,-Z1},
Figure BDA0001527294430000084
Figure BDA0001527294430000085
Figure BDA0001527294430000091
Figure BDA0001527294430000092
Figure BDA0001527294430000093
Figure BDA0001527294430000094
Λ22=-M,Λ26=-I,Λ29=R2-S233=-N,Λ39=R3-S3+G2N, Λ49=R4-S459=R5-S566=-γI,Λ69=R6-S679 =R7-S789=R8-S8,
Figure BDA00015272944300000910
若线性矩阵不等式(9)-(10)有可行解,则证明具有参数不确定、随机发生混合时滞和随机干扰的神经网络***是鲁棒无源的,否则证明***并不是鲁棒无源的。
本发明提出一种无源性的分析方法,在其中考虑了随机发生混合时滞、随机干扰和参数不确定对无源性分析的影响,构造李亚普诺夫函数完整利用了时滞的有效信息,相比于已有的神经网络动态***的无源性分析方法,本发明的分析方法能够一并考虑随机发生的混合时滞、随机干扰和参数不确定,得出了依赖于线性矩阵不等式解的无源性分析方法,达到抗非线性扰动的目的,并且该分析方法方便求解,易于实现。
附图说明
图1为本发明所述方法流程图;
图2是在
Figure BDA00015272944300000911
情形下,神经网络***的状态变化曲线x1(k)和x2(k);
图3是在
Figure BDA00015272944300000912
情形下,神经网络***的状态变化曲线x1(k)和x2(k);
图4是在
Figure BDA00015272944300000913
情形下,神经网络***的状态变化曲线x1(k)和x2(k);
具体实施方式
下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明的一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。
符号说明:
本发明中,MT表示矩阵M的转置。Rn表示n维欧几里得空间,Rn×m表示所有n×m阶实矩阵的集合。I和0分别表示单位矩阵、零矩阵。P>0表示一系列实对称正定矩阵。E{x}代表随机变量x的数学期望。||x||代表向量x的欧几里得范数。diag{A1,A2,…,An}表示对角块是矩阵A1,A2,…,An的块对角矩阵,符号*在对称块矩阵中表示对称项的省略。如果M为对称矩阵,λmax(M)表示矩阵M的最大特征值。若发明中某处没有明确指定矩阵维数,则假定其维数适合矩阵的代数运算。
结合图1,一种无源性的分析方法,包括以下步骤:
步骤一、建立具有随机发生混合时滞和随机干扰的神经网络***动态模型;
步骤二、根据所述动态模型,建立综合考虑随机发生混合时滞与随机干扰的无源性性能指标;
步骤三、根据所述无源性性能指标构建李雅普诺夫函数,进而求出李雅普诺夫函数差分的期望;
步骤四、根据步骤三中结果及神经网络***中相关约束条件,建立不等式关系;
步骤五、根据无源性判据,对具有随机发生混合时滞与随机干扰的神经网络***进行无源性分析;
步骤六、建立具有参数不确定、随机发生混合时滞和随机干扰的神经网络***的动态模型;
步骤七:根据步骤六建立的动态模型,重复步骤二至步骤四,根据鲁棒无源性判据,对具有参数不确定、随机发生混合时滞和随机干扰的神经网络***进行鲁棒无源性分析。
所述步骤一具体为:
建立具有随机发生混合时滞和随机干扰的神经网络***动态模型,其***状态空间形式为:
Figure BDA0001527294430000101
式中,
Figure BDA0001527294430000102
表示神经网络***的状态向量,u(k)为***的输入向量,
Figure BDA0001527294430000103
Figure BDA0001527294430000104
Figure BDA0001527294430000105
皆为非线性激励函数,y(k)为神经网络的测量输出,ω(k)为标准高斯噪声序列,
Figure BDA0001527294430000106
为扩散系数,τ(k)为时变时滞,满足
Figure BDA0001527294430000107
d为无穷分布时滞,τm(k)为有限分布时滞,满足dmin≤τm(k)≤dmax,A=diag{a1,a2,...,an}为实对角常矩阵,B=(bij)n×n为连接权值矩阵,C=(cij)n×n、D=(dij)n×n及E=(eij)n×n皆为时滞连接权值矩阵;
Figure BDA0001527294430000111
为相互独立的伯努利分布白序列,概率分布为
Figure BDA0001527294430000112
Figure BDA0001527294430000113
其中
Figure BDA0001527294430000114
Figure BDA0001527294430000115
为已知非负常数,q为已知正整数,常数μd满足下列条件:
Figure BDA0001527294430000116
非线性激励函数连续有界且满足以下约束条件:
Figure BDA0001527294430000117
Figure BDA0001527294430000118
Figure BDA0001527294430000119
式中,Gj +,Gj -,Fj +,Fj -,Hj +,Hj -为常数,gj(·),fj(·),hj(·)为第j个神经元的激励函数。
所述步骤二具体为:
建立无源性性能指标如下:
Figure BDA00015272944300001110
式中
Figure BDA00015272944300001111
γ>0为待求标量。
所述步骤三具体为:
构建李雅普诺夫函数如下:
Figure BDA00015272944300001112
其中:V1(k,x(k))=xT(k)Px(k),
Figure BDA00015272944300001113
Figure BDA00015272944300001114
Figure BDA00015272944300001115
Figure BDA0001527294430000121
式中P,Q1,Q2,Q3,Z1,Z2m为待求正定矩阵;
所述李雅普诺夫函数的差分的期望为:
Figure BDA0001527294430000122
式中,η(k)=x(k+1)-x(k),ΔV(k,x(k))=V(k+1,x(k+1))-V(k,x(k)),V(k,x(k))为k时刻的李雅普诺夫函数,ΔV(k,x(k))为k时刻的李雅普诺夫函数的差分,xT(k)为x(k)的转置,ηT(j)为η(j)的转置,fT(x(j))为f(x(j))的转置,hT(x(i))为h(x(i))的转置。
所述步骤四具体为:
建立不等式关系如下:
Figure BDA0001527294430000123
式中:
Figure BDA0001527294430000124
Figure BDA0001527294430000125
Figure BDA0001527294430000131
Figure BDA0001527294430000132
Figure BDA0001527294430000133
其中:
Figure BDA0001527294430000134
Figure BDA0001527294430000135
Figure BDA0001527294430000136
Figure BDA0001527294430000137
Π22=BTPB+τBTZ1B-M,
Figure BDA0001527294430000138
Figure BDA0001527294430000139
Π26=BTP+τBTZ1-I,Π29=R2-S2,
Figure BDA00015272944300001310
Figure BDA00015272944300001311
Π39=R3-S3+G2N,
Figure BDA00015272944300001312
Figure BDA00015272944300001313
Π49=R4-S4,
Figure BDA00015272944300001314
Figure BDA00015272944300001315
Π59=R5-S566=P+τZ1-γI,Π69=R6-S6,
Figure BDA00015272944300001316
Π79=R7-S7,
Figure BDA00015272944300001317
Π89=R8-S8,
Figure BDA00015272944300001318
Figure BDA00015272944300001319
Figure BDA00015272944300001320
Figure BDA00015272944300001321
Figure BDA00015272944300001322
Ri,Si,Wi(i=1,…,9)为待求矩阵,M,N,U,V为待求正定矩阵,λ为待求标量,ρ1,ρ2为标量,
Figure BDA0001527294430000145
为矩阵Z1的逆矩阵。
所述步骤五具体为:
所述无源性判据如下:
τZ1+P-λI≤0, (6)
Figure BDA0001527294430000141
式中:
Figure BDA0001527294430000142
Figure BDA0001527294430000143
Π3=diag{-Q1,-Q3,-Z1,-Z1,-Z1}。
对具有随机发生混合时滞与随机干扰的神经网络***进行无源性分析:
应用舒尔补引理于(7)式,可得:
Figure BDA0001527294430000144
综合(5)式可得
Figure BDA0001527294430000151
根据V(k,x(k))的定义,下式成立:
Figure BDA0001527294430000152
所以
Figure BDA0001527294430000153
成立,满足(2)式无源性指标,即若(6)-(7)式有可行解,证明***无源,否则,说明***并不是无源的。
所述步骤六具体为:
建立具有参数不确定、随机发生混合时滞和随机干扰的神经网络***的动态模型如下:
Figure BDA0001527294430000154
式中:K,A,B,C,D,E和Ni(i=1,2,3,4,5)为已知常矩阵,ΔA,ΔB,ΔC,ΔD,ΔE为时变矩阵且满足:[ΔA ΔB ΔC ΔD ΔE]=KF(k)[N1 N2 N3 N4 N5],F(k)为未知时变值函数,满足FT(k)F(k)≤I。
所述步骤七具体为:
所述鲁棒无源性判据如下:
τZ1+P-λI≤0, (9)
Figure BDA0001527294430000155
其中,ε>0为待求标量,
Figure BDA0001527294430000161
Figure BDA0001527294430000162
Figure BDA0001527294430000163
Figure BDA0001527294430000164
Figure BDA0001527294430000165
Figure BDA0001527294430000166
Σ22=diag{-Q1,-Q3,-Z1,-Z1,-Z1},
Figure BDA0001527294430000171
Figure BDA0001527294430000172
Figure BDA0001527294430000173
Figure BDA0001527294430000174
Figure BDA0001527294430000175
Figure BDA0001527294430000176
Figure BDA0001527294430000177
Λ22=-M,Λ26=-I,Λ29=R2-S233=-N,Λ39=R3-S3+G2N,
Figure BDA0001527294430000178
Λ49=R4-S4,
Figure BDA0001527294430000179
Λ59=R5-S566=-γI,Λ69=R6-S6,
Figure BDA00015272944300001710
Λ79=R7-S7,
Figure BDA00015272944300001711
Λ89=R8-S8,
Figure BDA00015272944300001712
若线性矩阵不等式(9)-(10)有可行解,则证明具有参数不确定、随机发生混合时滞和随机干扰的神经网络***是鲁棒无源的,否则证明***并不是鲁棒无源的。
采用本发明所述方法进行仿真验证:
***参数选为:
Figure BDA00015272944300001713
Figure BDA00015272944300001714
ρ1=0.01,ρ2=0.02,
Figure BDA00015272944300001715
Figure BDA00015272944300001716
F1 +=0.1,F1 -=-0.1,
Figure BDA00015272944300001717
N1=0.3I,N2=0.5I,N3=0.6I,N4=0.1I,N5=0.2I,K=0.4I
非线性激励函数选为:
Figure BDA0001527294430000181
此外,时滞发生的概率为
Figure BDA0001527294430000182
随机干扰为σ(x(k),x(k-τ(k)),k)=diag{-0.02sin(y(k-τ(k)))|x1(k)|,0.03sin(x2(k))}。
对(9)式进行求解,得到使式(8)满足鲁棒无源的性能指标γ,其部分求解结果如下:
Figure BDA0001527294430000183
λ=72.0598,γ=118.8594,ε=6.7541。
仿真效果展示在图2、3、4中:
图2是在
Figure BDA0001527294430000184
情形下,神经网络***的状态变化曲线x1(k)和x2(k);图3是在
Figure BDA0001527294430000185
情形下,神经网络***的状态变化曲线x1(k)和x2(k);图4是在
Figure BDA0001527294430000186
情形下,神经网络***的状态变化曲线x1(k)和x2(k)。
Figure BDA0001527294430000187
时,说明***时滞完全发生,由图2可见,曲线几乎未靠近过平衡点,***性能较差;当
Figure BDA0001527294430000188
时,由图4可见,曲线频繁靠***衡点,***性能较好;当
Figure BDA0001527294430000189
时,由图3可见,***性能介于二者之间。综上所述,对于具有随机发生混合时滞的随机神经网络***,所发明的无源性分析方法是有效、可行的。
以上对本发明所提供的一种无源性的分析方法,进行了详细介绍,本文中应用了具体个例对本发明的原理及实施方式进行了阐述,以上实施例的说明只是用于帮助理解本发明的方法及其核心思想;同时,对于本领域的一般技术人员,依据本发明的思想,在具体实施方式及应用范围上均会有改变之处,综上所述,本说明书内容不应理解为对本发明的限制。

Claims (1)

1.一种无源性的分析方法,其特征在于,包括以下步骤:
步骤一、建立具有随机发生混合时滞和随机干扰的神经网络***动态模型;
步骤二、根据所述动态模型,建立综合考虑随机发生混合时滞与随机干扰的无源性性能指标;
步骤三、根据所述无源性性能指标构建李雅普诺夫函数,进而求出李雅普诺夫函数差分的期望;
步骤四、根据步骤三中结果及神经网络***中相关约束条件,建立不等式关系;
步骤五、根据无源性判据,对具有随机发生混合时滞与随机干扰的神经网络***进行无源性分析;
步骤六、建立具有参数不确定、随机发生混合时滞和随机干扰的神经网络***的动态模型;
步骤七:根据步骤六建立的动态模型,重复步骤二至步骤四,根据鲁棒无源性判据,对具有参数不确定、随机发生混合时滞和随机干扰的神经网络***进行鲁棒无源性分析;
所述步骤一具体为:
建立具有随机发生混合时滞和随机干扰的神经网络***动态模型,其***状态空间形式为:
Figure FDA0002842925550000011
式中,
Figure FDA0002842925550000012
表示神经网络***的状态向量,u(k)为***的输入向量,
Figure FDA0002842925550000013
Figure FDA0002842925550000014
Figure FDA0002842925550000015
皆为非线性激励函数,y(k)为神经网络的测量输出,ω(k)为标准高斯噪声序列,
Figure FDA0002842925550000016
为扩散系数,τ(k)为时变时滞,满足
Figure FDA0002842925550000017
d为无穷分布时滞,τm(k)为有限分布时滞,满足dmin≤τm(k)≤dmax,A=diag{a1,a2,...,an}为实对角常矩阵,B=(bij)n×n为连接权值矩阵,C=(cij)n×n、D=(dij)n×n及E=(eij)n×n皆为时滞连接权值矩阵;
Figure FDA0002842925550000018
为相互独立的伯努利分布白序列,概率分布为
Figure FDA0002842925550000019
Figure FDA00028429255500000110
其中
Figure FDA00028429255500000111
Figure FDA00028429255500000112
为已知非负常数,q为已知正整数,常数μd满足下列条件:
Figure FDA0002842925550000021
非线性激励函数连续有界且满足以下约束条件:
Figure FDA0002842925550000022
Figure FDA0002842925550000023
Figure FDA0002842925550000024
式中,Gj +,Gj -,Fj +,Fj -,Hj +,Hj -为常数,gj(·),fj(·),hj(·)为第j个神经元的激励函数;
所述步骤二具体为:
建立无源性性能指标如下:
Figure FDA0002842925550000025
式中
Figure FDA0002842925550000026
γ>0为待求标量;
所述步骤三具体为:
构建李雅普诺夫函数如下:
Figure FDA0002842925550000027
其中:V1(k,x(k))=xT(k)Px(k),
Figure FDA0002842925550000028
Figure FDA0002842925550000029
Figure FDA00028429255500000210
Figure FDA00028429255500000211
式中P,Q1,Q2,Q3,Z1,Z2m为待求正定矩阵;
所述李雅普诺夫函数的差分的期望为:
Figure FDA00028429255500000212
式中,η(k)=x(k+1)-x(k),ΔV(k,x(k))=V(k+1,x(k+1))-V(k,x(k)),V(k,x(k))为k时刻的李雅普诺夫函数,ΔV(k,x(k))为k时刻的李雅普诺夫函数的差分,xT(k)为x(k)的转置,ηT(j)为η(j)的转置,fT(x(j))为f(x(j))的转置,hT(x(i))为h(x(i))的转置;
所述步骤四具体为:
建立不等式关系如下:
Figure FDA0002842925550000031
式中:
Figure FDA0002842925550000032
Figure FDA0002842925550000033
Figure FDA0002842925550000034
Figure FDA0002842925550000035
Figure FDA0002842925550000036
其中:符号*在对称块矩阵中表示对称项的省略;
Figure FDA0002842925550000037
Figure FDA0002842925550000041
Figure FDA0002842925550000042
Figure FDA0002842925550000043
Figure FDA0002842925550000044
Figure FDA0002842925550000045
Π26=BTP+τBTZ1-I,Π29=R2-S2,
Figure FDA0002842925550000046
Figure FDA0002842925550000047
Figure FDA0002842925550000048
Figure FDA0002842925550000049
Π49=R4-S4,
Figure FDA00028429255500000410
Figure FDA00028429255500000411
Π59=R5-S566=P+τZ1-γI,Π69=R6-S6,
Figure FDA00028429255500000412
Π79=R7-S7,
Figure FDA00028429255500000413
Π89=R8-S8,
Figure FDA00028429255500000414
Figure FDA00028429255500000415
Figure FDA00028429255500000416
Figure FDA00028429255500000417
Figure FDA00028429255500000418
Ri,Si,Wi为待求矩阵,i=1,…,9;M,N,U,V为待求正定矩阵,λ为待求标量,ρ1,ρ2为正标量,
Figure FDA00028429255500000419
为矩阵Z1的逆矩阵;
所述步骤五具体为:
所述无源性判据如下:
τZ1+P-λI≤0, (6)
Figure FDA0002842925550000051
式中:
Figure FDA0002842925550000052
Figure FDA0002842925550000053
Π3=diag{-Q1,-Q3,-Z1,-Z1,-Z1}
对具有随机发生混合时滞与随机干扰的神经网络***进行无源性分析:应用舒尔补引理于(7)式,可得:
Figure FDA0002842925550000054
综合(5)式可得
Figure FDA0002842925550000055
根据V(k,x(k))的定义,下式成立:
Figure FDA0002842925550000056
所以
Figure FDA0002842925550000061
成立,满足(2)式无源性指标,即若(6)-(7)式有可行解,证明***无源,否则,说明***并不是无源的;
所述步骤六具体为:建立具有参数不确定、随机发生混合时滞和随机干扰的神经网络***的动态模型如下:
Figure FDA0002842925550000062
式中:K,A,B,C,D,E和Ni为已知常矩阵,i=1,2,3,4,5;ΔA,ΔB,ΔC,ΔD,ΔE为时变矩阵且满足:[ΔA ΔB ΔC ΔD ΔE]=KF(k)[N1 N2 N3 N4 N5],F(k)为未知时变值函数,满足FT(k)F(k)≤I;
所述步骤七具体为:
所述鲁棒无源性判据如下:
τZ1+P-λI≤0, (9)
Figure FDA0002842925550000063
其中,ε>0为待求标量,
Figure FDA0002842925550000064
Figure FDA0002842925550000065
Figure FDA0002842925550000066
Figure FDA0002842925550000071
Figure FDA0002842925550000072
Figure FDA0002842925550000073
Σ22=diag{-Q1,-Q3,-Z1,-Z1,-Z1},
Figure FDA0002842925550000074
Figure FDA0002842925550000075
Figure FDA0002842925550000081
Figure FDA0002842925550000082
Figure FDA0002842925550000083
Figure FDA0002842925550000084
Figure FDA0002842925550000085
Λ22=-M,Λ26=-I,Λ29=R2-S233=-N,Λ39=R3-S3+G2N,
Figure FDA0002842925550000086
Λ49=R4-S4,
Figure FDA0002842925550000087
Λ59=R5-S566=-γI,Λ69=R6-S6,
Figure FDA0002842925550000088
Λ79=R7-S7,
Figure FDA0002842925550000089
Λ89=R8-S8,
Figure FDA00028429255500000810
若线性矩阵不等式(9)-(10)有可行解,则证明具有参数不确定、随机发生混合时滞和随机干扰的神经网络***是鲁棒无源的,否则证明***并不是鲁棒无源的。
CN201711444587.5A 2017-12-27 2017-12-27 一种无源性的分析方法 Active CN108153147B (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711444587.5A CN108153147B (zh) 2017-12-27 2017-12-27 一种无源性的分析方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711444587.5A CN108153147B (zh) 2017-12-27 2017-12-27 一种无源性的分析方法

Publications (2)

Publication Number Publication Date
CN108153147A CN108153147A (zh) 2018-06-12
CN108153147B true CN108153147B (zh) 2021-03-12

Family

ID=62463259

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711444587.5A Active CN108153147B (zh) 2017-12-27 2017-12-27 一种无源性的分析方法

Country Status (1)

Country Link
CN (1) CN108153147B (zh)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6064997A (en) * 1997-03-19 2000-05-16 University Of Texas System, The Board Of Regents Discrete-time tuning of neural network controllers for nonlinear dynamical systems
EP2447792A1 (en) * 2005-09-19 2012-05-02 Cleveland State University Controllers, observer, and applications thereof
CN102915385A (zh) * 2011-08-03 2013-02-06 复旦大学 一种基于时域梯形法差分的互连线模型降阶方法
CN104730921A (zh) * 2015-01-13 2015-06-24 河海大学常州校区 基于终端滑模的有源电力滤波器模糊神经网络控制方法
US9146557B1 (en) * 2014-04-23 2015-09-29 King Fahd University Of Petroleum And Minerals Adaptive control method for unmanned vehicle with slung load
CN106875009A (zh) * 2017-03-03 2017-06-20 深圳市唯特视科技有限公司 一种基于人工神经网络的混沌控制方法
CN107065555A (zh) * 2017-04-27 2017-08-18 成都理工大学 模糊随机大***的稳定性分析方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6064997A (en) * 1997-03-19 2000-05-16 University Of Texas System, The Board Of Regents Discrete-time tuning of neural network controllers for nonlinear dynamical systems
EP2447792A1 (en) * 2005-09-19 2012-05-02 Cleveland State University Controllers, observer, and applications thereof
CN102915385A (zh) * 2011-08-03 2013-02-06 复旦大学 一种基于时域梯形法差分的互连线模型降阶方法
US9146557B1 (en) * 2014-04-23 2015-09-29 King Fahd University Of Petroleum And Minerals Adaptive control method for unmanned vehicle with slung load
CN104730921A (zh) * 2015-01-13 2015-06-24 河海大学常州校区 基于终端滑模的有源电力滤波器模糊神经网络控制方法
CN106875009A (zh) * 2017-03-03 2017-06-20 深圳市唯特视科技有限公司 一种基于人工神经网络的混沌控制方法
CN107065555A (zh) * 2017-04-27 2017-08-18 成都理工大学 模糊随机大***的稳定性分析方法

Non-Patent Citations (8)

* Cited by examiner, † Cited by third party
Title
Hongyi Li 等.New passivity results for uncertain discrete-time stochastic neural networks with mixed time delays.《Neurocomputing》.2010,第73卷 *
Liangliang Li 等.Delay-dependent passivity analysis of impulsive neural networks with time-varying delays.《Neurocomputing》.2015,第168卷 *
M.Syed Ali 等.Passivity analysis of stochastic neural networks with leakage delay and Markovian jumping parameters.《Neu}ocomputing》.2016, *
彭诗友.几类时变时滞神经网络的无源性分析.《中国优秀硕士学位论文全文数据库信息科技辑》.2016,(第11期), *
朱进.递归神经网络的稳定性和无源性分析.《中国博士学位论文全文数据库信息科技辑》.2013,(第3期), *
沈坷婷.几类时滞神经网络的无源性分析.《中国优秀硕士学位论文全文数据库信息科技辑》.2012,(第7期), *
牟少帅.时滞***的稳定性分析.《中国优秀硕士学位论文全文数据库信息科技辑》.2012,(第7期), *
邵雪莹.几类时滞不确定神经网络的稳定性分析.《中国优秀硕士学位论文全文数据库基础科学辑》.2016,(第1期), *

Also Published As

Publication number Publication date
CN108153147A (zh) 2018-06-12

Similar Documents

Publication Publication Date Title
Yang et al. Distributed optimization based on a multiagent system in the presence of communication delays
Chen et al. Distributed Pareto optimization via diffusion strategies
Sakthivel et al. Synchronization and state estimation for stochastic complex networks with uncertain inner coupling
Wang et al. Stability of recurrent neural networks with time-varying delay via flexible terminal method
CN109088749B (zh) 一种随机通讯协议下复杂网络的状态估计方法
CN111812980B (zh) 基于未知输入观测器的离散切换***的鲁棒故障估计方法
Duan et al. Decentralized adaptive NN state-feedback control for large-scale stochastic high-order nonlinear systems
Jami'In et al. Deep searching for parameter estimation of the linear time invariant (LTI) system by using quasi-ARX neural network
Wu et al. Moment exponential stability of random delay systems with two-time-scale Markovian switching
Huang et al. Analysis and pinning control for generalized synchronization of delayed coupled neural networks with different dimensional nodes
Wang et al. Synchronization of generally uncertain Markovian inertial neural networks with random connection weight strengths and image encryption application
Yang et al. Resilient state estimation for nonlinear complex networks with time-delay under stochastic communication protocol
CN109688024A (zh) 基于随机通信协议的复杂网络弹性状态估计方法
CN108153147B (zh) 一种无源性的分析方法
Wang et al. Passivity of memristive BAM neural networks with leakage and additive time-varying delays
Duan et al. Optimization landscape of policy gradient methods for discrete-time static output feedback
Lee et al. Utility of edge-wise funnel coupling for asymptotically solving distributed consensus optimization
Sadamoto et al. Low-dimensional functional observer design for linear systems via observer reduction approach
Nekhoroshikh et al. On simple design of a robust finite-time observer
CN113820954A (zh) 一种广义噪声下复杂非线性***的容错控制方法
Wang et al. Robust Admissibilization for Discrete‐Time Singular Systems with Time‐Varying Delay
Ning et al. Robust decentralized H∞ control of multi-channel systems with norm-bounded parametric uncertainties
Wang et al. Multi-task total least-squares adaptation over networks
CN111339489A (zh) 一种有限域情形下多智能体***的控制器设计方法
Sun et al. Multiple delay-dependent guaranteed cost control for uncertain switched random nonlinear systems against intermittent sensor and actuator faults

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant