WO2022222229A1 - 耦合神经网络有界聚类投影同步调节控制方法及系统 - Google Patents

耦合神经网络有界聚类投影同步调节控制方法及系统 Download PDF

Info

Publication number
WO2022222229A1
WO2022222229A1 PCT/CN2021/097087 CN2021097087W WO2022222229A1 WO 2022222229 A1 WO2022222229 A1 WO 2022222229A1 CN 2021097087 W CN2021097087 W CN 2021097087W WO 2022222229 A1 WO2022222229 A1 WO 2022222229A1
Authority
WO
WIPO (PCT)
Prior art keywords
neural network
coupled
cluster
coupled neural
bounded
Prior art date
Application number
PCT/CN2021/097087
Other languages
English (en)
French (fr)
Inventor
汤泽
蒋晨辉
王艳
纪志成
Original Assignee
江南大学
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 江南大学 filed Critical 江南大学
Publication of WO2022222229A1 publication Critical patent/WO2022222229A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/06Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/061Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/23Clustering techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/06Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/063Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods

Definitions

  • the invention relates to the technical field of complex network synchronization control, in particular to a coupled neural network bounded clustering projection synchronization adjustment control method and system.
  • the projection synchronization method is of great significance in practical applications due to the different proportional coefficients of each node.
  • multi-ary communication achieves faster communication through projection synchronization.
  • Bounded synchronization is a special kind of synchronization, which is generally caused by factors such as parameter mismatch, internal interference, and external attacks. The system only needs to be consistent within a certain range.
  • cluster synchronization as a common synchronization phenomenon, describes the synchronization phenomenon that occurs in each subgroup.
  • a non-identical coupled neural network contains neural networks with different internal structures, and can also be divided into different clusters accordingly.
  • neural network is often used in the reasoning mechanism of decision support system, and the optimal solution is obtained through the sample information data according to the reasoning strategy generated by the neural network.
  • the problem of synchronization of coupled neural networks in the case of non-identity and parameter mismatch is rarely discussed. Due to the difference of the connection weight matrices of different neural networks, it is necessary to consider the characteristics of parameter mismatch.
  • the distributed control of the existing coupled neural network is more expensive than the pinning control, and the general cluster pinning control strategy aims to eliminate the interaction between different clusters and control the key nodes connecting multiple clusters. This type of control strategy is inefficient, has few applicable scenarios and cannot guarantee effective control.
  • the technical problem to be solved by the present invention is to overcome the high cost of synchronous control in the prior art to complete the non-identical coupled neural network control and the inability to perform effective control. .
  • the present invention provides a coupled neural network bounded clustering projection synchronization adjustment control method, including establishing a coupled neural network with multiple clusters of nonlinear, non-identical and mixed time-varying time delays, and for each Clustering to set a target neural network; establishing an error coupled neural network according to the coupled neural network and the target neural network; designing a pinned pulse feedback controller according to the error coupled neural network model, and selecting a corresponding function based on the pinned pulse feedback controller , in order to realize the bounded cluster projection synchronization between the neural network in each cluster and the target neural network; by building a network model and using the network model for numerical simulation, verify the existence of the target neural network and the coupled neural network. Boundary cluster projection synchronization effect.
  • the model of the coupled neural network is:
  • the model of the target neural network is:
  • the matrix measure ⁇ q (M) is defined:
  • I is an n-dimensional unit vector
  • q , q 1,2, ⁇ , represents the induced norm, and there is a constraint on the induced norm of the state vector of the target neural network: in is a normal number.
  • the model of the error coupled neural network is:
  • the expression of the restraint pulse feedback controller is:
  • the function is a Lyapunov function
  • the expression is:
  • the present invention also provides a coupled neural network bounded clustering projection synchronization adjustment control system applied to the above method, comprising: a building module for building a plurality of clusters with nonlinear, non-identical and mixed time-varying delays.
  • a coupled neural network, and a target neural network is set for each cluster; a setting module is used to establish an error coupled neural network according to the coupled neural network and the target neural network; a cluster synchronization module is used to according to the error coupled neural network
  • the network model designs a pinned impulse feedback controller, and selects a corresponding function based on the pinned impulse feedback controller to realize the bounded cluster projection synchronization between the neural network in each cluster and the target neural network; the simulation module is used to pass A network model is built and numerical simulation is performed using the network model to verify the effect of bounded cluster projection synchronization between the target neural network and the coupled neural network.
  • a pinned impulse controller is designed to realize the synchronization of coupled neural network clusters.
  • the controlled object feedback the error status information of the controlled object, so that effective control can be achieved with less consumption;
  • the value of the synchronization error bound is optimized.
  • the pulse intensity of the pinned pulse controller and the scale of the selected controlled object By controlling the pulse intensity of the pinned pulse controller and the scale of the selected controlled object, the existence of the synchronization error boundary can be ensured, and the synchronization error can be correspondingly reduced by adjusting the pulse frequency.
  • FIG. 1 is a structural diagram of a coupled neural network in Embodiment 2.
  • FIG. 1 is a structural diagram of a coupled neural network in Embodiment 2.
  • FIG. 2 is a pulse time sequence diagram in Embodiment 2.
  • FIG. 2 is a pulse time sequence diagram in Embodiment 2.
  • FIG. 3 is an effect diagram of the clustering pinning control strategy of the neural network in Embodiment 2.
  • FIG. 3 is an effect diagram of the clustering pinning control strategy of the neural network in Embodiment 2.
  • FIG. 4 is a graph showing the evolution of the error state of the neural network in Embodiment 2.
  • FIG. 5 is a state 1 evolution curve diagram of the neural network in cluster 1 in Example 2.
  • FIG. 5 is a state 1 evolution curve diagram of the neural network in cluster 1 in Example 2.
  • FIG. 6 is a state 2 evolution curve diagram of the neural network in cluster 1 in Example 2.
  • FIG. 6 is a state 2 evolution curve diagram of the neural network in cluster 1 in Example 2.
  • FIG. 7 is a state 1 evolution curve diagram of the neural network in cluster 2 in Example 2.
  • FIG. 7 is a state 1 evolution curve diagram of the neural network in cluster 2 in Example 2.
  • FIG. 8 is a state 2 evolution curve diagram of the neural network in cluster 2 in Example 2.
  • This embodiment provides a coupled neural network bounded clustering projection synchronization adjustment control method, including:
  • Step 1 Consider non-identical neural networks and divide them into multiple clusters, and set a target neural network for each sub-class.
  • the target neural network can be regarded as a leader, and other neural networks can be regarded as its followers By.
  • the following coupled neural network model with nonlinear, non-identical and mixed time-varying delays is established:
  • the matrices G and D satisfy the dissipation condition, that is, satisfy the and
  • u i (t) is the controller, which is designed in detail later.
  • the present invention introduces a matrix measure method, and defines the matrix measure ⁇ q (M) as follows
  • I is an n-dimensional unit vector,
  • q , q 1, 2, ⁇ represents the induced norm.
  • in is a normal number.
  • the normal number a is used to represent the projection factor, As a result, the following error-coupled neural network with nonlinearity, non-identity coupling and multiple delays is obtained:
  • the neural network synchronization problem in each cluster can be converted into a global stability problem of the error-coupled neural network, which is easier to handle.
  • Step 3 Assuming that the zth neural network in the coupled neural network is also the jth neural network sorted by the largest error in the ith cluster, define the symbol In order to realize the network synchronization between the neural network (Equation (1)) in each cluster and the target neural network (Equation (2)), based on the induced norm, each cluster is divided into two groups of neural networks according to the size of the state error. , grouped as follows:
  • (a) contains the larger induced norm in the ⁇ i -th cluster. a neural network, using to represent; (b) contains the ⁇ i -th cluster with a smaller induced norm a neural network, using To represent. selected
  • the internal neural network is controlled, and the state information of the error-coupled neural network is transmitted to it.
  • a pinned pulse feedback controller is designed, and its expression is:
  • the initial state can be defined as The following coupled neural network model with mixed time-varying delay, nonlinearity and non-identity:
  • the bounded cluster synchronization condition under the parameter mismatch condition of the coupled neural network with mixed delay and nonlinear and non-identity (equation (1)) is discussed. All mathematical formulations are based on Lyapunov stability theorem, mean pulse interval, matrix measure method and parameter variational method.
  • the present invention utilizes the designed pinning impulse controller (equation (4)) to obtain sufficient conditions for the bounded cluster projection synchronization between the neural network (equation (1)) and the target neural network (equation (2)).
  • Step 4 Based on the action of the pulse pinning controller, the following Lyapunov function is selected by the matrix measurement method:
  • inequality (9) can be written as:
  • the synchronization error can be derived using a linearization method:
  • is any value greater than zero and the function v(t) ⁇ V(t).
  • v(t) can be calculated as:
  • W(t,s) is based on the linear impulse system
  • the resulting Cauchy matrix using the idea of average pulse interval, can be calculated as: where Ta is the average pulse interval of the pulse sequence ⁇ .
  • inequality (16) holds under the condition t>0. Note that, even if inequality (16) does not hold under the condition t>0, there exists t * >0 such that inequality (16) holds at 0 ⁇ t ⁇ t * condition holds.
  • the synchronization error bound reaches its minimum value.
  • the appropriate pulse frequency can be selected so that there is This results in a smaller synchronization error bound.
  • the bounded convergence speed can be finally realized with the convergence speed of ⁇ under the action of the pinned impulse controller (Equation (5)) between the coupled neural network (Equation (1)) and the target neural network (Equation (2)).
  • Class projection synchronization where the synchronization error bound can be expressed as:
  • the synchronization effect between the target neural network and other neural networks is verified by building a network model and using this network model for numerical simulation.
  • Neural network is a network mathematical model that builds a structure similar to the connection of neurons in the brain for information processing.
  • the nonlinear mapping ability of neural network can be used to process data.
  • power electronic devices can be used to realize it.
  • transistors, resistors and capacitors are used to form a closed loop to simulate neuron circuits in memristive neural networks, and non-linear devices such as memristors and memristors are used to simulate the synapses of neural networks. Connect, the specific steps are as follows:
  • Step 1 Determine the coupled neural network model as follows:
  • the neural network with ideal parameters is selected as the synchronization target, and the target neural network model is determined as follows:
  • a coupled neural network composed of 6 neural networks is selected, as shown in Figure 1, wherein numbers 1, 2, 3, 4, 5, and 6 represent 6 neural networks and are divided into two clusters.
  • the specific control object selection scheme is to select a neural network with a larger error norm from the two clusters at each pulse time according to the pulse sequence when the coupled neural network is running, and control the The error feedback information in the controller acts on these two neural networks.
  • Step 3 Build the Simulink model of the coupled neural network (1), get the simulation results, and define the synchronization error of the neural network Figures 3-4 are obtained, which indicate that the error between any two nodes in the two clusters is within the synchronization error bound, that is, the bounded cluster projection synchronization is realized.
  • Figure 4 shows the state change curves of each neural network in the two clusters. It can be seen from Figure 5 to Figure 8 that due to the existence of non-identity and parameter mismatch features, not only in different clusters The state of the neural network has deviations, and the states of different neural networks in the same cluster also have deviations, but under the action of the pinning impulse controller, these state errors can be controlled within a certain range.
  • this embodiment provides a coupled neural network bounded clustering projection synchronization adjustment control system, the principle of which is similar to the coupled neural network bounded cluster projection synchronization adjustment control method, and will not be repeated.
  • the control system includes: a building module for building a coupled neural network with multiple clusters of nonlinear, non-identical and mixed time-varying time delays, and setting a target neural network for each cluster; a setting module for using to establish an error coupled neural network according to the coupled neural network and the target neural network; a cluster synchronization module is used to design a pinning pulse feedback controller according to the error coupled neural network model, and select a corresponding function based on the pinning pulse feedback controller , in order to realize the bounded cluster projection synchronization between the neural network in each cluster and the target neural network; the simulation module is used to verify the target neural network and the coupled neural network by building a network model and using the network model for numerical simulation Bounded cluster projection synchronization effect between networks.
  • the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • computer-usable storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions
  • the apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Neurology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • Feedback Control In General (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明涉及耦合神经网络有界聚类投影同步调节控制方法及系统。本发明包括建立具有非线性、非恒同和混合时变时滞的多个聚类的耦合神经网络,对每个聚类设定目标神经网络;根据所述耦合神经网络与目标神经网络建立误差耦合神经网络;根据所述误差耦合神经网络模型设计牵制脉冲反馈控制器,基于所述牵制脉冲反馈控制器选择相应函数,以实现每个聚类中神经网络与目标神经网络之间的有界聚类投影同步;通过搭建网络模型并利用所述网络模型进行数值仿真,验证目标神经网络与耦合神经网络之间的有界聚类投影同步效果。本发明控制成本低且控制精度高。

Description

耦合神经网络有界聚类投影同步调节控制方法及系统 技术领域
本发明涉及复杂网络同步控制技术领域,尤其是指耦合神经网络有界聚类投影同步调节控制方法及系统。
背景技术
近年来,复杂网络在各个领域的广泛应用,使得复杂网络的研究成为一个有趣的课题。通过校准系统参数或施加外部控制输入来迫使网络中的系统行为一致,称为同步现象。目前,混沌系统的聚类行为在图像处理、保密通信等工作中得到了广泛的应用,同步现象成为复杂网络研究中不可缺少的一部分。到目前为止,人们已经讨论了不同类型的同步现象,如有界同步、滞后同步、簇同步、相位同步、投影同步等。
在上述几种同步方法中,投影同步方法由于各节点的不同比例系数,在实际应用中具有重要意义。例如在数字通信领域,多进制通信通过投影同步实现更快的通信。有界同步是一种特殊的同步,一般由参数失配、内部干扰、外部攻击等因素引起,系统仅需要在一定范围内具有一致。此外,聚类同步作为一种常见的同步现象,描述了发生在各子群中的同步现象。
一般来说,很多复杂网络都很难独自实现同步。为了解决该问题,几种实现同步的有效方法已经被提出,包括牵制控制、脉冲控制、分布式控制、间歇控制等。其中脉冲控制作为一种高效节能的同步策略能够以低消耗实现有效控制,牵制控制也仅需要对网络中极少部分节点进行控制即可实现同步。因此,可以将以上两种控制方法联合使用以达到。
非恒同的耦合神经网络内部包含具有不同内部结构的神经网络,也可依此将其划分为不同聚类。在工业产品的方案设计过程中,神经网络常被应用于决策支持系统的推理 机构中,通过样本信息数据按照神经网络产生的推理策略得到最优方案。耦合神经网络在非恒同和参数不匹配的情况下实现同步的问题是很少被讨论的,由于不同神经网络的连接权值矩阵的差异性,考虑参数不匹配特征又是很有必要的。现有耦合神经网络的分布式控制相较于牵制控制成本较高,而一般的聚类牵制控制策略旨在消去不同聚类之间的相互作用并对连接多个聚类的关键节点进行控制,这类控制策略效率较低,适用场景较少且不能保证实现有效控制。
发明内容
为此,本发明所要解决的技术问题在于克服现有技术中完成非恒同的耦合神经网络控制的同步控制成本高且不能进行有效控制。。
为解决上述技术问题,本发明提供了耦合神经网络有界聚类投影同步调节控制方法,包括建立具有非线性、非恒同和混合时变时滞的多个聚类的耦合神经网络,对每个聚类设定目标神经网络;根据所述耦合神经网络与目标神经网络建立误差耦合神经网络;根据所述误差耦合神经网络模型设计牵制脉冲反馈控制器,基于所述牵制脉冲反馈控制器选择相应函数,以实现每个聚类中神经网络与目标神经网络之间的有界聚类投影同步;通过搭建网络模型并利用所述网络模型进行数值仿真,验证目标神经网络与耦合神经网络之间的有界聚类投影同步效果。
在本发明的一个实施例中,所述耦合神经网络的模型为:
Figure PCTCN2021097087-appb-000001
其中:
Figure PCTCN2021097087-appb-000002
是节点的状态向量;N个神经网络分为l个聚类,且存在N≥l>0,第i个神经网络和第j个神经网络在第z个聚类中,定义μ i=μ j=z,反之则有μ i≠μ j
Figure PCTCN2021097087-appb-000003
是第μ i个聚类中神经网络的连接权值矩阵;
Figure PCTCN2021097087-appb-000004
是神经元的外部输入向量;f k(·):R n→R n,k=1,2,3表示神经元的激活函数,其中有
Figure PCTCN2021097087-appb-000005
Figure PCTCN2021097087-appb-000006
Figure PCTCN2021097087-appb-000007
正常数σ 12是耦合神经网络的耦合强度;Γ,Υ表示耦合神经网络的内部耦合矩阵,Γ,Υ为单位矩阵;τ 1(t),τ 2(t)和τ 3(t)各自表示系统时变时滞,状态耦合时变时滞和分布式耦合时变时滞,存在0≤τ 1(t)≤τ 1,0≤τ 2(t)≤τ 2,0≤τ 3(t)≤τ 3,并定义最大时滞为τ=max{τ 1(t),τ 2(t),τ 3(t)};G=(g ij) m×m和D=(d ij) m×m是基于耦合神经网络拓扑结构的外部耦合矩阵,矩阵G,D满足耗散条件,即满足
Figure PCTCN2021097087-appb-000008
Figure PCTCN2021097087-appb-000009
在第i个神经网络与第j个神经网络之间存在连接,则有g ij=g ji>0(d ij=d ji>0),否则g ij=0(d ij=0);u i(t)是控制器。
在本发明的一个实施例中,所述目标神经网络的模型为:
Figure PCTCN2021097087-appb-000010
其中:
Figure PCTCN2021097087-appb-000011
是神经网络的状态向量,
Figure PCTCN2021097087-appb-000012
是神经网络的连接权值矩阵,且存在
Figure PCTCN2021097087-appb-000013
在本发明的一个实施例中,定义矩阵测度μ q(M):
Figure PCTCN2021097087-appb-000014
其中:I是一个n维单位向量,||·|| q,q=1,2,∞,表示诱导范数,对于所述目标神经网络的状态向量的诱导范数,存在约束:
Figure PCTCN2021097087-appb-000015
其中
Figure PCTCN2021097087-appb-000016
是一个正常数。
在本发明的一个实施例中,所述误差耦合神经网络的模型为:
Figure PCTCN2021097087-appb-000017
其中:
Figure PCTCN2021097087-appb-000018
Figure PCTCN2021097087-appb-000019
Figure PCTCN2021097087-appb-000020
Figure PCTCN2021097087-appb-000021
e i(t)=x i(t)-as μi(t)为误差向量,正常数a用来代表投影因子。
在本发明的一个实施例中,在第μ i个聚类中,当且仅当对于任意初始状态
Figure PCTCN2021097087-appb-000022
Figure PCTCN2021097087-appb-000023
存在正参数
Figure PCTCN2021097087-appb-000024
使得如下不等式成立
Figure PCTCN2021097087-appb-000025
在本发明的一个实施例中,定义为
Figure PCTCN2021097087-appb-000026
Figure PCTCN2021097087-appb-000027
得到如下具有混合时变时滞、非线性和非恒同误差耦合神经网络模型:
Figure PCTCN2021097087-appb-000028
其中:假设误差状态向量e i(t)是右连续的,存在
Figure PCTCN2021097087-appb-000029
Figure PCTCN2021097087-appb-000030
在本发明的一个实施例中,所述牵制脉冲反馈控制器表达式为:
Figure PCTCN2021097087-appb-000031
其中:脉冲强度ρ∈(-2,-1)∪(-1,0),δ(·)表示狄拉克函数;对于脉冲信号,时间序列ζ={t 1,t 2,…}是严格单调递增的。
在本发明的一个实施例中,所述函数为李雅普诺夫函数,表达式为:
Figure PCTCN2021097087-appb-000032
其中P是一个常正定矩阵。
本发明还提供一种应用于上述方法的耦合神经网络有界聚类投影同步调节控制系统,包括:构建模块,用于建立具有非线性、非恒同和混合时变时滞的多个聚类的耦合神经网络,并对每个聚类设定目标神经网络;设置模块,用于根据所述耦合神经网络与目标神经网络建立误差耦合神经网络;聚类同步模块,用于根据所述误差耦合神经网络模型设计牵制脉冲反馈控制器,基于所述牵制脉冲反馈控制器选择相应函数,以实现每个聚类中神经网络与目标神经网络之间的有界聚类投影同步;仿真模块,用于通过搭建网络模型并利用所述网络模型进行数值仿真,验证目标神经网络与耦合神经网络之间的有界聚类投影同步效果。
本发明的上述技术方案相比现有技术具有以下优点:
本发明所述的耦合神经网络有界聚类投影同步调节控制方法及系统,有如下优点:
1.充分考虑了耦合神经网络内不同神经网络的信号传输延迟和不同神经网络之间的内部差异性,构建了一类具有包含系统时变时滞、一般耦合时变时滞和分布式耦合时变时滞,在参数不匹配条件下,具有非线性和非恒同特点的耦合神经网络模型;
2.针对不同神经网络的非线性激活函数的不同,设计了一类用于实现耦合神经网络聚类同步的牵制脉冲控制器,通过选定不同聚类中具有较大状态误差的少部分神经网络作为被控对象,反馈被控对象的误差状态信息,从而可以在较少的消耗情况下实现有效控制;
3.基于李雅普诺夫稳定性定理、平均脉冲间隔的概念和一些线性化方法,利用不具有非负性的矩阵测度方法给出了更加准确的耦合神经网络有界聚类投影同步的判定条件,并通过参数变分法给出了有界聚类投影同步的指数收敛速度和对应的同步误差界,而与传统的给定脉冲上下确界来比,也降低了同步判定条件的保守型;
4.考虑了有界聚类投影同步的存在条件,对同步误差界的取值进行了优化。通过对牵制脉冲控制器的脉冲强度和选定被控对象的规模的控制,能够保证同步误差界的存在,而通过对脉冲频率的调整,可以相应得减小同步误差。
附图说明
为了使本发明的内容更容易被清楚的理解,下面根据本发明的具体实施例并结合附 图,对本发明作进一步详细的说明,其中
图1是实施例2中的耦合神经网络结构图。
图2是实施例2中的脉冲时间序列图。
图3是实施例2中神经网络的聚类牵制控制策略效果图。
图4是实施例2中神经网络误差状态演化曲线图。
图5是实施例2中聚类1中神经网络的状态1演化曲线图。
图6是实施例2中聚类1中神经网络的状态2演化曲线图。
图7是实施例2中聚类2中神经网络的状态1演化曲线图。
图8是实施例2中聚类2中神经网络的状态2演化曲线图。
具体实施方式
下面结合附图和具体实施例对本发明作进一步说明,以使本领域的技术人员可以更好地理解本发明并能予以实施,但所举实施例不作为对本发明的限定。
实施例1
本实施例提供耦合神经网络有界聚类投影同步调节控制方法,包括:
步骤一:考虑非恒同的神经网络并将其分为多个聚类,为每个子类设定一个目标神经网络,此时目标神经网络可以视为领导,而其他神经网络可视为其跟随者。首先建立如下具有非线性、非恒同和混合时变时滞的耦合神经网络模型:
Figure PCTCN2021097087-appb-000033
其中:
Figure PCTCN2021097087-appb-000034
是节点的状态向量;假定N个神经网络可以被分为l个聚类,且存在N≥l>0,若第i个神经网络和第j个神经网络在第z个聚类中,则定义μ i=μ j=z,反之则有μ i≠μ j
Figure PCTCN2021097087-appb-000035
是第μ i个聚类中神经网络的连接权值矩阵;
Figure PCTCN2021097087-appb-000036
是神经元的外部输入向量;f k(·):R n→R n,k=1,2,3表示神经元的激活函数,其中有
Figure PCTCN2021097087-appb-000037
Figure PCTCN2021097087-appb-000038
Figure PCTCN2021097087-appb-000039
正常数σ 12是耦合神经网络的耦合强度;Γ,Υ表示耦合神经网络的内部耦合矩阵,为了简单起见,假设Γ,Υ为单位矩阵;τ 1(t),τ 2(t)和τ 3(t)各自表示系统时变时滞,状态耦合时变时滞和分布式耦合时变时滞,存在0≤τ 1(t)≤τ 1,0≤τ 2(t)≤τ 2,0≤τ 3(t)≤τ 3,并定义最大时滞为τ=max{τ 1(t),τ 2(t),τ 3(t)};G=(g ij) m×m和D=(d ij) m×m是基于耦合神经网络拓扑结构的外部耦合矩阵,矩阵G,D满足耗散条件,即满足
Figure PCTCN2021097087-appb-000040
Figure PCTCN2021097087-appb-000041
Figure PCTCN2021097087-appb-000042
此外如果在第i个神经网络与第j个神经网络之间存在连接,则有g ij=g ji>0(d ij=d ji>0),否则g ij=0(d ij=0);u i(t)是控制器,在之后进行详细设计。
确认领导节点:由于本发明在各个聚类中确定了领导跟随的形式,因此需要在每个聚类中设定一个目标神经网络作为领导。确认如下第μ i个聚类的领导为:
Figure PCTCN2021097087-appb-000043
其中:
Figure PCTCN2021097087-appb-000044
是该神经网络的状态向量,
Figure PCTCN2021097087-appb-000045
是该神经网络的连接权值矩阵,且存在
Figure PCTCN2021097087-appb-000046
这体现了参数不匹配特征;在第μ i个聚类方程(式(1))中的所有神经网络都可以看作目标神经网络(式(2))的追随者。
为了获得更加准确的结果,本发明引入了矩阵测度方法,定义矩阵测度μ q(M)如下
Figure PCTCN2021097087-appb-000047
其中:I是一个n维单位向量,||·|| q,q=1,2,∞表示诱导范数。对于目标神经网络(式(2))的状态向量的诱导范数,存在约束:
Figure PCTCN2021097087-appb-000048
其中
Figure PCTCN2021097087-appb-000049
是一个正常数。
步骤二:通过传感器获得各节点的状态信息,基于投影同步则可以得到误差向量e i(t)=x i(t)-as μi(t)的状态信息,正常数a用来代表投影因子,从而得到如下具有非线性、非恒同耦合和多重时滞的误差耦合神经网络:
Figure PCTCN2021097087-appb-000050
其中:
Figure PCTCN2021097087-appb-000051
Figure PCTCN2021097087-appb-000052
Figure PCTCN2021097087-appb-000053
Figure PCTCN2021097087-appb-000054
Figure PCTCN2021097087-appb-000055
通过对所述误差耦合神经网络模型的处理,从而可以将各个聚类中的神经网络同步问题转换为一个误差耦合神经网络全局稳定性问题,更易于处理。
步骤三:假设耦合神经网络中的第z个神经网络同时也是第i个聚类中按最大误差排序的第j个神经网络,则定义符号
Figure PCTCN2021097087-appb-000056
为了实现各聚类中神经网络(式(1))与目标神经网络(式(2))之间的网络同步,基于诱导范数将各聚类中按状态误差大小排序分为两组神经网络,分组如下:
Figure PCTCN2021097087-appb-000057
Figure PCTCN2021097087-appb-000058
则(a)包含在第μ i个聚类中诱导范数较大的
Figure PCTCN2021097087-appb-000059
个神经网络,用
Figure PCTCN2021097087-appb-000060
来表示;(b)包含第μ i个聚类中诱导范数较小的
Figure PCTCN2021097087-appb-000061
个神经网络,用
Figure PCTCN2021097087-appb-000062
来表示。选定
Figure PCTCN2021097087-appb-000063
内的神经网络进行控制,向其传输误差耦合神经网络的状态信息,设计了一种牵制脉冲反馈控制器,其表达式为:
Figure PCTCN2021097087-appb-000064
其中:脉冲强度ρ∈(-2,-1)∪(-1,0),δ(·)表示狄拉克函数;对于脉冲信号,假设这时间序列ζ={t 1,t 2,…}是严格单调递增。
考虑第μ i个聚类中神经网络(式(1))与目标神经网络(式(2)),初始状态可以定义为
Figure PCTCN2021097087-appb-000065
可以进一步得到如下具有混合时变时滞、非线性和非恒同耦合神经网络模型:
Figure PCTCN2021097087-appb-000066
其中:假设误差状态向量e i(t)是右连续的,即存在
Figure PCTCN2021097087-appb-000067
Figure PCTCN2021097087-appb-000068
在第μ i个聚类中,当且仅当对于任意初始状态
Figure PCTCN2021097087-appb-000069
Figure PCTCN2021097087-appb-000070
和存在正参数
Figure PCTCN2021097087-appb-000071
使得如下不等式成立
Figure PCTCN2021097087-appb-000072
则表示神经网络(式(1))与目标神经网络(式(2))是有界聚类投影同步的。
讨论具有混合时滞和非线性、非恒同的耦合神经网络(式(1))在参数不匹配条件下的有界聚类同步条件。所有的数学表述都是基于李雅普诺夫稳定性定理、平均脉冲间隔、矩阵测度方法和参数变分法。本发明利用所设计的牵制脉冲控制器(式(4))获得神经网络(式(1))和目标神经网络(式(2))之间有界聚类投影同步的充分条件。
步骤四:基于脉冲牵制控制器的作用,利用矩阵测度方法选取如下李雅普诺夫函数:
Figure PCTCN2021097087-appb-000073
其中P是一个常正定矩阵。
对于
Figure PCTCN2021097087-appb-000074
根据集合
Figure PCTCN2021097087-appb-000075
和集合
Figure PCTCN2021097087-appb-000076
的定义,可以得出
Figure PCTCN2021097087-appb-000077
Figure PCTCN2021097087-appb-000078
其中:
Figure PCTCN2021097087-appb-000079
并且ρ∈(-2,-1)∪(-1,0),则有
Figure PCTCN2021097087-appb-000080
另一方面,对于
Figure PCTCN2021097087-appb-000081
沿着被控误差耦合神经网络(6)的轨迹对V(t)求导,可以得到
Figure PCTCN2021097087-appb-000082
基于线性化方法以及矩阵测度方法的性质,存在正常数ω 123,可以使得下列不等式成立:
Figure PCTCN2021097087-appb-000083
Figure PCTCN2021097087-appb-000084
从而,不等式(9)可以写为如下:
Figure PCTCN2021097087-appb-000085
其中:
Figure PCTCN2021097087-appb-000086
Figure PCTCN2021097087-appb-000087
Figure PCTCN2021097087-appb-000088
对于有界同步现象总是会存在一个同步误差界,如不等式(9)中的
Figure PCTCN2021097087-appb-000089
Figure PCTCN2021097087-appb-000090
基于目标神经网络的状态约束,使用线性化方法可以得出同步误差:
Figure PCTCN2021097087-appb-000091
将得到的同步误差界(式(11))代入不等式(10),可以得到下式:
Figure PCTCN2021097087-appb-000092
根据已知条件与比较引理,可以得到函数v(t)是满足下列脉冲系统的解:
Figure PCTCN2021097087-appb-000093
其中:ε是大于零的任意值并且函数v(t)≥V(t)。
之后,根据参数变分法,v(t)可以计算得:
Figure PCTCN2021097087-appb-000094
其中:W(t,s)是根据线性脉冲系统
Figure PCTCN2021097087-appb-000095
所得的柯西矩阵,利用平均脉冲间隔思想,计算可得:
Figure PCTCN2021097087-appb-000096
其中T a是脉冲序列ζ的平均脉冲间隔。
将柯西矩阵W(t,s)代入等式(14),可以计算得到:
Figure PCTCN2021097087-appb-000097
其中:
Figure PCTCN2021097087-appb-000098
利用上述参数,定义方程为
Figure PCTCN2021097087-appb-000099
Figure PCTCN2021097087-appb-000100
分别计算g(0 +),g(+∞)和导数g′(λ),计算结果如下:
Figure PCTCN2021097087-appb-000101
g(+∞)>0,
Figure PCTCN2021097087-appb-000102
以上结果表明,g(λ)在区间(0,+∞)内单调递增,并且在该区间内只有一个唯一解。对于
Figure PCTCN2021097087-appb-000103
ε>0,
Figure PCTCN2021097087-appb-000104
存在下式:
Figure PCTCN2021097087-appb-000105
接下来证明不等式(16)在t>0条件下成立,值的注意的是,即使不等式(16)在t>0条件下不成立,也存在t *>0使得不等式(16)在0<t<t *条件下成立。
将不等式(16)代入不等式(15),可以得出下式:
Figure PCTCN2021097087-appb-000106
Figure PCTCN2021097087-appb-000107
使得上式中ε→0,从而对于t>0,存在进一步结果为:
Figure PCTCN2021097087-appb-000108
这表明被控误差耦合神经网络(式(6))会在同步误差界范围内实现指数同步,同步误差界可以写为:
Figure PCTCN2021097087-appb-000109
因此,在牵制脉冲控制器(式(5))的作用下,耦合神经网络(式(1))与目标神经网络(式(2))之间可以以λ的收敛速度最终实现有界聚类投影同步。特别地,构建函数
Figure PCTCN2021097087-appb-000110
对其求导可得f(γ)在(0,1)范围内的极大值为:
Figure PCTCN2021097087-appb-000111
换言之,当
Figure PCTCN2021097087-appb-000112
时,同步误差界达到其极小值。而要对同步误差界进行调整只能选取合适的脉冲频率使得存在
Figure PCTCN2021097087-appb-000113
从而获取更小的同步误差界。
因此可以得到如下结论:
对于脉冲效应ζ={t 1,t 2,…},和假设平均脉冲间隔小于T a。存在
Figure PCTCN2021097087-appb-000114
Figure PCTCN2021097087-appb-000115
Figure PCTCN2021097087-appb-000116
Figure PCTCN2021097087-appb-000117
使不等式
Figure PCTCN2021097087-appb-000118
也就是说,耦合神经网络(式(1))与目标神经网络(式(2))之间可以在牵制脉冲控制器(式(5))的作用下以λ的收敛速度最终实现有界聚类投影同步,其中同步误差界可以表示为:
Figure PCTCN2021097087-appb-000119
通过搭建网络模型并利用此网络模型进行数值仿真,来验证目标神经网络与其他神经网络之间的同步效果。
实施例2
本实施例为了验证上述实施例1中方法的正确性,搭建网络模型来进行仿真验证。神经网络是构建类似于大脑神经元联接的结构进行信息处理的网络数学模型,利用神经网络的非线性映射能力可以对数据进行处理。同时可以应用电力电子器件对其进行实现,如在忆阻神经网络中利用晶体管、电阻和电容构成闭合环路模拟神经元电路,利用忆阻器、忆容器等非线性器件模拟神经网络的突触连接,具体步骤如下:
步骤1:确定耦合神经网络模型如下:
Figure PCTCN2021097087-appb-000120
其中:
Figure PCTCN2021097087-appb-000121
Figure PCTCN2021097087-appb-000122
Figure PCTCN2021097087-appb-000123
选取激活函数为f 1(u)=f 2(u)=f 3(u)=tanh(u)。
选取具有理想参数的神经网络作为同步目标,确定目标神经网络模型如下:
Figure PCTCN2021097087-appb-000124
其中:
Figure PCTCN2021097087-appb-000125
Figure PCTCN2021097087-appb-000126
为了验证本发明的正确性,选取6个神经网络组成的耦合神经网络,如图1所示,其中数字1,2,3,4,5,6表示6个神经网络并且被分为两个聚类,而具体的被控对象 的选择方案是在耦合神经网络运行时,依据脉冲序列在各个脉冲时刻时,在两个聚类中选取具有较大误差范数的神经网络各一个,并将控制器中的误差反馈信息作用在这两个神经网络上。
步骤2:根据已知,选取耦合矩阵G=[-2,1,1,0,0,0;1,-2,1,0,0,0;1,1,-3,1,0,0;0,0,1,-3,1,1;0,0,0,1,-2,1;0,0,0,1,1,-2]。此外,脉冲序列ζ={t 1,t 2,…}如图2所示,并假设平均脉冲间隔P a不超过0.02,脉冲控制强度为ρ=1.2。
步骤3:搭建耦合神经网络(1)的Simulink模型,得到仿真结果,并通过定义神经网络同步误差
Figure PCTCN2021097087-appb-000127
得到图3-图4,其表示两个聚类内任意两个节点之间的误差在同步误差界范围内,即实现了有界聚类投影同步。同时,图4中给出了两个聚类中各个神经网络的各状态变化曲线,从图5-图8可以看出,由于非恒同和参数不匹配特征的存在,不仅在不同聚类中的神经网络状态存在偏差,而且相同聚类中的不同神经网络的状态也是存在偏差的,但是在牵制脉冲控制器的作用下,能够将这些状态误差控制在一定范围内。
实施例3
基于同一发明构思,本实施例提供了耦合神经网络有界聚类投影同步调节控制系统,其解决问题的原理与所述耦合神经网络有界聚类投影同步调节控制方法类似,不再赘述。
所述控制系统包括:构建模块,用于建立具有非线性、非恒同和混合时变时滞的多个聚类的耦合神经网络,并对每个聚类设定目标神经网络;设置模块,用于根据所述耦合神经网络与目标神经网络建立误差耦合神经网络;聚类同步模块,用于根据所述误差耦合神经网络模型设计牵制脉冲反馈控制器,基于所述牵制脉冲反馈控制器选择相应函数,以实现每个聚类中神经网络与目标神经网络之间的有界聚类投影同步;仿真模块,用于通过搭建网络模型并利用所述网络模型进行数值仿真,验证目标神经网络与耦合神经网络之间的有界聚类投影同步效果。
本领域内的技术人员应明白,本申请的实施例可提供为方法、系统、或计算机程序产品。因此,本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且,本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计 算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器,使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中,使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品,该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上,使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理,从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
显然,上述实施例仅仅是为清楚地说明所作的举例,并非对实施方式的限定。对于所属领域的普通技术人员来说,在上述说明的基础上还可以做出其它不同形式变化或变动。这里无需也无法对所有的实施方式予以穷举。而由此所引申出的显而易见的变化或变动仍处于本发明创造的保护范围之中。

Claims (10)

  1. 耦合神经网络有界聚类投影同步调节控制方法,其特征在于,包括如下步骤:
    步骤一:建立具有非线性、非恒同和混合时变时滞的多个聚类的耦合神经网络,对每个聚类设定目标神经网络;
    步骤二:根据所述耦合神经网络与目标神经网络建立误差耦合神经网络;
    步骤三:根据所述误差耦合神经网络模型设计牵制脉冲反馈控制器,基于所述牵制脉冲反馈控制器选择相应函数,以实现每个聚类中神经网络与目标神经网络之间的有界聚类投影同步;
    步骤四:通过搭建网络模型并利用所述网络模型进行数值仿真,验证目标神经网络与耦合神经网络之间的有界聚类投影同步效果。
  2. 根据权利要求1所述的耦合神经网络有界聚类投影同步调节控制方法,其特征在于,所述耦合神经网络的模型为:
    Figure PCTCN2021097087-appb-100001
    其中:
    Figure PCTCN2021097087-appb-100002
    是节点的状态向量;N个神经网络分为l个聚类,且存在N≥l>0,第i个神经网络和第j个神经网络在第z个聚类中,定义μ i=μ j=z,反之则有μ i≠μ j
    Figure PCTCN2021097087-appb-100003
    是第μ i个聚类中神经网络的连接权值矩阵;
    Figure PCTCN2021097087-appb-100004
    是神经元的外部输入向量;f k(·):R n→R n,k=1,2,3表示神经元的激活函数,其中有
    Figure PCTCN2021097087-appb-100005
    Figure PCTCN2021097087-appb-100006
    Figure PCTCN2021097087-appb-100007
    正常数σ 12是耦合神经网络的耦合强度;Γ,Υ表示耦合神经网络的内部耦合矩阵,Γ,Υ为单位矩阵;τ 1(t),τ 2(t)和τ 3(t)各自表示系统时变时滞,状态耦合时变时滞和分布式耦合时变时滞, 存在0≤τ 1(t)≤τ 1,0≤τ 2(t)≤τ 2,0≤τ 3(t)≤τ 3,并定义最大时滞为τ=max{τ 1(t),τ 2(t),τ 3(t)};G=(g ij) m×m和D=(d ij) m×m是基于耦合神经网络拓扑结构的外部耦合矩阵,矩阵G,D满足耗散条件,即满足
    Figure PCTCN2021097087-appb-100008
    Figure PCTCN2021097087-appb-100009
    在第i个神经网络与第j个神经网络之间存在连接,则有g ij=g ji>0(d ij=d ji>0),否则g ij=0(d ij=0);u i(t)是控制器。
  3. 根据权利要求1所述的耦合神经网络有界聚类投影同步调节控制方法,其特征在于,所述目标神经网络的模型为:
    Figure PCTCN2021097087-appb-100010
    其中:
    Figure PCTCN2021097087-appb-100011
    是神经网络的状态向量,
    Figure PCTCN2021097087-appb-100012
    是神经网络的连接权值矩阵,且存在
    Figure PCTCN2021097087-appb-100013
  4. 根据权利要求3所述的耦合神经网络有界聚类投影同步调节控制方法,其特征在于,定义矩阵测度μ q(M):
    Figure PCTCN2021097087-appb-100014
    其中:I是一个n维单位向量,||·|| q,q=1,2,∞,表示诱导范数,对于所述目标神经网络的状态向量的诱导范数,存在约束:
    Figure PCTCN2021097087-appb-100015
    其中
    Figure PCTCN2021097087-appb-100016
    是一个正常数。
  5. 根据权利要求1所述的耦合神经网络有界聚类投影同步调节控制方法,其特征在于,所述误差耦合神经网络的模型为:
    Figure PCTCN2021097087-appb-100017
    其中:
    Figure PCTCN2021097087-appb-100018
    Figure PCTCN2021097087-appb-100019
    Figure PCTCN2021097087-appb-100020
    Figure PCTCN2021097087-appb-100021
    e i(t)=x i(t)-as μi(t)为误差向量,正常数a用来代表投影因子。
  6. 根据权利要求1所述的耦合神经网络有界聚类投影同步调节控制方法,其特征在于,在第μ i个聚类中,当且仅当对于任意初始状态
    Figure PCTCN2021097087-appb-100022
    存在正参数
    Figure PCTCN2021097087-appb-100023
    使得如下不等式成立
    Figure PCTCN2021097087-appb-100024
  7. 根据权利要求5所述的耦合神经网络有界聚类投影同步调节控制方法,其特征在于,定义为
    Figure PCTCN2021097087-appb-100025
    得到如下具有混合时变时滞、非线性和非恒同误差耦合神经网络模型:
    Figure PCTCN2021097087-appb-100026
    其中:假设误差状态向量e i(t)是右连续的,存在
    Figure PCTCN2021097087-appb-100027
    Figure PCTCN2021097087-appb-100028
  8. 根据权利要求1所述的耦合神经网络有界聚类投影同步调节控制方法,其特征在于,所述牵制脉冲反馈控制器表达式为:
    Figure PCTCN2021097087-appb-100029
    其中:脉冲强度ρ∈(-2,-1)∪(-1,0),δ(·)表示狄拉克函数;对于脉冲信号,时间序列ζ={t 1,t 2,…}是严格单调递增的。
  9. 根据权利要求1所述的耦合神经网络有界聚类投影同步调节控制方法,其特征在于, 所述函数为李雅普诺夫函数,表达式为:
    Figure PCTCN2021097087-appb-100030
    其中P是一个常正定矩阵。
  10. 一种应用于权利要求1-9任一所述方法的耦合神经网络有界聚类投影同步调节控制系统,其特征在于,包括:
    构建模块,用于建立具有非线性、非恒同和混合时变时滞的多个聚类的耦合神经网络,并对每个聚类设定目标神经网络;
    设置模块,用于根据所述耦合神经网络与目标神经网络建立误差耦合神经网络;
    聚类同步模块,用于根据所述误差耦合神经网络模型设计牵制脉冲反馈控制器,基于所述牵制脉冲反馈控制器选择相应函数,以实现每个聚类中神经网络与目标神经网络之间的有界聚类投影同步;
    仿真模块,用于通过搭建网络模型并利用所述网络模型进行数值仿真,验证目标神经网络与耦合神经网络之间的有界聚类投影同步效果。
PCT/CN2021/097087 2021-04-21 2021-05-31 耦合神经网络有界聚类投影同步调节控制方法及系统 WO2022222229A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110429664.X 2021-04-21
CN202110429664.XA CN112884136B (zh) 2021-04-21 2021-04-21 耦合神经网络有界聚类投影同步调节控制方法及系统

Publications (1)

Publication Number Publication Date
WO2022222229A1 true WO2022222229A1 (zh) 2022-10-27

Family

ID=76040634

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/097087 WO2022222229A1 (zh) 2021-04-21 2021-05-31 耦合神经网络有界聚类投影同步调节控制方法及系统

Country Status (2)

Country Link
CN (1) CN112884136B (zh)
WO (1) WO2022222229A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115860096A (zh) * 2022-12-08 2023-03-28 盐城工学院 一种混合时变时滞的惯性神经网络的指数同步控制方法
CN116107218A (zh) * 2023-04-10 2023-05-12 广东工业大学 基于脉冲牵制控制的随机复杂网络有限时间同步的方法
CN116400598A (zh) * 2023-04-07 2023-07-07 盐城工学院 一种惯性cg神经网络的同步控制方法
CN116400597A (zh) * 2023-04-07 2023-07-07 盐城工学院 一种二阶神经网络的同步控制方法
CN117335957A (zh) * 2023-10-10 2024-01-02 盐城工学院 一种bam忆阻神经网络修正函数投影同步的保密通信方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113792858B (zh) * 2021-09-13 2024-03-01 江南大学 耦合神经网络有界同步及其分布式控制方法
CN114260885B (zh) * 2022-01-27 2023-08-04 同济大学 一种蛇形机器人的仿生cpg运动调控系统及其方法
CN115378565B (zh) * 2022-08-19 2024-01-19 曲阜师范大学 一种用于双层耦合网络准同步的事件触发控制方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190080233A1 (en) * 2017-09-14 2019-03-14 Intel Corporation Synchronization scheduler of distributed neural network training
CN111523648A (zh) * 2020-04-27 2020-08-11 江南大学 含有聚类拓扑耦合的神经网络脉冲同步方法及系统
CN111814333A (zh) * 2020-07-08 2020-10-23 江南大学 奇异Lur’e网络聚类同步的牵制节点选择方法
CN112099357A (zh) * 2020-09-22 2020-12-18 江南大学 非连续复杂网络的有限时间聚类同步与牵制控制方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10291460B2 (en) * 2012-12-05 2019-05-14 Origin Wireless, Inc. Method, apparatus, and system for wireless motion monitoring
CN105184270B (zh) * 2015-09-18 2019-06-18 深圳市数字城市工程研究中心 一种基于脉冲耦合神经网络方法的道路信息遥感提取方法
CN111652271A (zh) * 2020-04-24 2020-09-11 华东交通大学 一种基于神经网络的非线性特征选择方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190080233A1 (en) * 2017-09-14 2019-03-14 Intel Corporation Synchronization scheduler of distributed neural network training
CN111523648A (zh) * 2020-04-27 2020-08-11 江南大学 含有聚类拓扑耦合的神经网络脉冲同步方法及系统
CN111814333A (zh) * 2020-07-08 2020-10-23 江南大学 奇异Lur’e网络聚类同步的牵制节点选择方法
CN112099357A (zh) * 2020-09-22 2020-12-18 江南大学 非连续复杂网络的有限时间聚类同步与牵制控制方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
TANG ZE; XUAN DELI; PARK JU H.; WANG YAN; FENG JIANWEN: "Impulsive Effects Based Distributed Synchronization of Heterogeneous Coupled Neural Networks", IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING, IEEE, vol. 8, no. 1, 8 December 2020 (2020-12-08), pages 498 - 510, XP011843691, DOI: 10.1109/TNSE.2020.3042781 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115860096A (zh) * 2022-12-08 2023-03-28 盐城工学院 一种混合时变时滞的惯性神经网络的指数同步控制方法
CN115860096B (zh) * 2022-12-08 2023-07-07 盐城工学院 一种混合时变时滞的惯性神经网络的指数同步控制方法
CN116400598A (zh) * 2023-04-07 2023-07-07 盐城工学院 一种惯性cg神经网络的同步控制方法
CN116400597A (zh) * 2023-04-07 2023-07-07 盐城工学院 一种二阶神经网络的同步控制方法
CN116400598B (zh) * 2023-04-07 2023-09-26 盐城工学院 一种惯性cg神经网络的同步控制方法
CN116400597B (zh) * 2023-04-07 2023-10-24 盐城工学院 一种二阶神经网络的同步控制方法
CN116107218A (zh) * 2023-04-10 2023-05-12 广东工业大学 基于脉冲牵制控制的随机复杂网络有限时间同步的方法
CN117335957A (zh) * 2023-10-10 2024-01-02 盐城工学院 一种bam忆阻神经网络修正函数投影同步的保密通信方法
CN117335957B (zh) * 2023-10-10 2024-03-26 盐城工学院 一种bam忆阻神经网络修正函数投影同步的保密通信方法

Also Published As

Publication number Publication date
CN112884136B (zh) 2022-05-13
CN112884136A (zh) 2021-06-01

Similar Documents

Publication Publication Date Title
WO2022222229A1 (zh) 耦合神经网络有界聚类投影同步调节控制方法及系统
Park et al. Recent advances in control and filtering of dynamic systems with constrained signals
Wen et al. Neuro-adaptive consensus tracking of multiagent systems with a high-dimensional leader
Wang et al. Distributed robust control of uncertain linear multi‐agent systems
Zhang et al. Distributed adaptive tracking synchronization for coupled reaction–diffusion neural network
Sun et al. Adaptive consensus tracking for linear multi‐agent systems with heterogeneous unknown nonlinear dynamics
Hu et al. Second-order consensus for heterogeneous multi-agent systems in the cooperation–competition network: A hybrid adaptive and pinning control approach
Chen et al. Leader‐following control for multiple inertial agents
Boulkroune et al. Adaptive fuzzy control‐based projective synchronization of uncertain nonaffine chaotic systems
Ye et al. Leader‐following consensus of general linear fractional‐order multiagent systems with input delay via event‐triggered control
Ren et al. Finite time synchronization of Markovian jumping stochastic complex dynamical systems with mix delays via hybrid control strategy
CN109471364B (zh) 一种带有执行器故障的非线性切换系统的可靠控制方法
Qin et al. Robust adaptive consensus of nonstrict-feedback multi-agent systems with quantized input and unmodeled dynamics
Huo et al. Fully distributed consensus for general linear multi‐agent systems with unknown external disturbances
Cui et al. Distributed containment control for nonlinear multiagent systems in pure‐feedback form
Zhang et al. Adaptive tracking synchronization for coupled reaction–diffusion neural networks with parameter mismatches
Yu et al. Distributed-delay-dependent exponential stability of impulsive neural networks with inertial term
Gong et al. Robust state estimation for delayed complex-valued neural networks
Xiong et al. Multilayered sampled-data iterative learning tracking for discrete systems with cooperative–antagonistic interactions
Liu et al. Impulsive H∞ synchronization for reaction–diffusion neural networks with mixed delays
CN113792858A (zh) 耦合神经网络有界同步及其分布式控制方法
Cao et al. Synchronization of multiple reaction–diffusion memristive neural networks with known or unknown parameters and switching topologies
Yan et al. Algebraic method of simplifying Boolean networks using semi‐tensor product of Matrices
Ji et al. Leader‐following consensus of fractional‐order multi‐agent systems via adaptive control
Li et al. Quasi-synchronization control of quaternion-valued fuzzy memristive neural networks

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21937454

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21937454

Country of ref document: EP

Kind code of ref document: A1