返回
类型 基础研究 预答辩日期 2018-05-20
开始(开题)日期 2017-01-13 论文结束日期 2018-01-22
地点 东南大学四牌楼校区自动化学院二楼会议室 论文选题来源 国家自然科学基金项目     论文字数 5 (万字)
题目 时变时滞神经网络稳定性及同步控制研究
主题词 神经网络,时变时滞,Lyapunov-Krasovskii泛函,有限时间同步,滑模控制
摘要 时滞神经网络是时滞系统的重要组成部分,其具有非常丰富的动力学行为。近年来,神经网络在信号与图像处理、联想记忆、模式识别、组合优化、自动控制和其它领域的广泛应用,引起了人们对神经网络动力学行为的大量关注,但时滞的存在对神经网络的动力学行为的影响不容忽视,其可能导致振荡行为、失稳现象甚至混沌现象。因此时滞神经网络的稳定性分析和同步控制已成为两大热门的研究课题,一系列具有较低保守性的时滞相关稳定性判据和同步控制方法得到发展,但仍有进一步研究和改进的空间,例如,稳定性判据中,矩阵的维数高和计算负担重;同步控制方法中,求解的控制器增益可能过大而无法物理实现。本论文针对时变时滞神经网络的稳定性判据和有限时间同步展开研究,在改进现有稳定性判据的同时也对有限时间同步控制方法进行了发展。本论文的主要研究成果如下: (1) 带有零下限时变时滞的神经网络稳定性判据 针对时变时滞神经网络的稳定分析问题,指出现有利用时滞神经元状态导数信息作为状态向量集研究存在的弊端,即时滞区间会发生转移和稳定性判据中矩阵的维数与计算负担会增加,提出一种基于改进的增广型Lyapunov-Krasovskii泛函(LKF)的时滞相关稳定分析方法。在LKF的导数估计中,时滞神经元状态导数信息不会出现,以致建立的时滞相关稳定性判据具有较低的保守性和计算负担。 (2) 带有非零下限时变时滞的神经网络稳定性判据 针对具有时变时滞的一般神经网络的稳定分析问题,充分考虑时变时滞下限为非零和时滞导数下限可测情况,利用时变时滞下限信息和神经元激活函数信息构造新的增广型LKF,提出一种基于新的增广型LKF的时滞相关稳定分析方法。基于互凸组合不等式和神经元激活函数的扇区约束条件,利用该方法建立的时滞相关稳定性判据具有更低的保守性和相对较低的计算负担。 (3) 基于一种滑模控制方法的时变时滞神经网络的有限时间同步 针对时变时滞神经网络的有限时间同步问题,基于驱动-响应概念和滑模控制理论,将同步误差直接定义为滑模流型,提出一种滑模控制方法。该方法相比现有线性状态或时滞状态反馈控制方法,具有以下几点优势:无需求解反馈控制增益;时滞神经网络的有限时间同步得到保证,而且同步时间短。 (4) 基于一种积分滑模控制方法的时变时滞神经网络的有限时间同步 针对时变时滞神经网络的有限时间同步问题,指出现有利用积分滑模控制方法研究存在的问题,提出一种改善的积分滑模控制方法。该方法相比现有积分滑模控制方法,具有以下几点优势:积分滑模流型结构更简单;无需求解状态反馈控制增益矩阵;时滞神经网络的有限时间同步得到保证。 (5) 基于一种自适应滑模控制方法的时变时滞神经网络的自适应同步 针对时变时滞神经网络的自适应同步问题,考虑驱动系统与响应系统中的外界常数输入向量不匹配情况,利用同步误差构造一种积分滑模流型,设计一种合适的自适应滑模控制器和对应的自适应律,提出一种自适应滑模控制方法。利用该方法能够消除不匹配的外界常数输入向量对时变时滞神经网络同步的影响。
英文题目 RESEARCH ON STABILITY AND SYNCHRONIZATION CONTROL FOR NEURAL NETWORKS WITH TIME-VARYING DELAYS
英文主题词 Neural networks, time-varying delays, Lyapunov-Krasovskii functional (LKF), finite-time synchronization, sliding mode control
英文摘要 Delayed neural networks are important component of delayed systems, which have very abundant dynamic behaviors. In recent years, much attention on the dynamic behaviors of neural networks has been paid due to their extensive applications in signal and image processing, associative memories, pattern recognition, combinatorial optimization, automatic control and other areas. However, the existence of time delay cannot be ignored for the dynamic behaviors of neural networks, which may lead to oscillation, instability and even chaotic phenomenon. Therefore, the stability analysis and synchronization control of delayed neural networks have become two active topics of research. Many less conservative delay- dependent stability criteria and synchronization control methods have been developed, but there is room for further investigation and improvement, for example, in the stability criteria, the matrices have high dimension and large computational burden; in the synchronization control methods, the obtained control gains may be too big to implement physically. This dissertation investigates the stability criteria and finite-time synchronization for neural networks with time-varying delays, which not only improves the existing stability criteria but also develops the finite-time synchronization control methods. The main research results of this paper are as follows: (1) Stability criteria of neural networks with zero lower bound of time-varying delays For the stability analysis of neural networks with time-varying delays, some drawbacks existed in the study of using the information of delay neuron state derivative as a state vector set are pointed out, namely, the delay integral is shifted, and the dimension and computational burden of the matrices in the stability criteria are increased. A delay-dependent stability analysis method based on improved augmented Lyapunov-Krasovskii functional (LKF) is proposed. In the estimation of the derivative of the LKF, the information of delay neuron state derivative is disappeared such that the established delay-dependent stability criteria have less conservatism and lower computational burden. (2) Stability criterion of neural networks with nonzero lower bound of time-varying delays For the stability analysis of generalized neural networks with time-varying delays, the cases that the lower bound of time-varying delay is nonzero and the the lower bound of delay derivative is known are fully considered, new augmented LKF is constructed by employing the information on lower bounds of time-varying delays and neuron activation functions, a delay-dependent stability analysis method that depends on the new augmented LKF is derived. Based on the reciprocally convex combination inequality and bounded conditions on the slope of neuron activation functions, the established delay-dependent stability criterion has much less conservatism and relatively low computational burden. (3) Finite-time synchronization of neural networks with time-varying delays based on a sliding mode control method For the finite-time synchronization problem of neural networks with time-varying delays, based on the drive-response concept and sliding mode control theory, the synchronization error is directly defined as a sliding manifold, a sliding mode control method is derived, which, compared with the existing linear state or delayed state feedback control methods, has the following advantages: there is no need to solve the unknown feedback control gain; the finite-time synchronization of delayed neural networks is guaranteed, and the synchronization time is shot. (4) Finite-time synchronization of neural networks with time-varying delays based on an integral sliding mode control method For the finite-time synchronization problem of neural networks with time-varying delays, some problems existed in the study of using integral sliding mode control method are pointed out, an improved integral sliding mode control method is proposed, which, compared with the existing integral sliding mode control methods, has the following advantages: the integral sliding manifold has simpler structure; there is no need to solve the unknown state feedback control gain matrix; the finite-time synchronization of delayed neural networks is guaranteed. (5) Adaptive synchronization of neural networks with time-varying delays based on an adaptive sliding mode control method For the adaptive synchronization problem of neural networks with time-varying delays, the case that the external constant input vectors of drive system and response system are mismatched is considered, an integral sliding manifold is constructed by employing the synchronization error, a suitable adaptive sliding mode controller and corresponding adaptive law are designed, an adaptive sliding mode control method is proposed, which can eliminate the affect caused by mismatched external constant input vectors.
学术讨论
主办单位时间地点报告人报告主题
自动化学院 2015.9 榴园宾馆 吴宏鑫 特征建模的理论与方法
自动化学院 2016.3 中心楼2楼会议室 Shi Ling My personal opinions on what’s important for a PG student
自动化学院 2016.3 中心楼2楼会议室 李文华 Control of small scale unmanned aircraft systems under wind conditions
自动化学院 2015.9 中心楼608 熊晶晶 Global fast dynamic terminal sliding mode control for a quadrotor UAV
自动化学院 2015.9 中心楼608 熊晶晶 Second order sliding mode control for a quadrotor UAV with input saturation
自动化学院 2017.1 中心楼102 熊晶晶 Delay-dependent stability criterion of recurrent neural networks with time-varying delays
自动化学院 2017.5 中心楼102 熊晶晶 Delay-dependent stability of neural networks with time-varying delays
自动化学院 2017.8 中心楼102 熊晶晶 Finite-time synchronization of neural networks with time-varying delays
     
学术会议
会议名称时间地点本人报告本人报告题目
2016 International Confernence on Control and Automation 2016.1.16 中国深圳 Second order sliding mode control for a class of underactuated sysems
2016 The 2nd International Conference on Control, Automation and Robotics 2016.4.29 中国香港 Sliding mode control for a quadrotor UAV with parameter uncertainties
第36届中国控制会议 2017.7.28 中国大连 Delay-dependent stability of neural networks with time-varying delays
     
代表作
论文名称
Global fast dynamic terminal sliding mode control for a quadrotor UAV
Discrete-time sliding mode control for a quadrotor UAV
Sliding mode control for a quadrotor UAV with parameter uncertainties
 
答辩委员会组成信息
姓名职称导师类别工作单位是否主席备注
曹云峰 正高 教授 博导 南京航空航天大学
卢子芳 正高 教授 硕导 南京邮电大学
魏海坤 正高 教授 博导 东南大学
费树岷 正高 教授 博导 东南大学
翟军勇 正高 教授 博导 东南大学
      
答辩秘书信息
姓名职称工作单位备注
杨万扣 其他 副研究员 东南大学