丁三波 王勇 舒紀(jì)超 耿艷利 王婕
摘要 通過設(shè)計(jì)連接權(quán)的更新律,研究時(shí)滯遞歸神經(jīng)網(wǎng)絡(luò)的[H∞]穩(wěn)定性問題,建立基于線性矩陣不等式的[H∞]穩(wěn)定性判據(jù)。所設(shè)計(jì)的連接權(quán)更新律可以減少時(shí)滯遞歸神經(jīng)網(wǎng)絡(luò)模型中參數(shù)不確定性的影響,進(jìn)而增強(qiáng)系統(tǒng)的魯棒性。將所提方法應(yīng)用到時(shí)滯遞歸神經(jīng)網(wǎng)絡(luò)的[H∞]同步控制問題,并建立相關(guān)的同步判據(jù)。最后,通過算例說明所提方法的有效性。
關(guān) 鍵 詞 神經(jīng)網(wǎng)絡(luò);[H∞]性能;穩(wěn)定性;同步控制
中圖分類號(hào) TP183? ? ?文獻(xiàn)標(biāo)志碼 A
Abstract This paper investigates the[H∞]stability problem of delayed recurrent neural networks by designing the update laws of connection weights, and the[H∞]stability criterion is established in terms of linear matrix inequality. The proposed update laws of connection weights reduce the influence of parameter uncertainties in the delayed recurrent neural network model which results in the increase of the robustness of system. The proposed method is applied to[H∞]synchronization control of delayed recurrent neural networks, and a synchronization criterion is developed accordingly. Finally, the number examples are provided to illustrate the effectiveness of the proposed method.
Key words recurrent neural networks;[H∞]performance; stability; synchronization control
0 引言
神經(jīng)網(wǎng)絡(luò)是研究人工智能的重要方法,也是實(shí)現(xiàn)智能控制的重要工具[1]。自1982年美國(guó)科學(xué)院院士Hopfield提出Hopfield遞歸神經(jīng)網(wǎng)絡(luò)以來,對(duì)該模型的相關(guān)研究始終是一個(gè)熱點(diǎn)話題。目前,該神經(jīng)網(wǎng)絡(luò)已經(jīng)廣泛應(yīng)用于優(yōu)化計(jì)算、圖像處理與模式識(shí)別等領(lǐng)域。穩(wěn)定是神經(jīng)網(wǎng)絡(luò)得以應(yīng)用的前提,有著重要的研究意義。因此,關(guān)于Hopfield遞歸神經(jīng)網(wǎng)絡(luò)的穩(wěn)定性分析得到學(xué)者的廣泛關(guān)注[2-8]。
隨著研究的不斷深入,Hopfield遞歸神經(jīng)網(wǎng)絡(luò)模型已經(jīng)得到演化。特別是,學(xué)者考慮了神經(jīng)網(wǎng)絡(luò)實(shí)現(xiàn)過程中存在的時(shí)滯問題[3-8]。一般來說,時(shí)滯是導(dǎo)致網(wǎng)絡(luò)不穩(wěn)定或者系統(tǒng)性能惡化的主要源頭之一。因此,對(duì)神經(jīng)網(wǎng)絡(luò)中不同種類的時(shí)滯穩(wěn)定性分析受到學(xué)者的青睞。另一方面,干擾無處不在,一個(gè)系統(tǒng)的[H∞]性能反映了其自身的抗干擾能力。因此,神經(jīng)網(wǎng)絡(luò)的[H∞]穩(wěn)定性研究有著重要的意義。如文獻(xiàn)[9]考慮了切換神經(jīng)網(wǎng)絡(luò)的[H∞]穩(wěn)定性問題;文獻(xiàn)[10-12]分別分析了時(shí)滯神經(jīng)網(wǎng)絡(luò)的[H∞]狀態(tài)估計(jì)問題;通過設(shè)計(jì)自適應(yīng)控制器,文獻(xiàn)[13-14]探究了時(shí)滯憶阻遞歸神經(jīng)網(wǎng)絡(luò)同步控制問題。
縱觀神經(jīng)網(wǎng)絡(luò)的[H∞]性能分析問題,現(xiàn)有文獻(xiàn)大多是通過設(shè)計(jì)狀態(tài)反饋控制器來實(shí)現(xiàn)系統(tǒng)的[H∞]指標(biāo)[10-14],而忽略了連接權(quán)的更新能力。進(jìn)而,如何結(jié)合神經(jīng)網(wǎng)絡(luò)連接權(quán)的更新能力來探究其[H∞]性能問題,是一個(gè)值得深入思考的話題?;诖嗽颍疚闹饕槍?duì)Hopfield時(shí)滯遞歸神經(jīng)網(wǎng)絡(luò),通過設(shè)計(jì)連接權(quán)矩陣的更新律來使系統(tǒng)達(dá)到[H∞]穩(wěn)定。同時(shí),主從同步控制考慮的是兩個(gè)系統(tǒng)的跟蹤問題,主要是通過對(duì)從系統(tǒng)施加控制,使其與主系統(tǒng)完全同步。本文將關(guān)于[H∞]穩(wěn)定的結(jié)果應(yīng)用于神經(jīng)網(wǎng)絡(luò)的同步控制問題。
1 模型描述
考慮以下時(shí)滯遞歸神經(jīng)網(wǎng)絡(luò)模型
2 主要結(jié)果
本節(jié)將針對(duì)時(shí)滯遞歸神經(jīng)網(wǎng)絡(luò)式(1),設(shè)計(jì)一種連接權(quán)更新律,建立一個(gè)新的LMI [H∞]穩(wěn)定性判據(jù),并將所得結(jié)果推廣到時(shí)滯神經(jīng)網(wǎng)絡(luò)的[H∞]同步控制問題。
2.1 [H∞]穩(wěn)定判據(jù)
5 結(jié)論
針對(duì)帶有參數(shù)擾動(dòng)和干擾輸入的時(shí)滯遞歸神經(jīng)網(wǎng)絡(luò),本文設(shè)計(jì)了連接權(quán)的參數(shù)更新律,并建立使得神經(jīng)元滿足[H∞]性能指標(biāo)的LMI穩(wěn)定性判據(jù)。該判據(jù)可以借助Matlab線性矩陣不等式工具包直接驗(yàn)證并計(jì)算連接權(quán)更新律的增益。與已有的文獻(xiàn)相比,本文所提出的分析方法,體現(xiàn)了連接權(quán)自身的學(xué)習(xí)能力。同時(shí),本文將所得穩(wěn)定性結(jié)果應(yīng)用于時(shí)滯遞歸神經(jīng)網(wǎng)絡(luò)[H∞]同步控制問題,建立了實(shí)現(xiàn)同步的充分判據(jù),給出了控制增益的求解方法。算例仿真說明了本文方法的有效性。本文的結(jié)果可以應(yīng)用于模式識(shí)別、圖像處理和安全通訊等領(lǐng)域。
參考文獻(xiàn):
[1]? ? ZHANG H G,WANG Z S,LIU D R. A comprehensive review of stability analysis of continuous-time recurrent neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems,2014,25(7):1229-1262.
[2]? ? NISHIYAMA K,SUZUKI K. Hinfinity-learning of layered neural networks[J]. IEEE Transactions on Neural Networks,2001,12(6):1265-1277.
[3]? ? FAYDASICOK O. New criteria for global stability of neutral-type Cohen-Grossberg neural networks with multiple delays[J]. Neural Networks,2020,125:330-337.
[4]? ? ARIK S. A modified Lyapunov functional with application to stability of neutral-type neural networks with time delays[J]. Journal of the Franklin Institute,2019,356(1):276-291.
[5]? ? LI Z C,YAN H C,ZHANG H,et al. Stability analysis for delayed neural networks via improved auxiliary polynomial-based functions[J]. IEEE Transactions on Neural Networks and Learning Systems,2019,30(8):2562-2568.
[6]? ? PARK J H. Robust stability of bidirectional associative memory neural networks with time delays[J]. Physics Letters A,2006,349(6):494-499.
[7]? ? 胡進(jìn),宋乾坤. 基于憶阻的時(shí)滯神經(jīng)網(wǎng)絡(luò)的全局穩(wěn)定性[J]. 應(yīng)用數(shù)學(xué)和力學(xué),2013,34(7):724-735.
[8]? ? DING S B,WANG Z S,WU Y M,et al. Stability criterion for delayed neural networks via Wirtinger-based multiple integral inequality[J]. Neurocomputing,2016,214:53-60.
[9]? ? AHN C K. An [H∞] approach to stability analysis of switched Hopfield neural networks with time-delay[J]. Nonlinear Dynamics,2010,60(4):703-711.
[10]? DING S B,WANG Z S,WANG J D,et al. [H∞] state estimation for memristive neural networks with time-varying delays:The discrete-time case[J]. Neural Networks,2016,84:47-56.
[11]? LIN W J,HE Y,ZHANG C K,et al. Stochastic finite-time [H∞] state estimation for discrete-time semi-Markovian jump neural networks with time-varying delays[J]. IEEE Transactions on Neural Networks and Learning Systems,2020,31(12):5456-5467.
[12]? RAKKIYAPPAN R,MAHESWARI K,VELMURUGAN G,et al. Event-triggered [H∞] state estimation for semi-Markov jumping discrete-time neural networks with quantization[J]. Neural Networks,2018,105:236-248.
[13]? WANG J L,QIN Z,WU H N,et al. Finite-time synchronization and [H∞] synchronization of multiweighted complex networks with adaptive state couplings[J]. IEEE Transactions on Cybernetics,2020,50(2):600-612.
[14]? WU H Q,ZHANG X W,LI R X,et al. Adaptive anti-synchronization and [H∞] anti-synchronization for memristive neural networks with mixed time delays and reaction-diffusion terms[J]. Neurocomputing,2015,168:726-740.
[15]? DING S B,WANG Z S. Event-triggered synchronization of discrete-time neural networks:a switching approach[J]. Neural Networks,2020,125:31-40.