On stability criteria for neural networks with time-varying delay using Wirtinger-based multiple integral inequality
Introduction
Neural networks have substantial capacity in many scientific areas such as signal processing, image decryption, pattern recognition, associative memories, fixed-point computations, optimization, feedback control, medical diagnosis, and financial applications [1], [2], [3]. It is well-known that the existence of time-delays often causes undesirable dynamic behavior, deteriorates system performance, and destroys stability. The occurrence of time-delay in neural networks cannot be avoided, due to the finite switching speed of amplifiers, the inherent communication time between the neurons, and so on. Thus, many researchers have paid their attention to analyze the stability of time-delayed neural networks [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19]. Until now, in order to reduce the conservatism of stability criteria, various methods are developed such as Park's inequality [20], multiple integral approach [21], model transformation [22], free-weighting matrices techniques [23], [24], convex combination technique [25], reciprocally convex optimization [26], and delay-partitioning approach [8], [9], [10], [27], [28].
It should be pointed out that, in the stability analysis on time-delayed neural networks based on Lyapunov approaches, Jensen's inequality is an essential technique to reduce conservatism of stability criteria, so it has been utilized for estimating upper bound of time derivative of constructed Lyapunov functional in most existing papers. It is clear that reducing Jensen's gap gives more feasible region of the stability of time-delay systems. However, until now, the following Jensen's gap is commonly used: where , a and b are positive constants satisfying . Only recently, there are a few works [29], [30], [31] to reduce Jensen's gap. In [29], [30], based on Wirtinger's inequalities, more closed lower bound of Jensen's inequality is proposed as where , and in order to apply Wirtinger inequalities which have some necessary assumptions, the choice of a particular signal has been proved, in which, an integration of states, , is involved in augmented vectors to make use of further information of the time-delay systems. Moreover, very recently, an improved result which reduces more Jensen's gap as has been presented in [31]. From the above, it is clear that , so the use of gives more feasible region than the use of and .
After presented Wirtinger-based integral approach in [31], many researchers have used it for the less conservative results on time-delay criteria [32], [33], [34], [35], [36], [37] In [34], the problem of stability analysis for neutral type neural networks with discrete and distributed delays was investigated by using the multiple integral, Wirtinger-based integral, and delay fragmentation approach. In [35], a criterion for the stability of recurrent neural networks with time-varying delay was derived by employing a triple integral term as an augmented Lyapunov–Krasovskii functional, and in order to derive the criterion, Wirtinger-based integral inequality and two zero-value free-weighting matrix equations are used. The authors in [36] studied the robust stochastic stability problem for a class of neutral-type uncertain neural networks with Markovian jumping parameters and time-varying delays in which the delay partitioning and Wirtinger-based integral inequality techniques were used to construct augmented Lyapunov–Krasovskii functional including several triple integral terms. Shao et al. [37] have proposed an filter design result which is mode- and delay-dependent for static neural networks with Markovian jumping parameters and time-varying delay, here, both triple integral terms in Lyapunov functional and Wirtinger-based integral approach were utilized as well. As seen in the above existing literature, however, Wirtinger-based integral approach was applied to only single integral term without any modifications. In those works, both double or multiple integral and Wirtinger-based integral approaches are employed together but used them individually, i.e. double or multiple integral terms were dealt with by Jensen's inequality approach even though Wirtinger-based integral approach gives more tighter bound than Jensen's one.
On the other hand, there are a few attempts to extend Wirtinger-based integral inequality recently. An attempt was made to combine it with delay fragmentation approach for time-delay systems in [38], in which every partitioned single integral terms were applied to Wirtinger-based integral approach, and less conservative results of the developed criterion were given by comparing other works. Authors of [29], [30], [31] proposed more general inequality to reduce Jensen's gap by using Bessel–Legendre inequality for the constant delay case [39]. Very recently, there was a work which is an extension of Wirtinger-based integral inequality to a double integral term [40].
Inspired by the above discussion, this paper mainly focuses on presenting a novel approach called Wirtinger-based multiple integral inequality for neural networks with time-varying delay. Based on a lemma in [40], a novel inequality which is multiple integral form of Wirtinger-based integral inequality is supposed in Lemma 5. In addition, in order to apply the derived new inequality to time-varying delayed neural networks, the reciprocally convex combination approach is generalized for high order case. Numerical examples are given to illustrate that the proposed methods are effective and lead to less conservative results.
Notations: is the n-dimensional Euclidean space, (respectively, 0) means that the matrix X is a real symmetric positive definite matrix (respectively, positive semi-definite). in a matrix represents the elements below the main diagonal of a symmetric matrix. In and denote the n×n and n×m dimensional identity and zero matrix, respectively. denotes block diagonal matrix. indicates . means that the elements of the matrix X include the values of f(t). For , denotes a basis for the null-space of X. indicates binomial coefficient.
Section snippets
Preliminaries
Consider the following time-varying delayed neural networks:where is the neuron state vector, n is the number of neurons in a neural network, means the neuron activation function, is a positive diagonal matrix, and are the interconnection matrices representing the weight coefficients of the neurons. The delay h(t) is a time-varying continuous function satisfying
Main results
Before proceeding further, are defined as block entry matrices, e.g. . Theorem 1 For given positive scalars h and , any scalar μ, and diagonal matrices and , the system (1) is asymptotically stable, if there exist positive definite matrices , (i=1,2), (), positive diagonal matrices ,…,, (i=1,2), and any matrices (i=1, …, M, j
Numerical example
In this section, three numerical examples are provided to show the improvement of the stability criteria. In the simulations, the results of unknown μ can be obtained by Theorem 1 with . Example 1 Consider the neural networks (1) with the following parameters: With the above parameters, Table 1 is given by applying Theorem 1, which shows the results of maximum bound of time-delay with various μ. Example 2 Consider the neural networks (1) with the
Conclusions
In this paper, the problem of delay-dependent stability analysis of neural networks with time-varying delay has been discussed. Based on Lemma 4 which is the Wirtinger-based integral inequality for a double integral term, a new lemma, Wirtinger-based multiple integral inequality, has been proposed in Lemma 5. By generalizing lower bound lemma for high order case, a new inequality, Lemma 5, can be applied to neural networks with time-varying delay. Three numerical examples have been given to
Acknowledgments
This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2013R1A1A2A10005201).
References (42)
- et al.
Improved delay-dependent stability criterion for neural networks with time-varying delay
Appl. Math. Comput.
(2011) - et al.
Improved stability criteria for neural networks with time-varying delay
Phys. Lett. A
(2009) - et al.
Global asymptotic stability analysis for delayed neural networks using a matrix-based quadratic convex approach
Neural Netw.
(2014) - et al.
Stability and dissipativity analysis of static neural networks with interval time-varying delay
J. Frankl. Inst.
(2015) An analysis of stability of neutral-type neural systems with constant time delays
J. Frankl. Inst.
(2014)- et al.
A multiple integral approach to stability of neutral time-delay systems
Appl. Math. Comput.
(2013) - et al.
Stability analysis for neutral systems with mixed delays
J. Comput. Appl. Math.
(2007) - et al.
Stability and robust stability for systems with a time-varying delay
Automatica
(2007) - et al.
Reciprocally convex approach to stability of systems with time-varying delays
Automatica
(2011) - et al.
Stability of stochastic neural networks of neutral type with Markovian jumping parametersa delay-fractioning approach
J. Frankl. Inst.
(2014)
Wirtinger-based integral inequalityapplication to time-delay systems
Automatica
Improved results on stability of linear systems with time-varying delays via Wirtinger-based integral inequality
J. Frankl. Inst.
Discrete Wirtinger-based inequality and its application
J. Frankl. Inst.
New stability analysis for neutral type neural networks with discrete and distributed delays using a multiple integral approach
J. Frankl. Inst.
Novel stability criteria for recurrent neural networks with time-varying delay
Neurocomputing
Improved delay-dependent robust stability analysis for neutral-type uncertain neural networks with Markovian jumping parameters and time-varying delays
Neurocomputing
Filter design of delayed static neural networks with Markovian jumping parameters
Neurocomputing
Stability of time-delay systems via Wirtinger-based double integral inequality
Automatica
Second-order reciprocally convex approach to stability of systems with interval time-varying delays
Appl. Math. Comput.
Neural Networks: A Comprehensive Foundation
Neural Networks for Optimization and Signal Processing
Cited by (81)
New results on stability analysis for a class of generalized delayed neural networks
2024, Applied Mathematics and ComputationMultiple integral techniques for stochastic stability analysis of Markovian jump systems: A unified uncertain transition rates
2022, Journal of the Franklin InstituteMemory feedback finite-time control for memristive neutral-type neural networks with quantization
2021, Chinese Journal of PhysicsStability analysis for delayed neural networks based on the augmented Lyapunov-Krasovskii functional with delay-product-type and multiple integral terms
2020, NeurocomputingCitation Excerpt :However, the existence of time delay may lead to instability or other unacceptable dynamic behaviors [11,12]. Thus, stability analysis of delayed neural networks (DNNs) has received extensive attention [13–20]. Lyapunov method is one of the most fruitful fields in the stability analysis of time-delay systems.