Abstract
This paper is concerned with positive solutions and global exponential stability of positive equilibrium of inertial neural networks with multiple time-varying delays. By utilizing the comparison principle via differential inequalities, we first explore conditions on damping coefficients and self-excitation coefficients to ensure that, with nonnegative connection weights and inputs, all state trajectories of the system initiating in an admissible set of initial conditions are always nonnegative. Then, based on the method of using homeomorphisms, we derive conditions in terms of linear programming problems via M-matrices for the existence, uniqueness, and global exponential stability of a positive equilibrium of the system. Two examples with numerical simulations are given to illustrate the effectiveness of the obtained results.
Similar content being viewed by others
References
Soulié FF, Gallinari P (1998) Industrial applications of neural networks. World Scientific Publishing, Singapore
Venketesh P, Venkatesan R (2009) A survey on applications of neural networks and evolutionary techniques in web caching. IETE Tech Rev 26:171–180
Mrugalski M, Luzar M, Pazera M, Witczak M, Aubrun C (2016) Neural network-based robust actuator fault diagnosis for a non-linear multi-tank system. ISA Trans 61:318–328
Witczak P, Patan K, Witczak M, Mrugalski M (2017) A neural network approach to simultaneous state and actuator fault estimation under unknown input decoupling. Neurocomputing 250:65–75
Kiakojoori S, Khorasani K (2016) Dynamic neural networks for gas turbine engine degradation prediction, health monitoring and prognosis. Neural Comput Appl 27:2157–2192
Gong M, Zhao J, Liu J, Miao Q, Jiao J (2016) Change detection in synthesis aperture radar images based on deep neural networks. IEEE Trans Neural Netw Learn Syst 27:125–138
Baldi P, Atiya AF (1995) How delays affect neural dynamics and learning. IEEE Trans Neural Netw 5:612–621
Lu H (2012) Chaotic attractors in delayed neural networks. Phys Lett A 298:109–116
Zhang H, Wang Z, Liu D (2014) A comprehensive review of stability analysis of continuous-time recurrent neural networks. IEEE Trans Neural Netw Learn Syst 25:1229–1262
Liu B (2015) Pseudo almost periodic solutions for CNNs with continuously distributed leakage delays. Neural Process Lett 42:233–256
Arik S (2016) Dynamical analysis of uncertain neural networks with multiple time delays. Int J Syst Sci 47:730–739
Li L, Yang YQ, Lin G (2016) The stabilization of BAM neural networks with time-varying delays in the leakage terms via sampled-data control. Neural Comput Appl 27:447–457
Liu B (2017) Global exponential convergence of non-autonomous SICNNs with multi-proportional delays. Neural Comput Appl 28:1927–1931
Manivannan R, Samidurai R, Sriraman R (2017) An improved delay-partitioning approach to stability criteria for generalized neural networks with interval time-varying delays. Neural Comput Appl 28:3353–3369
Hai-An LD, Hien LV, Loan TT (2017) Exponential stability of non-autonomous neural networks with heterogeneous time-varying delays and destabilizing impulses. Vietnam J Math 45:425–440
Lee TH, Trinh MH, Park JH (2017) Stability analysis of neural networks with time-varying delay by constructing novel Lyapunov functionals. IEEE Trans Neural Netw Learning Syst. https://doi.org/10.1109/TNNLS.2017.2760979
Wheeler DW, Schieve WC (1997) Stability and chaos in an inertial two neuron system. Phys D Nonlin Phenom 105:267–284
Koch C (1984) Cable theory in neurons with active linearized membrane. Biol Cybern 50:15–33
Babcock KL, Westervelt RM (1986) Stability and dynamics of simple electronic neural networks with added inertia. Phys D Nonlin Phenom 23:464–469
Tu Z, Cao J, Hayat T (2016) Matrix measure based dissipativity analysis for inertial delayed uncertain neural networks. Neural Netw 75:47–55
Wan P, Jian J (2017) Global convergence analysis of impulsive inertial neural networks with time-varying delays. Neurocomputing 245:68–76
Tu Z, Cao J, Alsaedi A, Alsaadi F (2017) Global dissipativity of memristor-based neutral type inertial neural networks. Neural Netw 88:125–133
Zhang G, Zeng Z, Hu J (2018) New results on global exponential dissipativity analysis of memristive inertial neural networks with distributed time-varying delays. Neural Netw 97:183–191
Ke Y, Miao C (2013) Stability analysis of inertial Cohen–Grossberg-type neural networks with time delays. Neurocomputing 117:196–205
Zhang Z, Quan Z (2015) Global exponential stability via inequality technique for inertial BAM neural networks with time delays. Neurocomputing 151:1316–1326
Cui N, Jiang H, Hu C, Abdurahman A (2018) Global asymptotic and robust stability of inertial neural networks with proportional delays. Neurocomputing 272:326–333
Tu Z, Cao J, Hayat T (2016) Global exponential stability in Lagrange sense for inertial neural networks with time-varying delays. Neurocomputing 171:524–531
Wang J, Tian L (2017) Global Lagrange stability for inertial neural networks with mixed time-varying delays. Neurocomputing 235:140–146
He X, Huang TW, Yu JZ, Li CD, Li CJ (2017) An inertial projection neural network for solving variational inequalities. IEEE Trans Cybern 47:809–814
Smith H (2008) Monotone dynamical systems: an introduction to the theory of competitive and cooperative systems. American Mathematical Society, Providence
Mózaryn J, Kurek JE (2010) Design of a neural network for an identification of a robot model with a positive definite inertia matrix. In: Artificial Intelligence and Soft Computing. Springer, Berlin
Ma GJ, Wu S, Cai GQ (2013) Neural networks control of the Ni-MH power battery positive mill thickness. Appl Mech Mater 411–414:1855–1858
Lu W, Chen T (2007) \(R^n_+\)-global stability of a Cohen–Grossberg neural network system with nonnegative equilibria. Neural Netw 20:714–722
Liu B, Huang L (2008) Positive almost periodic solutions for recurrent neural networks. Nonlinear Anal Real World Appl 9:830–841
Hien LV (2017) On global exponential stability of positive neural networks with time-varying delay. Neural Netw 87:22–2617
He Y, Ji MD, Zhang CK, Wu M (2016) Global exponential stability of neural networks with time-varying delay based on free-matrix-based integral inequality. Neural Netw 77:80–86
Arino O, Hbid ML, Ait Dads E (2002) Delay differential equations and applications. Springer, Dordrecht
Forti M, Tesi A (1995) New conditions for global stability of neural networks with application to linear and quadratic programming problems. IEEE Trans Circuits Syst-I: Fund 42:354–366
Hien LV, Son DT (2015) Finite-time stability of a class of non-autonomous neural networks with heterogeneous proportional delays. Appl Math Comput 251:14–23
Haykin S (1999) Neural networks: a comprehensive foundation. Prentice Hall, Upper Saddle River
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflicts of interest
The authors declare that no potential conflict of interest to be reported to this work.
Appendices
Appendix 1: Proof of Lemma 3
(Necessity) Let \(\eta _i>0\), \(\xi _i>0\), \(i\in [n]\), satisfying \(D_{\alpha }\succ 0\) and \(D_{\beta }\succ 0\). Then, we have
Observe that \(\xi _i(a_i-\xi _i)-b_i=\frac{1}{4}\left( a_i^2-4b_i\right) -\left( \xi _i-\frac{1}{2}a_i\right) ^2\). Thus,
(Sufficiency) Let condition (11) hold. Then, the constants \(\xi _i^l=\frac{a_i-\sqrt{a_i^2-4b_i}}{2}\) and \(\xi _i^u=\frac{a_i+\sqrt{a_i^2-4b_i}}{2}\) are well defined, \(\xi _i^l>0\), and \(\xi _i^u<a\). In addition, \(\xi _i(a_i-\xi _i)-b_i=(\xi _i^u-\xi _i)(\xi _i-\xi _i^l)>0\) for all \(\xi _i\in (\xi _i^l,\xi _i^u)\). Therefore,
for any \(\xi _i\in (\xi _i^l,\xi _i^u)\). It follows from (29) that \(D_{\alpha }\succ 0\) and \(D_{\beta }\succ 0\) for any \(\eta _i>0\) and \(\xi _i\in (\xi _i^l,\xi _i^u)\). The proof is completed.
Appendix 2: Proof of Theorem 1
For any initial condition in \({\mathscr {A}}_T\), it suffices to prove that the corresponding solution \(z(t)=(x^{\top }(t),y^{\top }(t))^{\top }\) of (5) is positive. To this end, we note at first that if \(y_i(t)\ge 0\), \(t\in [0,t_f)\), for some \(t_f>0\), then from the first equation in (5), we have
Thus, it is only necessary to show that \(y(t)=(y_i(t))\succeq 0\) for all \(t\ge 0\).
For a given \(\epsilon >0\), let \(z_{\epsilon }(t)=(x^{\top }_{\epsilon }(t),y^{\top }_{\epsilon }(t))^{\top }\) be solution (5) with initial functions \(\phi _i(.)\) and \(\psi _{i\epsilon }(.)=\psi _i(.)+\epsilon\). Since \(\phi (s)=(\phi _i(s))\succeq 0\) and \(\psi _{\epsilon }(s)=(\psi _{i\epsilon }(s))\succeq \epsilon {\mathbf {1}}_n\) for all \(s\in [-\tau ^+,0]\), where \({\mathbf {1}}_n\) denotes the vector in \({\mathbb {R}}^n\) with all entries equal one, it follows from (5) that \(y_{\epsilon }(t)=(y_{i\epsilon }(t))\succ 0\), \(t\in [0,t_f)\), for some small \(t_f>0\). By virtue of the contrary argument method, we assume that there exist a \(t_1>0\) and an index \(i\in [n]\) such that
and \(y_{\epsilon }(t)\succeq 0\) for all \(t\in [0,t_1]\). Then, by multiplying with \(e^{\alpha _it}\), it follows from (5) that
It follows from (31) and \(D_{\beta }e^{D_{\alpha }t}x_{\epsilon }(t)\succeq 0\), \(t\in [0,t_1]\), that
where \({\hat{f}}({\hat{x}}_{\epsilon }(s))=(f_j(x_{j\epsilon }(s-\tau _j(s)))\) and \(E_n\) is the identity matrix in \({\mathbb {R}}^{n\times n}\).
Since the vector fields \(F_1(x)=Cf(x)\) and \(F_2(x)=Df(x)\) are order-preserving, \(x_{\epsilon }(t)\succeq 0\) and \({\hat{x}}_{\epsilon }(t)\succeq 0\) for \(t\in [0,t_1]\), from (32), we have
where \(D_{\alpha \eta }=D_{\alpha }D_{\eta }\). Let \(t\uparrow t_1\) in (33), and note also that \(E_n-e^{-D_{\alpha }t_1}\succ 0\), we readily obtain
which gives a contradiction with (30). Therefore, \(x_{\epsilon }(t)\succeq 0\) and \(y_{\epsilon }(t)\succ 0\) for all \(t\ge 0\). Let \(\epsilon \downarrow 0\), we then obtain \(z(t)=\lim _{\epsilon \rightarrow 0}z_{\epsilon }(t)\succeq 0\). The proof is completed.
Appendix 3: Proof of Theorem 2
We define the following mappings
For any vectors \(\chi _1=\begin{bmatrix}x_1\\ y_1\end{bmatrix}\) and \(\chi _2=\begin{bmatrix}x_2\\ y_2\end{bmatrix}\) in \({\mathbb {R}}^{2n}\), we have
Therefore,
where \({\mathscr {S}}(x_1-x_2)={{\mathrm{diag}}}\{{{\mathrm{sgn}}}(x_{1i}-x_{2i})\}\).
By assumption (A1), and \(D_{\alpha \xi }-B=D_{\beta }D_{\eta }\succ 0\), similar to (34), we also have
where \(L_f={{\mathrm{diag}}}\{l_i^f\}\). From (34) and (35), we have
where \({\mathscr {M}}=\begin{pmatrix}D_{\xi }&-D_{\eta }\\ B-D_{\alpha \xi }-(C+D)L_f&D_{\alpha \eta }\end{pmatrix}\). On the other hand, it is clear that
Thus, combining (36) and (37) gives
Let \(\chi _0\) be a positive vector satisfying (14). Then, from (38), we have
If \({\mathscr {H}}(\chi _1)-{\mathscr {H}}(\chi _2)=0\) then, by (14) and (39), \(\chi _1=\chi _2\). This shows that the mapping \({\mathscr {H}}(.)\) is injective in \({\mathbb {R}}^{2n}\). In addition to this, let \(\chi _2=0\), from (39), we have
which ensures that \(\Vert {\mathscr {H}}(\chi _k)\Vert _{\infty }\rightarrow \infty\) for any sequence \(\{\chi _k\}\) in \({\mathbb {R}}^{2n}\) satisfying \(\Vert \chi _k\Vert _{\infty }\rightarrow \infty\). Thus, the continuous mapping \({\mathscr {H}}(.)\) is proper. By Lemma 2, \({\mathscr {H}}(.)\) is a homeomorphism onto \({\mathbb {R}}^{2n}\). Consequently, the equation \({\mathscr {H}}(\chi )=0\) has a unique solution \(\chi _*\in {\mathbb {R}}^{2n}\), which is a unique EP of (5).
Rights and permissions
About this article
Cite this article
Van Hien, L., Hai-An, L.D. Positive solutions and exponential stability of positive equilibrium of inertial neural networks with multiple time-varying delays. Neural Comput & Applic 31, 6933–6943 (2019). https://doi.org/10.1007/s00521-018-3536-8
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00521-018-3536-8