Global Exponential Stability of a Class of Memristor-Based RNN and Its Application to Design Stable Voltage Circuits

In this article, we study the global exponential stability of the equilibrium point for a class of memristor-based recurrent neural networks (MRNNs). The MRNNs are based on a realistic memristor model and can be implemented by a very large scale of integration circuits. By introducing a proper Lyapunov functional, it is proved that the equilibrium point of the MRNN is globally exponentially stable under two less conservative assumptions. Furthermore, an algorithm is proposed for the design of MRNN-based circuits with stable voltages. Finally, an illustration example is performed to show the validation of the proposed theoretical results; an MRNN-based circuit with stable voltages is designed according to the proposed algorithm.

1 Introduction

Recurrent networks have been one of the necessary tools to character system states since their wide applications in optimization (Li et al., 2021; Ma and Bian, 2021), games (Wu et al., 2019, 2021; Cheng et al., 2021), control (Yang et al., 2015; Jianmin et al., 2021; Toyoda and Wu, 2021), and so on (Wang et al., 2007; Shen et al., 2020; Shen and Raksincharoensak, 2021). In recent years, a new type of recurrent network was proposed based on a new two-terminal circuit element called the memristor (Chua, 1971; Strukov et al., 2008). Note that a memristor works like a biological synapse (Anthes, 2011; Qin et al., 2015) and has the ability of automatic information storage. Thus, memristors replaced resistors as synapses in recurrent neural networks, that is, memristor-based recurrent neural networks (MRNNs) (Anthes, 2011; Wen et al., 2013; Zhang et al., 2013). In recent years, the stability and stabilization of Boolean networks have been extensively investigated (Chen et al., 2018; Guo et al., 2019, 2021).

MRNNs have been a promising architecture in neuromorphic systems by virtue of their non-volatility, high-density, and physical storable feature. According to the realistic structure of MRNNs, several different mathematical models for MRNNs were proposed (Hu and Wang, 2010; Wu et al., 2011; Li et al., 2014; Chen et al., 2015; Jianmin et al., 2019). Meanwhile, notice that the MRNN, a special recurrent network, depends on the stability of its equilibrium points in application scenarios. Therefore, many interesting works were addressed to analyze the stability for the MRNNs (Hu and Wang, 2010; Wu et al., 2011; Li et al., 2014; Chen et al., 2015; Jianmin et al., 2019). A mathematical model of MRNN was proposed, and its global uniform asymptotic stability was investigated in a Lyapunov sense (Hu and Wang, 2010). A simple model of MRNN was introduced by Wu et al. (2011) by means of the typical current–voltage characteristics of memristors. A stochastic MRNN was proposed by Li et al. (2014) based on the work by Wang et al. (2007), in which therewas some unavoidable noise in real networks. Furthermore, the global exponential stability for the stochastic MRNN was studied under the framework of Filppov’s solution; three sufficient conditions with the form of linear inequalities were provided to determine the global exponential stability of the stochastic MRNN. The global asymptotic stability and synchronization of a class of fractional-order memristor-based delayed neural networks were investigated by Chen et al. (2015). The existence and global exponential stability were discussed by Jianmin et al. (2019) for an uncertain MRNN with mixed time delay under two assumptions.

Motivated by the aforementioned works, the global exponential stability of the equilibrium point is investigated for a class of MRNNs with time-varying delay, and its application to stabilize the voltage in a circuit network is carried out in this study. A sufficient condition is obtained for the global exponential stability of MRNNs. Based on this condition, an algorithm is proposed to stabilize the voltage of the MRNN-based circuit. The time-varying delay was considered in the activation functions of MRNN in this study. In addition, the activation functions in the MRNN are not necessarily non-decreasing, while the activation functions are non-decreasing in the works by Hu and Wang (2010); Wu et al. (2011); Li et al. (2014); Chen et al. (2015); Jianmin et al. (2019). Thus, the MRNN considered in this study is the extension from the view of activation functions compared with those in the works by Hu and Wang (2010); Wu et al. (2011); Li et al. (2014); Chen et al. (2015); Jianmin et al. (2019). Meanwhile, the stable voltage is a necessary prerequisite for obtaining high-quality electric energy in power systems, such as wind power converters (Kobravi et al., 2007). Consequently, the obtained theretical results are successfully applied to design the MRNN-based circuit system with global exponential stability, which makes it possible to apply the MRNN to power converters.

The structure of this article is given as follows: an MRNN with time-varying delay and some notations is introduced in Section 2. In Section 3, the global exponential stability of the equilibrium point for the MRNN is obtained, and an example is given to show the effectiveness of the obtained results. Then, an algorithm to design the MRNN-based circuit with stable voltage is proposed, and a simple application is carried out in Section 4. Finally, the main conclusions are given in Section 5.

2 Memristor-Based Recurrent Neural Network

In this section, some notations are introduced, and an MRNN is described under two assumptions based on the mathematical models by Wen et al. (2013); Jianmin et al. (2019).

Notation: $\mathbb{R}$ denotes the set of real numbers. $x={(x_1,x_2,\dots ,x_m)}^T$ is an m − dimensional column, and the superscript T stands for the transpose operator. $\Vert x\Vert ≔{({\sum }_{i=1}^m{x}_i^2)}^{1/2}$. $A=(a_{ij})\in {\mathbb{R}}^{m\times m}$ is a matrix. $\Vert A\Vert =\sqrt{{\lambda }_M(A^{T}A)}$, where λ_M(A) represents the maximum eigenvalue of A. $I\in {\mathbb{R}}^{m\times m}$ stands for an identity matrix. For a real symmetric matrix A, A > 0 (A < 0) means that A is positive (negative) definite.

Consider the following MRNN, which was originated from Wen et al. (2013),

$\begin{array}{cc}\hfill C_i{\dot{x}}_i\left(t\right)=& -\left[\sum _{j=1}^n\left(\frac{1}{R_{fij}}+\frac{1}{R_{gij}}\right)+W_i\left(x_i\left(t\right)\right)\right]x_i\left(t\right)\hfill \\ \hfill & +\sum _{j=1}^n\frac{{\mathrm{sign}}_{ij}}{R_{fij}}{f}_j\left(x_j\left(t\right)\right)+\sum _{j=1}^n\frac{{\mathrm{sign}}_{ij}}{R_{gij}}{g}_j\left(x_j\left(t-{\tau }_j\left(t\right)\right)\right)+I_i.\hfill \\ \hfill & \hfill \end{array}(1)$

Here, f_j (⋅) is the activation function, τ_j (⋅) is the time-varying delay, C_i is the capacitance of the capacitor, and x_i(t) is the voltage of the capacitor. $R_{f_{ij}}$ is the resistor between the feedback function f_j (x_j(t)) and the state x_i(t), and $R_{g_{ij}}$ is the resistor between the feedback function g_j (x_j (t − τ_j(t))) and the state x_i(t). sign_ij is defined as

${\mathrm{sign}}_{ij}=\left\{\begin{array}{cccc}\hfill & 1,\hfill & \hfill & \text{if\hspace{0.17em}}i\ne j;\hfill \\ \hfill & 0,\hfill & \hfill & \text{if\hspace{0.17em}}i=j.\hfill \end{array}\right.(2)$

W_i[x_i(t)] is the memductance of the i − th memristor satisfying

$W_i\left(x_i\left(t\right)\right)=\left\{\begin{array}{cccc}\hfill & W_i^{\prime },\hfill & \hfill & \text{if\hspace{0.17em}}{x}_i\left(t\right)\le 0;\hfill \\ \hfill & W_i^{\prime \prime },\hfill & \hfill & \text{if\hspace{0.17em}}{x}_i\left(t\right)>0.\hfill \end{array}\right.(3)$

I_i is an external input or bias and i, j = 1, 2, …, n. Let

$\widetilde{W_i}=\frac{W_i^{\prime \prime }-W_i^{\prime }}{2C_i}.(4)$

From Jianmin et al. (2019), the MRNN (Eq. 1) is transformed into:

$\begin{array}{cc}\hfill {\dot{x}}_i\left(t\right)& =-d_i{x}_i\left(t\right)-\widetilde{W_i}|x_i\left(t\right)|+\sum _{j=1}^n{a}_{ij}{f}_j\left(x_j\left(t\right)\right)\\ & +\sum _{j=1}^n{b}_{ij}{g}_j\left(x_j\left(t-{\tau }_j\left(t\right)\right)\right)+U_i.\end{array}(5)$

Here,

$\begin{array}{cc}\hfill d_i& =\sum _{j=1}^n\left[\frac{1}{C_i{R}_{f_{ij}}}+\frac{1}{C_i{R}_{g_{ij}}}+\frac{W_i^{\prime }+W_i^{\prime \prime }}{2C_i}\right],\hfill \\ \hfill a_{ij}& =\frac{{\mathrm{sign}}_{ij}}{C_i{R}_{f_{ij}}},b_{ij}=\frac{{\mathrm{sign}}_{ij}}{C_i{R}_{g_{ij}}},U_i=\frac{I_i}{C_i}.\hfill \end{array}(6)$

Next, let D = diag{d₁, d₂, …, d_n}, $\tilde{W}=\mathrm{diag}\{\widetilde{W_1},\widetilde{W_2},\dots ,\widetilde{W_n}\}$, $A={(a_{ij})}_{n\times n}$, $B={(b_{ij})}_{n\times n}$, $|x(t)|={(|x_1(t)|,|x_2(t)|,\dots ,|x_n(t)|)}^T$, $\tau (t)={({\tau }_1(t),{\tau }_2(t),\dots ,{\tau }_n(t))}^T$, and $U={(U_1,U_2,\dots ,U_n)}^T.$ Then, Eq. 5 is rewritten as:

$\begin{array}{c}\hfill \dot{x}\left(t\right)=-Dx\left(t\right)-\tilde{W}|x\left(t\right)|+Af\left(x\left(t\right)\right)+Bg\left(x\left(t-\tau \left(t\right)\right)\right)+U.\end{array}(7)$

In addition, there are two assumptions and one lemma, which will be needed in the sequel, for the MRNN (Eq. 7). The first assumption about the activation function f_i is from Wen et al. (2013). The second assumption about the time-varying delay τ_j is from Wen et al. (2013).

S₁. For i ∈ {1, 2, …, n}, the activation function f_i is bounded continuous, and $\forall r_1,r_2\in \mathbb{R}$, there exists real number l_i > 0 such that

$0\le \frac{f_i\left(r_1\right)-f_i\left(r_2\right)}{r_1-r_2}\le l_i.(8)$

Here, we set L_f = diag{l₁, l₂, …, l_n}.

For i ∈ {1, 2, …, n}, the activation function g_i is bounded continuous, g_i (0) = 0, and $\forall r_1,r_2\in \mathbb{R}$, there exists real number $l_i^{\prime }>0$ such that

$-l_i^{\prime }\le \frac{g_i\left(r_1\right)-g_i\left(r_2\right)}{r_1-r_2}\le l_i^{\prime }.(9)$

Here, we set $L_g=\mathrm{diag}\{l_1^{\prime },l_2^{\prime },\dots ,l_n^{\prime }\}$.

S₂. For i ∈ {1, 2, …, n}, τ_i(t) satisfies

$0\le {\tau }_i\left(t\right)\le {\bar{\tau }}_i,{\dot{\tau }}_i\left(t\right)\le {\mu }_i<1.(10)$

Here, we let $\bar{\tau }=\max \{{\bar{\tau }}_1,\dots ,{\bar{\tau }}_n\}$, and μ = max{μ₁, …, μ_n}.

Remark 1. From Eq. 9, the activation functions g_i [x_i(t)] are non-monotonic in this study. On the other hand, we notice that the activation functions of MRNNs in the works by Hu and Wang (2010; Wu et al. (2011); Li et al. (2014); Chen et al. (2015); Jianmin et al. (2019) are non-decreasing. Thus, Eq. 1 is the extension from the view of activation functions compared with those references.

3 Globally Exponential Stability

In this section, we will prove that the MRNN (Eq. 1) is globally exponentially stable under the assumptions S₁ and S₂. A sufficient condition with the form of linear matrix inequalities can be obtained for globally exponential stability of MRNN by constructing a suitable Lyapunov functional.

Theorem 1. Assume that S₁ and S₂ hold. If there exist a matrix P = diag{p₁, p₂, …, p_n} > 0, a constant k > 0, and small enough constants ξ > 0 and ϑ > 0 such that

$\begin{array}{cc}\hfill & \mathrm{\Phi }≔\xi P-2PD+2P|\tilde{W}|+\left(1+\frac{1}{k}\right)I+2\vartheta \left(\xi L_{f}P+L_{f}P|\tilde{W}|\right)\hfill \\ & +\frac{\left(1+\vartheta \right)}{1-\mu }\Vert PB{\Vert }^2{e}^{\xi \bar{\tau }}{L}_g{L}_g<0,\\ \hfill & \mathrm{\Psi }≔-2\vartheta L_f^{-1}PD+\vartheta \left(PA+A^{T}P\right)+\vartheta I+k\Vert PA{\Vert }^{2}I<0.\hfill \end{array}(11)$

Then, the equilibrium point of the MRNN (Eq. 1) is globally exponentially stable.

Proof. To simplify the proof, we make the following transformation:

$\begin{array}{c}\hfill z=x-x^{*},\end{array}(12)$

where $x^{*}$ is the equilibrium point of the MRNN (Eq. 1). Then, the MRNN (Eq. 1) can be rewritten equivalently as

$\begin{array}{cc}\hfill \dot{z}\left(t\right)& =-Dz\left(t\right)-\tilde{W}\left(|z\left(t\right)+x^{*}|-|x^{*}|\right)\\ & +A\mathcal{f}\left(z\left(t\right)\right)+B\mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right),\end{array}(13)$

where $\mathcal{f}$(z(t)) = f(z(t) + x^*) − f (x^*) and $\mathcal{g}$(z(t − τ(t))) = g(z(t − τ(t)) + x^*) − g (x^*). It is obvious that $\mathcal{f}$_i (0) = 0 and $\mathcal{g}$_i (0) = 0. By the assumption S₁, we get

$\begin{array}{cc}\hfill & {\mathcal{f}}^T\left(z\left(t\right)\right)z\left(t\right)\ge {\mathcal{f}}^T\left(z\left(t\right)\right)L_f^{-1}\mathcal{f}\left(z\left(t\right)\right),\hfill \\ \hfill & {\mathcal{f}}^T\left(z\left(t\right)\right)z\left(t\right)\le z^T\left(t\right)L_{f}z\left(t\right),\hfill \\ \hfill & \mathcal{g}\left(z\left(t\right)\right)\le L_g|z\left(t\right)|.\hfill \end{array}(14)$

We define a Lyapunov functional as follows:

$\mathcal{V}\left(t,z\right)={\mathcal{V}}_0\left(t,z\right)+{\mathcal{V}}_1\left(t,z\right)+{\mathcal{V}}_2\left(t,z\right),(15)$

where

$\begin{array}{cc}\hfill {\mathcal{V}}_0\left(t,z\right)& =e^{\xi t}{z}^{T}Pz,\hfill \\ \hfill {\mathcal{V}}_1\left(t,z\right)& =2\vartheta e^{\xi t}\sum _{i=1}^n{p}_i{\int }_0^{z_i}{\mathcal{f}}_i\left(s\right)ds,\hfill \\ \hfill {\mathcal{V}}_2\left(t,z\right)& =\eta \sum _{i=1}^n{\int }_{t-{\tau }_i\left(t\right)}^t{\mathcal{g}}_i^2\left(z_i\left(s\right)\right)e^{\xi \left(s+{\bar{\tau }}_i\right)}ds.\hfill \end{array}(16)$

Here, ξ, ϑ are small positive constants, and η is a positive constant to be determined.First, calculating the time derivative of ${\mathcal{V}}_0(t,z)$ along the trajectories of the MRNN (Eq. 13), we have

$\begin{array}{cc}\hfill \frac{d}{dt}{\mathcal{V}}_0\left(t,z\left(t\right)\right)& =\xi e^{\xi t}{z}^T\left(t\right)Pz\left(t\right)+2e^{\xi t}{z}^T\left(t\right)P\dot{z}\left(t\right)\hfill \\ \hfill & =\xi e^{\xi t}{z}^T\left(t\right)Pz\left(t\right)-2e^{\xi t}{z}^T\left(t\right)PDz\left(t\right)\hfill \\ & -2e^{\xi t}{z}^T\left(t\right)P\tilde{W}\left(|z\left(t\right)+x^{*}|-|x^{*}|\right)\\ \hfill & +2e^{\xi t}{z}^T\left(t\right)PA\mathcal{f}\left(z\left(t\right)\right)\hfill \\ & +2e^{\xi t}{z}^T\left(t\right)PB\mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right).\end{array}(17)$

In addition,

$\begin{array}{cc}\hfill -2e^{\xi t}{z}^T\left(t\right)P\tilde{W}\left(|z\left(t\right)+x^{*}|-|x^{*}|\right)& \le 2e^{\xi t}|z\left(t\right){|}^{T}P|\tilde{W}|\hfill \\ & \times |\left(|z\left(t\right)+x^{*}|-|x^{*}|\right)|\\ \hfill & \le 2e^{\xi t}|z\left(t\right){|}^{T}P|\tilde{W}|\cdot |z\left(t\right)|\hfill \\ \hfill & =2e^{\xi t}{z}^T\left(t\right)P|\tilde{W}|z\left(t\right),\hfill \end{array}(18)$

$\begin{array}{cc}& 2e^{\xi t}{z}^T\left(t\right)PA\mathcal{f}\left(z\left(t\right)\right)\\ & \le e^{\xi t}\left[\frac{1}{k}{z}^T\left(t\right)z\left(t\right)+{\mathcal{f}}^T\left(z\left(t\right)\right)k\Vert PA{\Vert }^2\mathcal{f}\left(z\left(t\right)\right)\right],\end{array}(19)$

$\begin{array}{cc}2e^{\xi t}{z}^T\left(t\right)PB\mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right)& \le e^{\xi t}\left[z^T\left(t\right)z\left(t\right)+{\mathcal{g}}^T\left(z\left(t-\tau \left(t\right)\right)\right)\right.\\ & \times \left.\Vert PB{\Vert }^2\mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right)\right].\end{array}(20)$

Here, the parameter k is a positive constant. Substituting Eqs 18–20 into Eq. 17, we obtain

$\begin{array}{cc}\hfill \frac{d}{dt}{\mathcal{V}}_0\left(t,z\left(t\right)\right)& \le e^{\xi t}{z}^T\left(t\right)\left[\xi P-2PD+2P|\tilde{W}|+\left(1+\frac{1}{k}\right)I\right]z\left(t\right)\hfill \\ \hfill & +e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)k\Vert PA{\Vert }^2\mathcal{f}\left(z\left(t\right)\right)+e^{\xi t}{\mathcal{g}}^T\left(z\left(t-\tau \left(t\right)\right)\right)\hfill \\ & \times \Vert PB{\Vert }^2\mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right).\end{array}(21)$

Second, by calculating the time derivative of ${\mathcal{V}}_1(t,z)$ along the trajectories of the MRNN (Eq. 13), it follows

$\begin{array}{cc}\hfill \frac{d}{dt}{\mathcal{V}}_1\left(t,z\left(t\right)\right)& =2\xi \mathrm{\vartheta }{e}^{\xi t}\sum _{i=1}^n{p}_i{\int }_0^{z_i}{\mathcal{f}}_i\left(s\right)ds+2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)P\dot{z}\left(t\right)\hfill \\ \hfill & \le 2\xi \vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)Pz\left(t\right)-2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)PDz\left(t\right)\hfill \\ & -2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)P\tilde{W}\left(|z\left(t\right)+x^{*}|-|x^{*}|\right)\\ \hfill & +2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)PA\mathcal{f}\left(z\left(t\right)\right)+2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)PB\hfill \\ & \times \mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right).\end{array}(22)$

By Eq. 14, we have

$\begin{array}{cc}\hfill & 2\xi \vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)Pz\left(t\right)\le 2\xi \vartheta e^{\xi t}{z}^T\left(t\right)L_{f}Pz\left(t\right),\hfill \\ \hfill & -2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)PDz\left(t\right)\le -2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)L_f^{-1}PD\mathcal{f}\left(z\left(t\right)\right),\hfill \\ \hfill & \hfill \end{array}(23)$

$\begin{array}{cc}\hfill & -2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)P\tilde{W}\left(|z\left(t\right)+x^{*}|-|x^{*}|\right)\hfill \\ \hfill & \le 2\vartheta e^{\xi t}|z\left(t\right){|}^T{L}_{f}P|\tilde{W}\Vert z\left(t\right)|\hfill \\ \hfill & =2\vartheta e^{\xi t}{z}^T\left(t\right)L_{f}P|\tilde{W}|z\left(t\right).\hfill \end{array}(24)$

Notice that

$\begin{array}{cc}\hfill & 2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)PB\mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right)\hfill \\ \hfill & \le \vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)\mathcal{f}\left(z\left(t\right)\right)+e^{\xi t}{\mathcal{g}}^T\left(z\left(t-\tau \left(t\right)\right)\right)\vartheta \Vert PB{\Vert }^2\hfill \\ \hfill & \times \mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right).\hfill \end{array}(25)$

Substituting Eqs 23–25 into Eq. 22, we have

$\begin{array}{cc}\hfill & \frac{d}{dt}{\mathcal{V}}_1\left(t,z\left(t\right)\right)\hfill \\ \hfill & \le 2\xi \vartheta e^{\xi t}{z}^T\left(t\right)L_{f}Pz\left(t\right)-2e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)\vartheta L_f^{-1}PD\mathcal{f}\left(z\left(t\right)\right)\hfill \\ \hfill & +2\vartheta e^{\xi t}{z}^T\left(t\right)L_{f}P|\tilde{W}|z\left(t\right)+2\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)PA\mathcal{f}\left(z\left(t\right)\right)\hfill \\ \hfill & +\vartheta e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)\mathcal{f}\left(z\left(t\right)\right)+e^{\xi t}{\mathcal{g}}^T\left(z\left(t-\tau \left(t\right)\right)\right)\vartheta \Vert PB{\Vert }^2\hfill \\ & \times \mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right)\\ \hfill & =2\vartheta e^{\xi t}{z}^T\left(t\right)\left[\xi L_{f}P+L_{f}P|\tilde{W}|\right]z\left(t\right)+2e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)\hfill \\ & \times \left[-\vartheta L_f^{-1}PD+\vartheta PA+\frac{1}{2}\vartheta I\right]\mathcal{f}\left(z\left(t\right)\right)\\ \hfill & +e^{\xi t}{\mathcal{g}}^T\left(z\left(t-\tau \left(t\right)\right)\right)\vartheta \Vert PB{\Vert }^2\mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right).\hfill \end{array}(26)$

Third, calculating the time derivative of ${\mathcal{V}}_2(t,z)$ along the trajectories of the MRNN (Eq. 13), we have

$\begin{array}{cc}\hfill & \frac{d}{dt}{\mathcal{V}}_2\left(t,z\left(t\right)\right)\hfill \\ \hfill & =\eta \sum _{i=1}^n{e}^{\xi \left(t+{\bar{\tau }}_i\right)}{\mathcal{g}}_i^2\left(z_i\left(t\right)\right)-\eta \sum _{i=1}^n\left(1-{\dot{\tau }}_i\left(t\right)\right)e^{\xi \left(t-{\tau }_i\left(t\right)+{\bar{\tau }}_i\right)}\hfill \\ & \times {\mathcal{g}}_i^2\left(z_i\left(t-{\tau }_i\left(t\right)\right)\right)\\ \hfill & \le \eta e^{\xi \left(t+\bar{\tau }\right)}{\mathcal{g}}^T\left(z\left(t\right)\right)\mathcal{g}\left(z\left(t\right)\right)-\eta \left(1-\mu \right)e^{\xi t}{\mathcal{g}}^T\left(z\left(t-\tau \left(t\right)\right)\right)\hfill \\ & \times \mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right)\\ \hfill & \le e^{\xi t}z{\left(t\right)}^T\left[\eta e^{\xi \bar{\tau }}{L}_g{L}_g\right]z\left(t\right)-\eta \left(1-\mu \right)e^{\xi t}{\mathcal{g}}^T\left(z\left(t-\tau \left(t\right)\right)\right)\hfill \\ & \times \mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right).\end{array}(27)$

Hence, by Eqs 21, 26, 27, we have

$\begin{array}{cc}\hfill & \frac{d}{dt}\mathcal{V}\left(t,z\left(t\right)\right)\hfill \\ \hfill & =\frac{d}{dt}{\mathcal{V}}_0\left(t,z\left(t\right)\right)+\frac{d}{dt}{\mathcal{V}}_1\left(t,z\left(t\right)\right)+\frac{d}{dt}{\mathcal{V}}_2\left(t,z\left(t\right)\right)\hfill \\ \hfill & \le e^{\xi t}{z}^T\left(t\right)\left[\xi P-2PD+2P|\tilde{W}|+\left(1+\frac{1}{k}\right)I\right.\hfill \\ & \left.+2\vartheta \left(\xi L_{f}P+L_{f}P|\tilde{W}|\right)+\eta e^{\xi \bar{\tau }}{L}_g{L}_g\right]z\left(t\right)\\ \hfill & +e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)\left[-2\vartheta L_f^{-1}PD+\vartheta \left(PA+A^{T}P\right)+\vartheta I+k\Vert PA{\Vert }^{2}I\right]\hfill \\ \hfill & \times \mathcal{f}\left(z\left(t\right)\right)+e^{\xi t}{\mathcal{g}}^T\left(z\left(t-\tau \left(t\right)\right)\right)\left[\left(1+\vartheta \right)\Vert PB{\Vert }^2-\eta \left(1-\mu \right)\right]\hfill \\ \hfill & \times \mathcal{g}\left(z\left(t-\tau \left(t\right)\right)\right).\hfill \end{array}(28)$

Let $\eta =\frac{(1+\vartheta )}{1-\mu }\Vert PB{\Vert }^2$ in Eq. 28. It means that

$\begin{array}{cc}\hfill & \frac{d}{dt}\mathcal{V}\left(t,z\left(t\right)\right)\hfill \\ \hfill & \le e^{\xi t}{z}^T\left(t\right)\left[\xi P-2PD+2P|\tilde{W}|+\left(1+\frac{1}{k}\right)I\right.\hfill \\ & \left.+2\vartheta \left(\xi L_{f}P+L_{f}P|\tilde{W}|\right)+\frac{\left(1+\vartheta \right)}{1-\mu }\Vert PB{\Vert }^2{e}^{\xi \bar{\tau }}{L}_g{L}_g\right]z\left(t\right)\\ \hfill & +e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)\left[-2\vartheta L_f^{-1}PD+\vartheta \left(PA+A^{T}P\right)+\vartheta I\right.\hfill \\ \hfill & \left.+k\Vert PA{\Vert }^{2}I\right]\mathcal{f}\left(z\left(t\right)\right)\hfill \\ \hfill & =e^{\xi t}{z}^T\left(t\right)\mathrm{\Phi }z\left(t\right)+e^{\xi t}{\mathcal{f}}^T\left(z\left(t\right)\right)\mathrm{\Psi }\mathcal{f}\left(z\left(t\right)\right).\hfill \end{array}(29)$

Since Φ < 0, Ψ < 0, and by Eq. 29, we have

$\begin{array}{c}\hfill \frac{d}{dt}\mathcal{V}\left(t,z\left(t\right)\right)\le 0,\end{array}(30)$

which means that $e^{\xi t}{z}^T(t)Pz(t)={\mathcal{V}}_0(t,z(t))\le \mathcal{V}(t,z(t))\le \mathcal{V}(0,z(0))$. More precisely,

$\begin{array}{c}\hfill \Vert x\left(t\right)-x^{*}\Vert =\Vert z\left(t\right)\Vert \le M{e}^{-\frac{\xi }{2}t},\end{array}(31)$

where $M={[p\mathcal{V}(0,z(0))]}^{\frac{1}{2}}$ and p = max{p_i: i = 1, …, n}, that is, the unique equilibrium point $x^{*}$ of the MRNN (Eq. 1) is globally exponentially stable.Remark 2. Motivated by the representation of the Lyapunov functional in the work by Jianmin et al. (2019), we construct a new Lyapunov functional $\mathcal{V}(t,x(t))$, in order to overcome the difficulty brought by the nonmonotone activation functions in MRNN (Eq. 1) in the proof of Theorem 1.Now, we give an example to illustrate that the equilibrium point of the MRNN is globally exponentially stable when the conditions in Theorem 1 are satisfied.Example 1. Consider an MRNN (Eq. 1) with four state voltages, for which the parameter values of MRNN (Eq. 1) are originated from the work by Jianmin et al. (2019), especially the capacitors C₁ = 2, C₂ = 3, C₃ = 2, and C₄ = 7; the external inputs I₁ = 9, I₂ = 3, I₃ = 9.5, and I₄ = 6; the memductances $W_1^{\prime }=1$, $W_2^{\prime }=3$, $W_3^{\prime }=9.5$, and $W_4^{\prime }=1$ for x_i(t) ≤ 0; the memductances $W_1^{\prime \prime }=4$, $W_2^{\prime \prime }=1.5$, $W_3^{\prime \prime }=2$, and $W_4^{\prime \prime }=3.5$ for x_i(t) ≥ 0; and the resistors $R_f≔(R_{f_{ij}})$ and $R_g≔(R_{g_{ij}})$ are given as follows:

$R_f=\left[\begin{array}{cccc}\hfill 1\hfill & \hfill 3\hfill & \hfill 1.5\hfill & \hfill 2\hfill \\ \hfill 2\hfill & \hfill 4\hfill & \hfill 6\hfill & \hfill 12\hfill \\ \hfill 1.8\hfill & \hfill 2.6\hfill & \hfill 3.5\hfill & \hfill 4\hfill \\ \hfill 7\hfill & \hfill 3\hfill & \hfill 4\hfill & \hfill 3\hfill \end{array}\right],$

$R_g=\left[\begin{array}{cccc}\hfill 1\hfill & \hfill 3\hfill & \hfill 1.5\hfill & \hfill 2\hfill \\ \hfill 2\hfill & \hfill 2\hfill & \hfill 2\hfill & \hfill 2\hfill \\ \hfill 1.8\hfill & \hfill 2.6\hfill & \hfill 3.5\hfill & \hfill 4\hfill \\ \hfill 7\hfill & \hfill 3\hfill & \hfill 4\hfill & \hfill 3\hfill \end{array}\right].$

Next, by the aforementioned parameters and Eqs 2–6, it follows that D, $\tilde{W}$, U, A, and B. Let the activation functions

$f_i\left(x_i\left(t\right)\right)=\frac{1}{2}\left(|x_i\left(t\right)+1|-|x_i\left(t\right)-1|\right),$

$g_i\left(x_i\left(t\right)\right)=\sin \left(x_i\left(t\right)\right),$

and the time-varying delays

${\tau }_i\left(t\right)=1.8+0.5\sin t,$

for i = 1, 2, 3, 4. It is obvious that the assumptions S₁ and S₂ are satisfied. Then, by assumptions S₁ and S₂, we have $L=\mathrm{diag}\{\mathrm{1,1,1,1}\},\bar{\tau }=2.3$, and μ = 0.5.Now, by fixing the parameters k = 1000, ξ = 0.001, and ϑ = 0.001 in Theorem 1 and substituting the matrices $A,B,D,\tilde{W}$ into the linear matrix inequalities (Eq. 11), we get a positive definite diagonal matrix

$P=diag\left\{0.0499,0.0499,0.0499,0.0499\right\},$

namely, by Theorem 1, the equilibrium point of the MRNN (Eq. 1) is globally exponentially stable.The initial values of the neural network (Eq. 1) are set at (0.1,0.1,0.1,0.1)^T, (0.5,0.5,0.5,0.5)^T, and (0.9,0.9,0.9,0.9)^T. The solution trajectories of Eq. 1 are illustrated in Figure 1. From Figure 1, we see that the equilibrium point of the MRNN is globally exponentially stable, which shows the validation of the obtained result from Theorem 1.

FIGURE 1

FIGURE 1. Solution trajectories of the MRNN (1).

4 An Algorithm to Design the MRNN-Based Circuit With Stable Voltages

Note that the stable voltage is a necessary prerequisite for obtaining high-quality electric energy in power systems. In this section, the two linear inequalities in Theorem 1 are used to design the MRNN-based circuit with globally exponentially stable voltages, which make it possible to apply the MRNN to power converters. The design process is described by the following four steps:

Step 1. Fix the values of capacitor C_i, external input I_i, and the resistors $R_{f_{ij}}$ and $R_{g_{ij}}$ in Eq. 1 for i, j = 1, 2, …, n.

Step 2. For the given time-varying delay τ_i(t) and the activation functions f_i, g_i, calculate the matrices L_f, L_g in the assumption S₁ and the parameters $\bar{\tau }$ and μ in the assumption S₂.

Step 3. Determine the parameters $W_i^{\prime }$ and $W_i^{\prime \prime }$ in the memductance W_i(x_i(t)) of the i − th memristor in Eq. 1 for i = 1, 2, …, n.

• Fix a matrix P > 0 and the parameters k, ξ, and ϑ in Theorem 1.

• Substitute C_i, I_i, $R_{f_{ij}}$, and $R_{g_{ij}}$ into a_ij, b_ij, and U_i in Eq. 6 to obtain matrices A, B, D, and U.

• Substitute the matrices P, D, A, B, and U into the linear matrix inequalities (11).

• Solve Eq. 11 to obtain the matrix $\tilde{W}$.

• Calculate $W_i^{\prime }$ and $W_i^{\prime \prime }$ by the d_i and ${\tilde{W}}_i$ in Eq. 5.

Step 4. By substituting C_i, I_i, $R_{f_{ij}}$, $R_{g_{ij}}$, $W_i^{\prime }$, and $W_i^{\prime \prime }$ into Eq. 1, the MRNN-based circuit with stable voltages is obtained.

Remark 3. From Step 3, the parameters $W_i^{\prime }$ and $W_i^{\prime \prime }$ in the MRNN (Eq. 1) can be determined at the same time by the parameter d_i and ${\tilde{W}}_i$ for i = 1, 2, …, n in (6). Consequently, we can select or make the memristor guarantee the MRNN-based circuit with stable voltage when the other elements are given beforehand by means of the proposed algorithm.Next, we will design an MRNN-based circuit with four stable voltages by the proposed algorithm, where the activation functions and some of the parameters in the MRNN-based circuit in this example are the same as those in the first example.Example 2. It is declared that the activation functions f_i(x_i(t)), g_i(x_i(t)), the time-varying delay τ_i(t), and the values of parameters C_i, I_i, $R_{f_{ij}}$, and $R_{g_{ij}}$ for the MRNN-based circuit are the same as those in the first example. Next, by Step 3, we determine the values of $W_i^{\prime }$ and $W_i^{\prime \prime }$ in the memductance W_i[x_i(t)] of the i − th memristor in Eq. 1 for i, j = 1, 2, 3, 4.

• Fix the values of parameters k = 1000, ξ = 0.001, and ϑ = 0.001 in Theorem 1 and a matrix P = diag{5, 5, 5, 5} > 0 and D = diag{30, 20, 25, 10}. Substitute the matrices P, D, B, and U into the linear matrix inequalities (11). Then, solve Eq. 11 to obtain the matrix $\tilde{W}$:

$\tilde{W}=\left\{25.5148,15.9082,20.7115,6.3017\right\}.$

• Calculate $W_i^{\prime }$ and $W_i^{\prime \prime }$ by d_i and ${\tilde{W}}_i$ in Eq. 6, especially $W_1^{\prime \prime }=60.9574$, $W_2^{\prime \prime }=56.3541$, $W_3^{\prime \prime }=47.3558$, $W_4^{\prime \prime }=17.1508$; $W_1^{\prime }=35.4426$, $W_2^{\prime }=40.4459$, $W_3^{\prime }=26.6443$, and $W_4^{\prime }=10.8492$.

By Step 4, substituting C_i, I_i, $R_{f_{ij}}$, $R_{g_{ij}}$, $W_i^{\prime }$, and $W_i^{\prime \prime }$ into Eq. 1, we obtain the MRNN-based circuit with stable voltages. The initial values of Eq. 1 are given as same as those in Example 1. The solution trajectories of the designed MRNN-based circuit are depicted in Figure 2, which means that we obtain a MRNN-based circuit with stable voltages through selecting the suitable parameter values in the memductance of the memristor.

FIGURE 2

FIGURE 2. Solution trajectories of the designed MRNN-based circuit.

5 Conclusion

In this study, the global exponential stability of the equilibrium point of the MRNN is investigated for a class of general activation functions. A sufficient condition with the form of linear matrix inequalities is obtained for the global exponential stability. Furthermore, the proposed results are applied to design the MRNN-based circuits with stable voltages. From the view of the MRNN-based circuit, some elements of the MRNN-based circuit with stable voltages can be determined by the proposed algorithm. Note that the earth’s environmental pollution and the lack of energy restrict the survival and development of the human society. Wind energy, an environment-friendly renewable resource, has become one of the effective ways to solve these two difficulties. The conversion of wind energy into electric energy can rely on wind power converters. The mathematical model of the power system of new wind turbines was described by a recurrent network. Thus, further research will focus on transforming the output voltage of the wind power converter to ensure the stable amplitude of its output voltage based on MRNN with stability.

Data Availability Statement

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.

Author Contributions

ZY contributed to the globally exponential stability of MRNN by considering the proper assumptions and constructing a suitable Lyapunov functional. YL drafted the manuscript and contributed to the algorithm of design of the MRNN-based circuit with stable voltages, experiments, and conclusions. All authors agree to be accountable for the content of the work.

Funding

This work was financially supported by the China Postdoctoral Science Foundation (Grant No. 2020M670785).

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher’s Note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors, and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Anthes, G. (2011). Memristors. Commun. ACM 54, 22–24. doi:10.1145/1897852.1897859

CrossRef Full Text | Google Scholar

Chen, L., Wu, R., Cao, J., and Liu, J.-B. (2015). Stability and Synchronization of Memristor-Based Fractional-Order Delayed Neural Networks. Neural Networks 71, 37–44. doi:10.1016/j.neunet.2015.07.012

PubMed Abstract | CrossRef Full Text | Google Scholar

Chen, S., Wu, Y., Macauley, M., and Sun, X.-M. (2018). Monostability and Bistability of Boolean Networks Using Semitensor Products. IEEE Trans. Control. Netw. Syst. 6, 1379.

Google Scholar

Cheng, D., Wu, Y., Zhao, G., and Fu, S. (2021). A Comprehensive Survey on STP Approach to Finite Games. J. Syst. Sci. Complex 34, 1666–1680. doi:10.1007/s11424-021-1232-8

CrossRef Full Text | Google Scholar

Chua, L. (1971). Memristor-the Missing Circuit Element. IEEE Trans. Circuit Theor. 18, 507–519. doi:10.1109/tct.1971.1083337

CrossRef Full Text | Google Scholar

Guo, Y., Wu, Y., and Gui, W. (2021). Stability of Discrete-Time Systems under Restricted Switching via Logic Dynamical Generator and Stp-Based Mergence of Hybrid States. IEEE Trans. Automatic Control. doi:10.1109/tac.2021.3105319

CrossRef Full Text | Google Scholar

Guo, Y., Zhou, R., Wu, Y., Gui, W., and Yang, C. (2019). Stability and Set Stability in Distribution of Probabilistic Boolean Networks. IEEE Trans. Automatic Control. 64, 736–742. doi:10.1109/TAC.2018.2833170

CrossRef Full Text | Google Scholar

Hu, J., and Wang, J. (2010). “Global Uniform Asymptotic Stability of Memristor-Based Recurrent Neural Networks with Time Delays,” in IEEE Congress on Cumputational Intelligence (Spain): Barcelona), 2127–2134. doi:10.1109/ijcnn.2010.5596359

CrossRef Full Text | Google Scholar

Jianmin, W., Fengqiu, L., and Sitian, Q. (2021). Exponential Stabilization of Memristor-Based Recurrent Neural Networks with Disturbance and Mixed Time Delays via Periodically Intermittent Control. Int. J. Control Automation Syst. 19, 2284–2296. doi:10.1007/s12555-020-0083-8

CrossRef Full Text | Google Scholar

Jianmin, W., Fengqiu, L., and Sitian, Q. (2019). Global Exponential Stability of Uncertain Memristor-Based Recurrent Neural Networks with Mixed Time Delays. Int. J. Machine Learn. Cybernetics 10, 743–755. doi:10.1007/s13042-017-0759-4

CrossRef Full Text | Google Scholar

Kobravi, K., Kinsner, W., and Filizadeh, S. (2007). Analysis of Bifurcation and Stability in a Simple Power System Using Matcont. Can. Conf. Electr. Comput. Eng., 1150–1154. doi:10.1109/ccece.2007.292

CrossRef Full Text | Google Scholar

Li, H., Shao, S., Qin, S., and Yang, Y. (2021). Neural Networks with Finite-Time Convergence for Solving Time-Varying Linear Complementarity Problem. Neurocomputing 439, 146–158. doi:10.1016/j.neucom.2021.01.015

CrossRef Full Text | Google Scholar

Li, J., Hu, M., and Guo, L. (2014). Exponential Stability of Stochastic Memristor-Based Recurrent Neural Networks with Time-Varying Delays. Neurocomputing 138, 92–98. doi:10.1016/j.neucom.2014.02.042

CrossRef Full Text | Google Scholar

Ma, L., and Bian, W. (2021). A Novel Multiagent Neurodynamic Approach to Constrained Distributed Convex Optimization. IEEE Trans. Cybern. 51, 1322–1333. doi:10.1109/TCYB.2019.2895885

PubMed Abstract | CrossRef Full Text | Google Scholar

Qin, S., Wang, J., and Xue, X. (2015). Convergence and Attractivity of Memristor-Based Cellular Neural Networks with Time Delays. Neural Networks 63, 223–233. doi:10.1016/j.neunet.2014.12.002

PubMed Abstract | CrossRef Full Text | Google Scholar

Shen, X., and Raksincharoensak, P. (2021). Pedestrian-aware Statistical Risk Assessment. IEEE Trans. Intell. Transport. Syst., 1–9. doi:10.1109/TITS.2021.3074522

CrossRef Full Text | Google Scholar

Shen, X., Zhang, X., and Raksincharoensak, P. (2020). Probabilistic Bounds on Vehicle Trajectory Prediction Using Scenario Approach. IFAC-PapersOnLine 53, 2385–2390. doi:10.1016/j.ifacol.2020.12.038

CrossRef Full Text | Google Scholar

Strukov, D. B., Snider, G. S., Stewart, D. R., and Williams, R. S. (2008). The Missing Memristor Found. Nature 453, 80–83. doi:10.1038/nature06932

PubMed Abstract | CrossRef Full Text | Google Scholar

Toyoda, M., and Wu, Y. (2021). Mayer-type Optimal Control of Probabilistic Boolean Control Network with Uncertain Selection Probabilities. IEEE Trans. Cybern. 51, 3079–3092. doi:10.1109/tcyb.2019.2954849

PubMed Abstract | CrossRef Full Text | Google Scholar

Wang, Z., Lauria, S., Fang, J. a., and Liu, X. (2007). Exponential Stability of Uncertain Stochastic Neural Networks with Mixed Time-Delays. Chaos, Solitons & Fractals 32, 62–72. doi:10.1016/j.chaos.2005.10.061

CrossRef Full Text | Google Scholar

Wen, S., Bao, G., Zeng, Z., Chen, Y., and Huang, T. (2013). Global Exponential Synchronization of Memristor-Based Recurrent Neural Networks with Time-Varying Delays. Neural Networks 48, 195–203. doi:10.1016/j.neunet.2013.10.001

PubMed Abstract | CrossRef Full Text | Google Scholar

Wu, A., Zeng, Z., Zhu, X., and Zhang, J. (2011). Exponential Synchronization of Memristor-Based Recurrent Neural Networks with Time Delays. Neurocomputing 74, 3043–3050. doi:10.1016/j.neucom.2011.04.016

CrossRef Full Text | Google Scholar

Wu, Y., Cheng, D., Ghosh, B. K., and Shen, T. (2019). Recent Advances in Optimization and Game Theoretic Control for Networked Systems. Asian J. Control. 21, 2493–2512. doi:10.1002/asjc.2303

CrossRef Full Text | Google Scholar

Wu, Y., Guo, Y., and Toyoda, M. (2021). Policy Iteration Approach to the Infinite Horizon Average Optimal Control of Probabilistic Boolean Networks. IEEE Trans. Neural Netw. Learn. Syst. 32, 2910–2924. doi:10.1109/tnnls.2020.3008960

PubMed Abstract | CrossRef Full Text | Google Scholar

Yang, S., Guo, Z., and Wang, J. (2015). Robust Synchronization of Multiple Memristive Neural Networks with Uncertain Parameters via Nonlinear Coupling. IEEE Trans. Syst. Man. Cybern, Syst. 45, 1077–1086. doi:10.1109/tsmc.2014.2388199

CrossRef Full Text | Google Scholar

Zhang, G., Shen, Y., Yin, Q., and Sun, J. (2013). Global Exponential Periodicity and Stability of a Class of Memristor-Based Recurrent Neural Networks with Multiple Delays. Inf. Sci. 232, 386–396. doi:10.1016/j.ins.2012.11.023

CrossRef Full Text | Google Scholar