- Research
- Open Access
- Published:
Weighted pseudo almost periodic solutions for Clifford-valued neutral-type neural networks with leakage delays on time scales
Advances in Difference Equations volume 2020, Article number: 286 (2020)
Abstract
In this paper, we consider a class of Clifford-valued neutral-type neural networks with leakage delays on time scales. We do not decompose the networks under consideration into real-valued systems, but we directly study the Clifford-valued networks. We first establish the existence of weighted pseudo almost periodic solutions of this class of neural networks by the theory of calculus on time scales and the Banach fixed point theorem. Then, we study the global exponential stability of weighted pseudo almost periodic solutions of this class of neural networks by using inequality techniques and the proof by contradiction. Finally, we give an example to illustrate the feasibility of the obtained results.
1 Introduction
After nearly half a century of development, neural network theory has been widely and successfully applied in many research fields, such as associative memory, pattern recognition, automatic control, signal processing, auxiliary decision-making, artificial intelligence, and so on [1–6]. With the development of artificial intelligence technology, the integration of neural network and fuzzy logic, expert system, genetic algorithm, wavelet, chaos, rough set, fractal, evidence theory, grey system, and other technologies has become an important development trend of intelligent technology, which has a good development prospect. Although neural networks have been widely used in many fields, their development is not very mature, and there are still some problems to be further studied. For example, the basic theoretical framework of neural computing and the research of physiological level still need to be in-depth, the research of new models and structures, the better combination of neural network technology and other technologies, etc.
As far as the new neural network models are concerned, Clifford-valued neural networks, which are neural networks whose state variables and connection weights are all Clifford numbers, were first proposed in the reference [7]. They include real-valued neural networks, complex-valued neural networks, and quaternion-valued neural networks as their special cases, and they have more advantages than real-valued networks (see [8]). Recently, more and more attention has been paid to the study of Clifford-valued neural networks [9–13]. However, because Clifford-valued neural networks are a kind of multi-dimensional neural networks, and the multiplication of Clifford numbers does not satisfy the commutativity law, it is more difficult to study the dynamics of Clifford-valued neural networks than that of real-valued and complex-valued neural networks. Nevertheless, it is well known that the dynamics of neural networks is of great significance to the design and implementation of neural networks. Recently, in [12–14], by decomposing Clifford-valued systems into real-valued systems, the stability of the equilibrium, anti-periodicity, and \(S^{P}\)-almost periodicity of three types of Clifford-valued neural networks have been studied, respectively. At present, it is rare to study the dynamics of Clifford-valued neural networks by direct method, that is, not decomposing the Clifford-valued neural networks into real-value systems [15, 16]. In particular, until now, no paper has been published on the dynamics of Clifford-valued neural networks on time scales by direct method.
On the one hand, time delays are inevitable in real systems, and the existence of time delays may change the dynamic behavior of systems. The three types of time delays usually considered in neural network systems are transmission delay, leakage delay, and neutral-type delay. Therefore, neural network models with these types of delays are widely studied [17–21].
On the other hand, as we all know, the continuous-time systems and the discrete-time systems have the same importance in theory and practice, and the discrete-time systems are more convenient for computer processing than the continuous-time ones. So it is necessary to study the continuous-time systems and the discrete-time systems at the same time. The time scale calculus theory can unify the research of continuous analysis and discrete analysis, so the study of neural networks on time scales has important theoretical and practical significance.
In addition, in the design, implementation, and application of neural networks, the existence and stability of periodic solutions, almost periodic solutions, pseudo almost periodic solutions, and almost automorphic solutions of non-autonomous neural networks are of equal importance to the existence and stability of equilibrium solutions of autonomous neural networks. The weighted pseudo almost periodicity is the generalization of almost periodicity and pseudo almost periodicity, so it is of great theoretical significance and practical value to study the existence and stability of weighted pseudo almost periodic solutions of neural networks [22–24].
Based on the above discussion, one can see that studying Clifford-valued neural networks on time scales cannot only unify the research of real-valued neural networks, complex-valued neural networks, and quaternion-valued neural networks, but also unify the research of continuous-time neural networks and discrete-time neural networks. At the same time, the weighted pseudo almost periodic oscillation is one of the important dynamic behaviors of neural networks. Therefore, the main purpose of this paper is to study the existence and global exponential stability of weighted pseudo almost periodic solutions for a class of Clifford-valued neutral-type neural networks with leakage delays on time scales by direct method. The results of this paper are new. The method can be used to study the existence and stability of almost periodic function solutions of other types of neural networks on time scales.
This paper is organized as follows: In Sect. 2, we introduce some definitions, preliminary lemmas and give a model description. In Sect. 3, we study the existence and global exponential stability of weighted pseudo almost periodic solutions for a class of Clifford-valued neural networks on time scales. In Sect. 4, we investigate the global exponential stability of weighted pseudo almost periodic solutions of this class of neural networks. In Sect. 5, we give an example to demonstrate the feasibility of our results. This paper ends with a brief conclusion in Sect. 6.
2 Preliminaries and model description
The real Clifford algebra over \(\mathbb{R}^{m}\) is defined as
where \(\varOmega =\{\emptyset , 1,2,\ldots ,A,\ldots ,12\cdots m\}\), \(e_{A}=e_{\iota _{1}}e_{\iota _{2}}\cdots e_{\iota _{\nu }}\) with \(A=\iota _{1}\iota _{2}\cdots \iota _{\nu }, 1\leq \iota _{1}<\iota _{2}< \cdots <\iota _{\nu }\leq m\). Moreover, \(e_{\emptyset }=e_{0}=1\) and \(e_{\iota }, \iota =1,2,\ldots ,m\), are called Clifford generators and satisfy the following relations:
For \(x=\sum_{A}x^{A}\in \mathcal{A}\), we define \(\|x\|_{\mathcal{A}}=\max_{A\in \varOmega }\{|x^{A}|\}\), and for \(x=(x_{1},x_{2},\ldots ,x_{n})^{T}\in \mathcal{A}^{n}\), we define \(\|x\|_{\mathcal{A}^{n}}=\max_{1\leq p\leq n}\{\|x_{p}\|_{ \mathcal{A}}\}\), then \((\mathcal{A},\|\cdot \|_{\mathcal{A}})\) and \((\mathcal{A}^{n},\|\cdot \|_{\mathcal{A}^{n}})\) are Banach spaces. For more information about Clifford algebra, we refer the reader to [25].
A time scale \(\mathbb{T}\) is an arbitrary nonempty closed subset of the real set \(\mathbb{R}\). We use \(\rho (t)\) and \(\nu (t)\) to denote the backward jump operator and the graininess function of \(\mathbb{T}\), respectively.
Definition 2.1
([26])
Assume that \(f:\mathbb{T} \rightarrow \mathbb{R}\) is a function, and let \(t\in \mathbb{T}_{{k}}\). Then we define \(f^{\nabla }(t)\) to be the number (provided it exists) with the property that given any \(\varepsilon >0\), there is a neighborhood U of t (i.e., \(U=(t-\delta ,t+\delta )\cap \mathbb{T}\) for some \(\delta >0\)) such that
for all \(s\in U\). We call \(f^{\nabla }(t)\) the nabla derivative of f at t.
The derivative of function \(z=\sum_{A}z^{A}e_{A}:\mathbb{T}\rightarrow \mathcal{A}\) is given by \(z^{\nabla }(t)=\sum_{A}(z^{A})^{\nabla }(t)e_{A}\), where \(z^{A}:\mathbb{T}\rightarrow \mathbb{R}\).
Let \(\mathcal{R}_{\nu }=\{f: \mathbb{T}\rightarrow \mathbb{R}: 1-\nu (t)r(t) \neq 0, \forall t\in \mathbb{T}_{k}\}\) and \(\mathcal{R}_{\nu }^{+}=\{r\in \mathcal{R}_{\nu }:1-\nu (t)r(t)>0, \forall t\in \mathbb{T}\}\). Then, for \(p \in \mathcal{R}_{\nu }\), we define the nabla exponential function by
where
For \(p,q\in \mathcal{R}_{\nu }\), we define a circle plus addition by \((p \oplus _{\nu } q)(t)= p(t)+q(t)-\nu (t) p(t)q(t)\) for all \(t \in \mathbb{T}_{k}\). For \(p\in \mathcal{R}_{\nu }\), we define a circle minus p by \(\ominus _{\nu } p=-\frac{p}{1-\nu p}\). For the basic theory and notation of the time scale theory, we refer the reader to [26, 27].
Definition 2.2
([28])
A time scale \(\mathbb{T}\) is called an almost periodic time scale if
Remark 2.1
Obviously, if \(\mathbb{T}\) is called an almost periodic time scale, then \(\inf \mathbb{T}=-\infty \) and \(\sup \mathbb{T}=+\infty \).
Lemma 2.1
Let\(t\in \mathbb{T}\), where\(\mathbb{T}\)is an almost periodic time scale. Iftis left-dense, then for every\(h\in \varPi \), \(t+h\)is also left-dense. Similarly, iftis left-scattered, then for every\(h\in \varPi \), \(t+h\)is also left-scattered.
Proof
If t is left-dense, then there exists \(\{s_{n}\}\subset \mathbb{T}\) such that
Since \(\mathbb{T}\) is an almost periodic time scale, we have
for each \(h\in \varPi \). Thus, we obtain that
so \(\rho (t+h)=t+h\). Consequently, \(t+h\) is left-dense.
Now, we shall show that if t is left-scattered, then \(t+h\) is also left-dense. Otherwise, there exists \(h\in \varPi \) such that \(t+h\) is left-dense. Then there exists \(\{s_{n}+h\}\subset \mathbb{T}\) such that
Hence,
Since \(\mathbb{T}\) is an almost periodic time scale, we have \(\{s_{n}\}\subset \mathbb{T}\). That is, \(\rho (t)=t\), which contradicts the fact that t is left-scattered. Hence, \(t+h\) is also left-scattered for every \(h\in \varPi \). The proof is complete. □
Lemma 2.2
Let\(\mathbb{T}\)be an almost periodic time scale and\(h\in \varPi \), then for each\(t\in \mathbb{T}\),
Proof
If t is left-dense, then, by Lemma 2.1, it is obvious that (1) holds.
If t is left-scattered, then, by Lemma 2.1, for each \(h\in \varPi \), \(t+h\) is also left-scattered. Hence, for each \(h\in \varPi \), \(\rho (t+h)< t+h\). That is,
Since \(\mathbb{T}\) is an almost periodic time scale and \(h\in \varPi \), we have that \(\rho (t+h)-h\in \mathbb{T}\). According to (2) and the definition of the backward jump operator, for each \(h\in \varPi \), we have \(\rho (t)\geq \rho (t+h)-h\). Hence,
On the other hand, since t is left-scattered, \(\rho (t)< t\). Hence,
Since \(h\in \varPi \), \(\rho (t)+h\in \mathbb{T}\). By (4) and the definition of the backward jump operator, we obtain that
By (3) and (5), we obtain that the first equality of (1) holds. Similarly, one can prove that the second equality of (1) also holds. The proof is complete. □
In the following, we always assume that \(\mathbb{T}\) is an almost periodic time scale.
Lemma 2.3
If\(-a\in \mathcal{R}^{+}_{\nu }\)and\(t,s\in \mathbb{T}\), \(\tau \in \varPi \), then\(\hat{e}_{-a}(t+\tau ,\rho (s+\tau ))-\hat{e}_{-a}(t,\rho (s)) =\int ^{t}_{ \rho (s)}\hat{e}_{-a}(t,\rho (\theta )) (a(\theta )-a(\theta + \tau ) )\hat{e}_{-a}(\theta +\tau ,\rho (s+\tau ))\nabla \theta \).
Proof
From \((\hat{e}_{-a}(t,s))^{\nabla }=-a(t)\hat{e}_{-a}(t,s)\) it follows that
Multiplying both sides of (6) by \(\hat{e}_{-a}(\rho (s),\rho (t))\) and integrating over \([\rho (s),t]_{\mathbb{T}}\), we obtain
Noting that \([\hat{e}_{p}(c,\cdot ) ]^{\nabla }=-p [e_{p}(c,\cdot ) ]^{\rho }\) and by Lemma 2.2, \(\rho (t+\tau )=\rho (t)+\tau \), we have
Multiplying both sides of (7) by \(\hat{e}_{-a}(t,\rho (s))\), we obtain
The proof is complete. □
In the sequel, we denote by \(BC(\mathbb{T},\mathcal{A}^{n})\) the set of all bounded continuous functions from \(\mathbb{T}\) to \(\mathcal{A}^{n}\) and by \(C^{1}_{\nabla }(\mathbb{T},\mathcal{A}^{n})\) the set of all functions from \(\mathbb{T}\) to \(\mathcal{A}^{n}\) with continuous ∇-derivatives. Obviously, \(BC(\mathbb{T},\mathcal{A}^{n})\) with the norm \(\|x\|_{\infty }=\sup_{t\in \mathbb{T}}\|x(t)\|_{\mathcal{A}^{n}}\) is a Banach space.
Similar to Definition 3.9 in [28], we give the following definition.
Definition 2.3
A function \(f \in BC(\mathbb{T},\mathcal{A}^{n})\) is called almost periodic on \(\mathbb{T}\) if the ε-translation set of
is a relatively dense set in \(\mathbb{R}\) for all \(\varepsilon >0\); that is, for any given \(\varepsilon >0\), there exists a constant \(l(\varepsilon )>0\) such that each interval of length \(l(\varepsilon )\) contains at least one \(\tau (\varepsilon )\in E\{\varepsilon ,f\}\) such that
We denote by \(AP(\mathbb{T},\mathcal{A}^{n})\) the set of all almost periodic functions defined on \(\mathbb{T}\).
Similar to the proofs of the corresponding results in [28], one can easily prove the following.
Lemma 2.4
If\(f,g\in AP(\mathbb{T},\mathcal{A})\), then\(f+g,fg\in AP(\mathbb{T},\mathcal{A})\).
Lemma 2.5
If\(f\in C(\mathcal{A},\mathcal{A})\)satisfies the Lipschitz condition, \(x\in AP(\mathbb{T},\mathcal{A}) \), then\(f(x(\cdot ))\in AP(\mathbb{T},\mathcal{A})\).
Lemma 2.6
If\(x\in AP(\mathbb{T},\mathcal{A})\)and\(\tau \in AP(\mathbb{T},\mathcal{A})\), then\(x(\cdot -\tau (\cdot ))\in AP(\mathbb{T},\mathcal{A})\).
Lemma 2.7
([29])
Let\(f_{i}\in AP(\mathbb{T},\mathbb{X}_{i})\), where\(\mathbb{X}_{i}\)is a Banach space, \(i = 1,\ldots , n\). Then, for every\(\varepsilon >0\), all the functions\(f_{1}, f_{2}, \ldots , f_{n}\)have a common set ofε-almost periods.
Let \(\mathbb{U}\) be the set of all functions \(\varsigma:\mathbb{T}\rightarrow (0,+\infty )\) that are locally ∇-integrable over \(\mathbb{T}\) such that \(\varsigma >0\) almost everywhere. From now on, for \(\varsigma \in \mathbb{U}\) and \(r\in \mathbb{T}\) with \(r>0\), we denote
The space of weights \(\mathbb{U}_{\infty }\) is defined by
Noting that \(\lim_{r\rightarrow +\infty }u(r,1)=+\infty \), \(\mathbb{U}_{\infty }\neq \emptyset \).
Fix \(\varsigma \in \mathbb{U}_{\infty }\), set
Similar to Definition 3.1 in [30], we introduce the following definition.
Definition 2.4
Fix \(\varsigma \in \mathbb{U}_{\infty }\). A function \(f\in BC(\mathbb{T},\mathcal{A}^{n})\) is called weighted pseudo almost periodic if \(f=g+h\), where \(g\in AP(\mathbb{T},\mathcal{A}^{n})\) and \(h\in PAP_{0}(\mathbb{T},\mathcal{A}^{n},\varsigma )\).
We denote by \(PAP(\mathbb{T},\mathcal{A}^{n},\varsigma )\) the set of all weighted pseudo almost periodic functions from \(\mathbb{T}\) to \(\mathcal{A}^{n}\).
Denote
where \(\overline{\theta }=\sup_{t\in \mathbb{T}}\theta (t)\).
Lemma 2.8
Let\(\varsigma \in \mathbb{H}_{\infty }\). If\(x\in PAP(\mathbb{T},\mathcal{A},\varsigma )\), \(\theta \in C^{1}_{\nabla }(\mathbb{T},\varPi )\cap AP(\mathbb{T},\varPi )\), and\(\alpha:=\inf_{t\in \mathbb{T}}(1-\theta ^{\nabla }(t))>0\), then\(x(\cdot -\theta (\cdot ))\in PAP(\mathbb{T},\mathcal{A},\varsigma )\).
Proof
Since \(x\in PAP(\mathbb{T},\mathcal{A},\varsigma )\), by Definition 2.4, we have \(x=x_{1}+x_{0}\), where \(x_{1}\in AP(\mathbb{T},\mathcal{A})\) and \(x_{0}\in PAP_{0}(\mathbb{T},\mathcal{A},\varsigma )\). Then we have
By Lemma 2.6, we get \(x_{1}\in AP(\mathbb{T},\mathcal{A})\). Now, we will show that \(x_{0}\in PAP_{0}(\mathbb{T},\mathcal{A},\varsigma )\). In fact,
which implies that \(x_{0}(\cdot -\theta (\cdot ))\in PAP_{0}(\mathbb{T},\mathcal{A}, \varsigma )\). Therefore, \(x(\cdot -\theta (\cdot ))\in PAP(\mathbb{T},\mathcal{A},\varsigma )\). The proof is completed. □
Similar to the proofs of the corresponding results in [30], it is not difficult to prove the following three lemmas.
Lemma 2.9
Let\(\varsigma \in \mathbb{H}_{\infty }\). If\(\alpha \in \mathbb{R}\), \(f,g\in PAP(\mathbb{T},\mathcal{A},\varsigma )\), then\(\alpha f, f+g,fg\in PAP(\mathbb{T},\mathcal{A},\varsigma )\).
Lemma 2.10
Let\(\varsigma \in \mathbb{H}_{\infty }\). If\(f \in C(\mathcal{A},\mathcal{A})\)satisfies the Lipschitz condition and\(\varphi \in PAP(\mathbb{T},\mathcal{A},\varsigma )\), then\(f(x(\cdot ))\in PAP(\mathbb{T},\mathcal{A},\varsigma )\).
Lemma 2.11
If\(\varsigma \in \mathbb{H}_{\infty }\), then\((PAP(\mathbb{T},\mathcal{A},\varsigma ),\|\cdot \|_{\infty })\)is a Banach space.
In this paper, we consider the following Clifford-valued neutral-type neural networks with leakage delays on time scale \(\mathbb{T}\):
where \(i \in \{1,2,\ldots ,n\}=:I\); n corresponds to the number of units in the neural network; \(x_{i}(t)\in \mathcal{A}\) is the state vector of the ith unit at time t; \(a_{i}(t)>0\) represents the rate with which the ith unit will reset its potential to the resting state in isolation when disconnected from the network and external inputs; \(b_{ij}(t)\), \(c_{ij}(t)\in \mathcal{A}\) are the delay connection weight and the neutral delay connection weight from neuron j to neuron i at time t, respectively; \(\delta _{i}(t)\geq 0\) denotes the leakage delay at time t and satisfies \(t-\delta _{i}(t)\in \mathbb{T}\) for \(t\in \mathbb{T}\), \(\eta _{ij}(t),\tau _{ij}(t)\geq 0\) correspond to the transmission delays at time t and satisfy \(t-\eta _{ij}(t)\) and \(t-\tau _{ij}(t) \in \mathbb{T}\) for \(t\in \mathbb{T}\); \(u_{i}(t)\in \mathcal{A}\) is the external input at time t; \(f_{j},g_{j}:\mathcal{A}\rightarrow \mathcal{A}\) are the activation functions of signal transmission.
For convenience, we introduce the following notation:
The initial values of system (8) are as follows:
where \(\varphi _{i}:[t_{0}-\vartheta ,t_{0}]_{\mathbb{T}}\rightarrow \mathcal{A}\) is a bounded and ∇-differentiable function.
Throughout this paper, we assume that the following conditions hold:
- \((\mathrm{A}_{1})\):
For \(i,j\in I\), \(a_{i}\in AP(\mathbb{R},\mathbb{R}^{+})\) with \(-a_{i}\in \mathcal{R}_{\nu }^{+}\), \(\delta _{i},\tau _{ij}, \eta _{ij}\in C^{1}(\mathbb{R},\varPi )\cap AP( \mathbb{R},\mathbb{R}^{+})\) with \(\inf_{t\in \mathbb{R}} \{(1-\delta ^{\nabla }_{i}(t)),(1- \tau ^{\nabla }_{ij}(t)),(1-\eta ^{\nabla }_{ij}(t)) \}>0\), \(b_{ij}\), \(c_{ij}\in PAP(\mathbb{T},\mathcal{A},\varsigma )\) and \(u_{i}\in PAP(\mathbb{T},\mathcal{A},\varsigma )\).
- \((\mathrm{A}_{2})\):
There exist positive constant numbers \(L^{f}_{j},L^{g}_{j}\) such that, for any \(u,v\in \mathcal{A}\) and \(j\in I\),
$$\begin{aligned} \bigl\Vert f_{j}(u)-f_{j}(v) \bigr\Vert _{\mathcal{A}}\leq L^{f}_{j} \Vert u-v \Vert _{\mathcal{A}}, \qquad \bigl\Vert g_{j}(u)-g_{j}(v) \bigr\Vert _{\mathcal{A}}\leq L^{g}_{j} \Vert u-v \Vert _{ \mathcal{A}}. \end{aligned}$$- \((\mathrm{A}_{3})\):
There exists a positive constant ϒ such that
$$\begin{aligned} \max_{i\in I} \biggl\{ \frac{P_{i}}{a^{-}_{i}},P_{i} \biggl(1 + \frac{a^{+}_{i}}{a^{-}_{i}} \biggr) \biggr\} \leq \varUpsilon\quad \text{and}\quad \max _{i\in I} \biggl\{ \frac{Q_{i}}{a^{-}_{i}}, Q_{i} \biggl(1+ \frac{a^{+}_{i}}{a^{-}_{i}} \biggr) \biggr\} =:\kappa < 1, \end{aligned}$$where
$$\begin{aligned} &P_{i}=a^{+}_{i}\delta ^{+}_{i} \varUpsilon + \sum_{j=1}^{n}b^{+}_{ij} \bigl(L^{f}_{j}\varUpsilon + \bigl\Vert f(0) \bigr\Vert _{\mathcal{A}} \bigr) +\sum_{j=1}^{n}c^{+}_{ij} \bigl(L^{g}_{j}\varUpsilon + \bigl\Vert g(0) \bigr\Vert _{\mathcal{A}} \bigr)+u^{+}_{i}, \\ &Q_{i}=a^{+}_{i}\delta ^{+}_{i}+ \sum_{j=1}^{n}b^{+}_{ij}L^{f}_{j} +\sum_{j=1}^{n}c^{+}_{ij}L^{g}_{j},\quad i\in I. \end{aligned}$$
3 The existence of weighted pseudo almost periodic solutions
Let \(\mathbb{E}= \{\varphi \in C^{1}_{\nabla }(\mathbb{T},\mathcal{A}^{n}): \varphi ,\varphi ^{\nabla }\in PAP(\mathbb{T},\mathcal{A}^{n}, \varsigma ) \}\) with the norm
then \(\mathbb{E}\) is a Banach space.
Theorem 3.1
Let\(\varsigma \in H_{\infty }\). Assume that\((\mathrm{A}_{1})\)–\((\mathrm{A}_{3})\)hold. Then system (8) has a unique weighted pseudo almost periodic solution in\(\mathbb{E}_{0}=\{\varphi \in \mathbb{E}:\|\varphi \|_{\mathbb{E}} \leq \varUpsilon \}\), whereϒis mentioned in\((\mathrm{A}_{3})\).
Proof
Firstly, it is easy to see that if \(x=(x_{1},x_{2},\ldots ,x_{n})^{T}\in BC(\mathbb{T},\mathcal{A}^{n})\) is a solution of the following system
then x is also a solution of system (8).
Secondly, we define a mapping \(\varPhi:\mathbb{E}\rightarrow BC(\mathbb{T},\mathcal{A}^{n})\) by
where \(\varphi \in \mathbb{E}\), \(i\in I\),
We will prove that Φ maps \(\mathbb{E}\) into itself. To this end, let
By Lemmas 2.8 and 2.9, we find that \(\varGamma _{i}\in PAP(\mathbb{T},\mathcal{A},\varsigma ), i\in I\). Let \(\varGamma _{i}=\varGamma ^{1}_{i}+\varGamma ^{0}_{i}\), where \(\varGamma ^{1}_{i}\in AP(\mathbb{T},\mathcal{A})\), \(\varGamma ^{0}_{i}\in PAP_{0}(\mathbb{T},\mathcal{A},\varsigma )\), then we have
We will prove that \(F^{1}_{i}\in AP(\mathbb{T},\mathcal{A})\) and \(F^{0}_{i}\in PAP_{0}(\mathbb{T},\mathcal{A},\varsigma )\) for \(i\in I\).
In fact, since \(a_{i}\in AP(\mathbb{T},\mathbb{R}), \varGamma ^{1}_{i}\in AP(\mathbb{T}, \mathcal{A})\), by Lemma 2.7, for every \(\varepsilon >0\), there exists \(\xi \in \varPi \) such that
Hence, from the definition of \(F^{1}_{i}\) and Lemma 2.3, we have
which implies that \(F^{1}_{i}\in AP(\mathbb{T},\mathcal{A})\) for all \(i\in I\).
Since \(\varGamma ^{0}_{i}\in PAP_{0}(\mathbb{T},\mathcal{A},\varsigma ), i\in I\) and by Lebesgue’s dominated convergence theorem, we have
which implies that \(F^{0}_{i}\in PAP_{0}(\mathbb{T},\mathcal{A},\varsigma )\) for all \(i\in I\). Therefore, \(\varPhi _{i}\varphi \in PAP(\mathbb{T},\mathcal{A},\varsigma )\), \(i\in I\). Moreover, noting that for \(i\in I\),
\(a_{i}\in AP(\mathbb{T},\mathbb{R})\), and \(\varPhi _{i}\varphi , \varGamma _{i} \in PAP(\mathbb{T},\mathcal{A}, \varsigma )\), we can conclude that \(\varPhi _{i}^{\nabla }\varphi \in PAP(\mathbb{T},\mathcal{A},\varsigma )\) for \(i\in I\).
Thirdly, we will prove that Φ is a self-mapping defined on \(\mathbb{E}_{0}\). In fact, for each \(\varphi \in \mathbb{E}_{0}\), we have
and
Hence, in view of \((\mathrm{A}_{3})\), we have \(\|\varPhi \varphi \|_{\mathbb{E}}\leq \varUpsilon \), that is, \(\varPhi (\mathbb{E}_{0})\subset \mathbb{E}_{0}\).
Finally, we shall prove that Φ is a contraction mapping. In fact, for any \(\varphi =(\varphi _{1},\varphi _{2},\ldots , \varphi _{n})^{T}\), \(\psi =(\psi _{1},\psi _{2},\ldots ,\psi _{n})^{T}\in \mathbb{E}_{0}\), we have
and
By \((\mathrm{A}_{3})\), we have
Hence, Φ is a contraction mapping. Therefore, system (8) has a unique weighted pseudo almost periodic in the set \(\mathbb{E}_{0}=\{\varphi \in \mathbb{E}_{0}:\|\varphi \|_{\mathbb{E}} \leq \varUpsilon \}\). This completes the proof of Theorem 3.1. □
4 The stability of weighted pseudo almost periodic solutions
In this section, we study the global exponential stability of the unique weighted pseudo almost periodic solution of system (8) by using the proof by contradiction and the technique of inequalities.
Definition 4.1
A solution x with the initial value φ of system (8) is called globally exponentially stable if there exist positive constants λ with \(\ominus _{\nu }\lambda \in \mathcal{R}_{\nu }^{+}\) and \(\tilde{C}>1\) such that for each solution y with the initial value ψ satisfies
where
Theorem 4.1
Let conditions\((\mathrm{A}_{1})\)–\((\mathrm{A}_{3})\)hold, then system (8) has a unique weighted pseudo almost periodic solution that is globally exponentially stable.
Proof
According to Theorem 3.1, system (8) has a weighted pseudo almost periodic solution \(x=(x_{1},x_{2},\ldots ,x_{n})^{T}\). Assume that its initial value is \(\varphi =(\varphi _{1},\varphi _{2},\ldots ,\varphi _{n})^{T}\). Suppose that \(y=(y_{1},y_{2},\ldots ,y_{n})^{T}\) is an arbitrary solution of system (8) with the initial value \(\psi =(\psi _{1},\psi _{2},\ldots ,\psi _{n})^{T}\). Then, by (8), we have
Multiplying both sides of (9) by \(\hat{e}_{-a_{i}}(t_{0},\rho (t))\) and integrating over \([t_{0},t]_{\mathbb{T}}\), we obtain
Denote
and
Then, by (\(\mathrm{A}_{3}\)), for \(i\in I\), we have
and
Because \(\varTheta _{i}\) and \(\varTheta ^{\ast }_{i}\) are continuous on \([0,+\infty )\) and \(\varTheta _{i}(\omega ), \varTheta ^{\ast }_{i}(\omega ) \rightarrow - \infty \), as \(\omega \rightarrow +\infty \), so there exist \(\xi _{i}\), \(\xi ^{\ast }_{i}\) such that \(\varTheta _{i}(\xi _{i})=\varTheta ^{\ast }_{i}(\xi ^{\ast }_{i})=0\) and \(\varTheta _{i}(\omega )>0\) for \(\omega \in (0,\xi _{i})\), \(\varTheta ^{\ast }_{i}(\omega )>0\) for \(\omega \in (0,\xi ^{\ast }_{i})\), \(i\in I\). Let \(c=\min_{i\in I}\{\xi _{i},\xi ^{\ast }_{i}\}\), we have \(\varTheta _{i}(c)\geq 0\), \(\varTheta ^{\ast }_{i}(c)\geq 0\), \(i\in I\). Hence, we can take a positive constant \(0<\lambda <\min \{c,\min_{i\in I}\{a^{-}_{i}\} \}\) with \(\ominus _{\nu }\lambda \in \mathcal{R}_{\nu }^{+}\) such that
which imply that
and
Denote \(\tilde{C}=\max_{ i\in I} \{\frac{a^{-}_{i}}{Q_{i}} \}\), then by (\(\mathrm{A}_{3}\)) we have \(\tilde{C}>1\). Hence,
Since \(\hat{e}_{\ominus _{\nu }\lambda }(t,t_{0})>1\) for \(t\in [t_{0}-\vartheta ,t_{0}]_{\mathbb{T}}\), it is easy to see that
We claim that
To prove that (11) holds, we first show that, for any \({\varrho }>1\), the following inequality holds:
that is, for \(i\in I\),
In fact, if (13) is not true, then there exist some \(i\in I\) and \(t_{1}\in (t_{0},+\infty )_{\mathbb{T}}\) such that
and
Hence, one can choose a constant \(c\geq 1\) such that
and
By (10), (14), (15), and \(\tilde{C}>1\), we have
Similarly, in view of (10), we have
which contradicts (14), and so (13) holds. Letting \({\varrho }\rightarrow 1\), we obtain that (11) holds. Hence, the weighted pseudo almost periodic solution of system (8) is globally exponentially stable. The proof of Theorem 4.1 is completed. □
5 An example
In this section, we give an example to illustrate the feasibility of our results obtained in this paper.
Example 5.1
In system (8), let \(m=3\), \(n=2\), the weight \(\varrho (t)=e^{-|t|}\), and take the coefficients as follows:
If \(\mathbb{T}=\mathbb{R}\), then we take
and if \(\mathbb{T}=\mathbb{Z}\), then we take
By calculating, we have \(a^{-}_{1}=0.94\), \(a^{-}_{2}=0.87\), \(a^{+}_{1}=0.96\), \(a^{+}_{2}=0.93\), \(L^{f}_{1}=L^{f}_{2}=L^{g}_{1}=L^{g}_{2}=\frac{1}{70}\), \(\|f(\mathrm{0})\|_{\mathcal{A}}=\|g(\mathrm{0})\|_{\mathcal{A}}= \frac{3}{400}\), \(b^{+}_{11}=b^{+}_{12}=0.004\), \(b^{+}_{21}=b^{+}_{22}=0.003\), \(c^{+}_{11}=c^{+}_{22}=0.005\), \(c^{+}_{12}=c^{+}_{21}=0.004\), \(u^{+}_{1}=0.5\), \(u^{+}_{2}=0.4\).
Take \(\varUpsilon =3\), for \(i,j=1,2\), when \(\mathbb{T}=\mathbb{R}\),
and
when \(\mathbb{T}=\mathbb{Z}\),
and
Hence, whether \(\mathbb{T}=\mathbb{R}\) or \(\mathbb{T}=\mathbb{Z}\), all the conditions of Theorems 3.1 and 4.1 are satisfied. Therefore, system (8) has a weighted pseudo almost periodic solution, which is globally exponentially stable (see Figs. 1–8).
\(\mathbb{T}=\mathbb{R}\). Transient states of the solutions \(x_{i}^{0}(t)\) and \(x_{i}^{1}(t)\) of system (8) with the initial values \(x_{i}^{0}(0)=(0.1,-0.2)^{T}, (-0.4,0.3)^{T}, (0.7,-0.8)^{T}\) and \(x_{i}^{1}(0)=(-0.3,0.4)^{T}, (0.2,-0.5)^{T}, (-0.1,0.5)^{T}\), \(i=1,2\)
\(\mathbb{T}=\mathbb{R}\). Transient states of the solutions \(x_{i}^{2}(t)\) and \(x_{i}^{3}(t)\) of system (8) with the initial values \(x_{i}^{2}(0)=(-0.2,-0.3)^{T}, (0.1,0.25)^{T}, (0.2,-0.1)^{T}\) and \(x_{i}^{3}(0)=(0.1,-0.5)^{T}, (0.5,1)^{T}, (-0.7,-0.3)^{T}\), \(i=1,2\)
\(\mathbb{T}=\mathbb{R}\). Transient states of the solutions \(x_{i}^{12}(t)\) and \(x_{i}^{13}(t)\) of system (8) with the initial values \(x_{i}^{12}(0)=(0.3,-0.4)^{T}, (0.6,0.4)^{T}, (-0.1,-0.7)^{T}\) and \(x_{i}^{13}(0)=(-0.1,0.2)^{T}, (-0.5,0.4)^{T}, (-0.2,0.5)^{T}\), \(i=1,2\)
\(\mathbb{T}=\mathbb{R}\). Transient states of the solutions \(x_{i}^{23}(t)\) and \(x_{i}^{123}(t)\) of system (8) with the initial values \(x_{i}^{23}(0)=(0.3,1)^{T}, (-0.7,0.7)^{T}, (-1,-0.3)^{T}\) and \(x_{i}^{123}(0)=(0.4,-0.5)^{T}, (0.6,0.2)^{T}, (-0.1,-0.3)^{T}\), \(i=1,2\)
\(\mathbb{T}=\mathbb{Z}\). Transient states of the solutions \(x_{i}^{0}(n)\) and \(x_{i}^{1}(n)\) of system (8) with the initial values \(x_{i}^{0}(0)=(0.2,-0.1)^{T}, (-0.3,0.3)^{T}, (0.5,-0.4)^{T}\) and \(x_{i}^{1}(0)=(-0.4,0.2)^{T}, (0.5,-0.5)^{T}, (-0.6,-0.1)^{T}\), \(i=1,2\)
\(\mathbb{T}=\mathbb{Z}\). Transient states of the solutions \(x_{i}^{2}(n)\) and \(x_{i}^{3}(n)\) of system (8) with the initial values \(x_{i}^{2}(0)=(0.2,-0.1)^{T}, (-0.25,0.15)^{T}, (0.1,-0.2)^{T}\) and \(x_{i}^{3}(0)=(0.1,0.5)^{T}, (-0.5,-0.2)^{T}, (0.3,-0.4)^{T}\), \(i=1,2\)
\(\mathbb{T}=\mathbb{Z}\). Transient states of the solutions \(x_{i}^{12}(n)\) and \(x_{i}^{13}(n)\) of system (8) with the initial values \(x_{i}^{12}(0)=(0.3,-0.2)^{T}, (-0.6,0.5)^{T}, (-0.1,-0.3)^{T}\) and \(x_{i}^{13}(0)=(0.1,0.3)^{T}, (-0.2,-0.15)^{T}, (-0.5,-0.1)^{T}\), \(i=1,2\)
\(\mathbb{T}=\mathbb{Z}\). Transient states of the solutions \(x_{i}^{23}(n)\) and \(x_{i}^{123}(n)\) of system (8) with the initial values \(x_{i}^{23}(0)=(0.2,-0.3)^{T}, (-0.7,-0.5)^{T}, (0.5,0.7)^{T}\) and \(x_{i}^{123}(0)=(0.4,0.2)^{T}, (-0.05,-0.3)^{T}, (-0.4,0.25)^{T}\), \(i=1,2\)
Remark 5.1
There are no existing results that can be applied to Example 5.1.
6 Conclusion
In this paper, we have considered a class of Clifford-valued neutral-type neural networks with leakage delays on time scales. By using the Banach fixed point theorem, the theory of calculus on time scales, inequality techniques, and the proof by contradiction, we have established the existence and exponential stability of weighted pseudo almost periodic solutions for this class of neural networks. An example has been given to show the feasibility of our results. This is the first paper to study the existence and global exponential stability of weighted pseudo almost periodic solutions for Clifford-valued neutral-type neural networks with leakage delays on time scales. The study of Clifford-valued neural networks on time scales cannot only unify the research of real-valued neural networks, complex-valued neural networks, and quaternion-valued neural networks, but also unify the research of continuous-time neural networks and discrete-time neural networks. Our method of this paper can be applied to study other types of the Clifford-valued neural networks on times scales, such as BAM neural networks on time scales, shunting inhibitory cellular neural networks on time scales, high-order Hopfield neural networks, and so on.
References
Michel, A.N., Farrell, J.A.: Associative memories via artificial neural networks. IEEE Control Syst. Mag. 10(3), 6–17 (1990)
Lewis, F.L., Jagannathan, S., Yeildirek, A.: Neural Network Control of Robot Manipulators and Nonlinear Systems. Taylor & Francis, London (1999)
Cichocki, A., Amari, S.I.: Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications. Wiley, New York (2002)
Chen, M.S., Wang, H.C.: A decision-enhanced pattern classifier based on neural network approach. Pattern Recognit. Lett. 13(5), 315–323 (1992)
Ripley, B.D.: Pattern Recognition and Neural Networks. Cambridge University Press, Cambridge (2007)
Uhr, L., Honavar, V.: Artificial Intelligence and Neural Networks: Steps Toward Principled Integration. Academic Press, Boston (1994)
Pearson, J., Bisset, D.: Back propagation in a Clifford algebra. J. Artif. Neural Netw. 2, 413–416 (1992)
Buchholz, S.: A theory of neural computation with Clifford algebras. Ph.D. Thesis, University of Kiel (2005)
Li, B., Li, Y.: Existence and global exponential stability of almost automorphic solution for Clifford-valued high-order Hopfield neural networks with leakage delays. Complexity 2019, Article ID 6751806 (2019)
Hitzer, E., Nitta, T., Kuroe, Y.: Applications of Clifford’s geometric algebra. Adv. Appl. Clifford Algebras 23(2), 377–404 (2013)
Zhu, J., Sun, J.: Global exponential stability of Clifford-valued recurrent neural networks. Neurocomputing 173, 685–689 (2016)
Liu, Y., Xu, P., Lu, J.Q., Liang, J.L.: Global stability of Clifford-valued recurrent neural networks with time delays. Nonlinear Dyn. 84(2), 767–777 (2016)
Shen, S., Li, Y.: \(S^{p}\)-Almost periodic solutions of Clifford-valued fuzzy cellular neural networks with time-varying delays. Neural Process. Lett. 51(2), 1749–1769 (2020)
Li, Y., Xiang, J.: Existence and global exponential stability of anti-periodic solution for Clifford-valued inertial Cohen–Grossberg neural networks with delays. Neurocomputing 332, 259–269 (2019)
Li, B., Li, Y.: Existence and global exponential stability of pseudo almost periodic solution for Clifford-valued neutral high-order Hopfield neural networks with leakage delays. IEEE Access 7, 150213–150225 (2019)
Li, Y., Wang, Y., Li, B.: The existence and global exponential stability of μ-pseudo almost periodic solutions of Clifford-valued semi-linear delay differential equations and an application. Adv. Appl. Clifford Algebras 29(5), 105 (2019)
Shi, P., Li, F., Wu, L., Lim, C.C.: Neural network-based passive filtering for delayed neutral-type semi-Markovian jump systems. IEEE Trans. Neural Netw. Learn. Syst. 28(9), 2101–2114 (2016)
Liu, Y., Zhang, D., Lu, J.: Global exponential stability for quaternion-valued recurrent neural networks with time-varying delays. Nonlinear Dyn. 87(1), 553–565 (2017)
Arbi, A., Cao, J.: Pseudo-almost periodic solution on time-space scales for a novel class of competitive neutral-type neural networks with mixed time-varying delays and leakage delays. Neural Process. Lett. 46(2), 719–745 (2017)
Wang, F., Liu, M.: Global exponential stability of high-order bidirectional associative memory (BAM) neural networks with time delays in leakage terms. Neurocomputing 177, 515–528 (2016)
Weera, W., Niamsup, P.: Novel delay-dependent exponential stability criteria for neutral-type neural networks with non-differentiable time-varying discrete and neutral delays. Neurocomputing 173, 886–898 (2016)
Yang, G., Wan, W.: Weighted pseudo almost periodic solutions for cellular neural networks with multi-proportional delays. Neural Process. Lett. 49(3), 1125–1138 (2019)
Xu, Y.: Weighted pseudo-almost periodic delayed cellular neural networks. Neural Comput. Appl. 30(8), 2453–2458 (2018)
Xu, Y.: Exponential stability of weighted pseudo almost periodic solutions for HCNNs with mixed delays. Neural Process. Lett. 46(2), 507–519 (2017)
Brackx, F., Delanghe, R., Sommen, F.: Clifford Analysis. Pitman Advanced Publishing Program, Boston (1982)
Bohner, M., Peterson, A.: Dynamic Equations on Time Scales, an Introduction with Applications. Birkhäuser, Boston (2001)
Bohner, M., Peterson, A.C.: Advances in Dynamic Equations on Time Scales. Birkhäuser, Boston (2003)
Li, Y., Wang, C.: Uniformly almost periodic functions and almost periodic solutions to dynamic equations on time scales. Abstr. Appl. Anal. 2011, Article ID 341520 (2011)
Levitan, B.M., Zhikov, V.V.: Almost Periodic Functions and Functional Differential Equations. Cambridge University Press, Cambridge (1982)
Li, Y., Zhao, L.: Weighted pseudo almost periodic functions on time scales with applications to cellular neural networks with discrete delays. Math. Methods Appl. Sci. 40(6), 1905–1921 (2017)
Acknowledgements
Not applicable.
Availability of data and materials
Data sharing not applicable to this article as no datasets were generated or analysed during the current study.
Funding
This work is supported by the National Natural Science Foundation of China under Grant 11861072.
Author information
Authors and Affiliations
Contributions
The two authors contributed equally to the manuscript and typed, read, and approved the final manuscript.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Consent for publication
Not applicable.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Shen, S., Li, Y. Weighted pseudo almost periodic solutions for Clifford-valued neutral-type neural networks with leakage delays on time scales. Adv Differ Equ 2020, 286 (2020). https://doi.org/10.1186/s13662-020-02754-2
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13662-020-02754-2
Keywords
- Weighted pseudo almost periodic solutions
- Clifford-valued neutral-type cellular neural networks
- Global exponential stability
- Time scales