Abstract
In this article, the issue of neural adaptive decentralized finite-time prescribed performance (FTPP) control is investigated for interconnected nonlinear time-delay systems. First, to bypass the potential singularity difficulties, the hyperbolic tangent function and the radial basis function neural networks are integrated to handle the unknown nonlinear items. Then, an adaptive FTPP control strategy is developed, where an improved fractional-order filter is applied to tackle the tremendous “amount of calculation” and eliminate the filter error simultaneously. Furthermore, by considering the impact of bandwidth limitation, an adaptive self-triggered control law is designed, in which the next trigger instant is determined through the current information. Ultimately, it can be demonstrated that the proposed control scheme not only guarantees that all states of the closed-loop system are semi-globally uniformly ultimately bounded, but also that the system output is confined to a small area in finite time. Two simulation examples are carried out to verify the effectiveness and superiority of the proposed method.
Similar content being viewed by others
Explore related subjects
Discover the latest articles, news and stories from top researchers in related subjects.Avoid common mistakes on your manuscript.
1 Introduction
Given that some practical systems, such as robotic systems and industrial process systems, etc, are modeled as interconnected nonlinear systems. Therefore, the investigation of interconnected nonlinear systems has attracted widespread interest from scholars. Initially, the centralized control approach is impractical and suffers from the computational burden of redundancy due to the utilization of overall state information. Notably, the decentralized-based adaptive backstepping control schemes [1,2,3] in each subsystem have capable of independently handling the control assignment of the interconnected system via its local information. Meanwhile, the majority of existing decentralized control results require that the interconnection term of the interconnected system contains all states of the entire interconnected system, which is referred to as strong interconnections. Hence, an adaptive decentralized control method was investigated for interconnected nonlinear systems with strong interconnections in [4]. Due to the existence of unknown nonlinearities in interconnected nonlinear systems, with the aid of radial basis function neural networks (RBFNNs) and fuzzy logic systems, the adaptive decentralized control schemes were presented in [2, 3]. Generally, the backstepping technique entails the issue of complexity in analytical computation stemming from the iterative calculation of the virtual control function. To this end, the dynamic surface control (DSC) schemes [5,6,7] were proposed to address the explosive calculation by introducing a first-order filter. Although the above DSC methods solve the above issues, filter errors between the filter output and the virtual control function were not eliminated. Thus, the command filter backstepping control (CFBC) methods [8, 9] were constructed by means of an error compensation signal. Meanwhile, Song et al. [10] developed an adaptive fuzzy secure control solution for nonlinear systems, where a fractional-order filter (FOF) was introduced to eradicate the effect of filter errors. Noteworthy, the above CFBC-based control results neglect the impact of the communication burden.
Until today, event-triggered control (ETC) plays an important role in the realm of transmission resource constraints, where the measurement error satisfies the trigger criterion, the control signal will be updated and applied to the system. Targeting the aircraft wing rock motion [11], an adaptive event-triggered control was presented to economize communication resources. [12] discussed the prescribed-time synchronization issue for nonlinear systems via ETC strategy. Unfortunately, the event-triggered mechanism is imposed to continuously monitor the trigger criteria, which requires relatively unavailable hardware situations for industrial applications. With this in mind, an adaptive self-triggered control solution was investigated in [13], which can calculate the next trigger instant following the current information. Meanwhile, [14] reported a neural adaptive self-triggered control method for nonlinear systems with unmeasurable states.
To further handle the limitation of bandwidth in networked control, quantized control was used to degrade the communication rate to satisfy that the system can operate normally within the specified bandwidth. Initially, a logarithmic quantizer was investigated in [15], where the bandwidth of the constrained transmission was alleviated. However, the chattering phenomenon will inevitably occur in the quantization control of continuous-time systems. Soon afterward, the hysteresis quantizer was developed in [16], which can decrease the risk of chattering. In [17], an adaptive fuzzy quantized control was proposed for nonlinear systems. In addition, networked control may result in discontinuous control signals at two trigger instants, which will drastically affect the control system’s performance. Hence, it is imperative to concentrate on the trade-off between communication cost and tracking precision.
It is widely acknowledged that prescribed performance control (PPC) enables tracking errors to converge to a predefined range. Particularly, Sun et al. [18] and Song et al. [19] investigated the adaptive CFBC methods for the different nonlinear systems, where transient and steady performances were maintained. In [20], an adaptive PPC scheme was analyzed for nonlinear systems. Soon afterward, an improved performance function in comparison to PPC is proposed, called the finite-time prescribed performance (FTPP) function, which guarantees that the tracking error is confined to a small origin in finite time. In particular, [21] developed a fuzzy adaptive FTPP control strategy for nonlinear systems with dynamic uncertainty. Nevertheless, the above results do not both consider the time delay and unmodeled dynamics that are ubiquitous in modern industrial applications.
Naturally speaking, if this obstacle is not overcome, the system performance will deteriorate and even resulting in the instability of the closed-loop system. Thus, many remarkable results have been reported to ensure the stability of the systems (see [22,23,24,25,26] and reference therein). Among them, by introducing a dynamic signal, an adaptive decentralized tracking control strategy was addressed for nonlinear systems with dynamical uncertainties in [23]. In [24], with the help of Lyapunov–Krasovskii functional, an adaptive neural control scheme was discussed for interconnected nonlinear systems with time delay. Meanwhile, Li et al. [26] developed an adaptive CFBC method for nonlinear time-delay systems.
Guided by the foregoing analysis, this paper developed the neural adaptive FTPP quantized control strategy by utilizing the FOF for interconnected nonlinear time-delay systems with unmodeled dynamics and self-triggered input. The highlights of this article are enumerated below:
-
1.
Different from the event-triggered mechanism [10, 11, 26], the investigated self-triggered control solution was developed, where the next trigger instant was determined by the current information. In addition, the effect of bandwidth limitation was synthesized in the interconnected nonlinear time-delay system as a challenging issue in comparison to [7, 16, 17]. However, the networked control mentioned above may affect the performance of the system. Tactfully, the FTPP function was considered to optimize the transient and steady-state performance of the interconnected nonlinear system.
-
2.
In contrast to the integer-order filter in [9, 18, 26, 27], an improved FOF was introduced such that the tremendous “ amount of calculation ” was avoided and the filter performance was skillfully enhanced. Unlike [7, 21, 28], which only considered the PPC issue or error compensation signal, the proposed self-triggered quantized control scheme in this paper takes FTPP and error compensation signal into account to eliminate the filter errors simultaneously.
-
3.
To address the potential singularity problem that may exist in interconnected nonlinear systems, hyperbolic tangent functions, and RBFNNs were introduced in [16, 26]. Furthermore, this article extends both time delay and unmodeled dynamics to interconnected nonlinear systems, which makes it applicable to more general situations [3, 7, 23].
2 System formulation
Consider the following interconnected nonlinear time-delay plant
where \(\bar{x}_{m,i} = [x_{m,1}, \ldots , x_{m,i}]^ T \in R^i\) and \(y_m \in R\) are the state vector and output vector of the mth subsystem, respectively. \(\check{x}_m=[x_{m,1}, \ldots , x_{m,n_m}]^ T \in R^{n_m}\), \(\bar{y}=[y_1, \ldots , y_M]^ T \in R^M\). \(f_{m,i}(\cdot )\) and \(h_{m,i}(\cdot )\) with \(1\le m \le M\) and \(1 \le i \le n_m\) are the unknown smooth nonlinear functions, \(g_{m,i}(\bar{y})\) denotes the unknown smooth interconnected term between the mth subsystem and other subsystems. \(\tau _{m,i}(t)\) represents the time-varying delay satisfying \(\mid \tau _{m,i}(t)\mid \le \bar{\tau }_{m,i}<\infty\) and \(\mid \dot{\tau }_{m,i}(t)\mid \le \check{\tau }_{m,i}<1\), where \(\bar{\tau }_{m,i}>0\) and \(\check{\tau }_{m,i}>0\) are constants. The \(\psi _m\)-dynamics and \(\Delta _{m,i}(\cdot )\) denote the unmodeled dynamics and dynamic disturbances, respectively. \(q_m\) and \(\Delta _{m,i}(\cdot )\) indicate the Lipschitz continuous functions. \(q(u_m)\) and \(u_m\) indicate the quantized control signal and quantized input signal. Meanwhile, the hysteresis quantizer is described as follow:
where \({\vartheta }_m=(1+\hslash _{m}){\text{sgn}}(u_m)\), \(u_{m}^w=\delta _{m}^{1-w}u_{m}^{\rm{min}}\ (w=1,2,\ldots )\) with \(0<\delta _{m}<1\) stands for a measure of quantization density and \(\hslash _{m}=\frac{1-\delta _{m}}{1+\delta _{m}}\), \(u_{m}^{\rm{min}}\) indicates the scope of the dead-zone for \(q(u_m)\) and \(q(u_m)\) is in the set \(\bigcup _m=(0,\pm u_{m}^w,\pm u_{m}^w(1+\hslash _{m}))\).
To support the controller design, we need the following assumptions, lemmas, and definitions.
Definition 1
[29] Assume \(F(t):[t_0,+\infty )\rightarrow R\) is a continuous function together with its fractional derivative of order \(P_m\) under Caputo’s definition is expressed as:
where \(n_m\) denotes an integer such that \({n_m-1}\le {P_m}\le n_m\). \(\Gamma (x)=\int _{0}^{+\infty }\tau ^{x-1}e^{-\tau }\text {d}\tau (x>0)\) denotes Euler’s Gamma function with \(\Gamma (1)=1\).
Definition 2
[28] A smooth finite-time performance function (FTPF) \(\ell _m(t)\) satisfies the following characteristics: \((1)\ \ell _m(t)>0\), \((2)\ \dot{\ell }_m(t)\le 0,\) \((3)\ \lim \nolimits _{t \rightarrow T_f} \ell _m(t)=\ell _{mT_f}>0\) and \(\ell _m(t)=\ell _{mT_f}>0\) for any \(t>T_f\) with \(T_f\) and \(\ell _{mT_f}\) are the settling time and the arbitrarily small constant, respectively.
Lemma 1
[30] To further analyze the quantization impact, the hysteretic quantizer is reconstructed as \(q(u_m(t))=(1-\bar{\kappa }_{m})u_m(t)+\bar{\kappa }_{m} \bar{\varsigma }_m(t)\) such that the nonlinearity function \(\bar{\varsigma }_m(t)\) satisfies the following inequalities:
where \(0<\bar{\kappa }_{m}<1\) is an adjustable constant to be designed.
Assumption 1
[31] There exist unknown non-negative smooth functions \(\varphi _{m,i1}(\Vert \bar{x}_{m,i}\Vert )\) and \(\varphi _{m,i2}(\Vert \psi _i\Vert )\), the dynamic disturbance \(\Delta _{m,i}(\psi _i,\check{x}_m)\) satisfying
Assumption 2
[32] Let us consider the unmodeled dynamics \(\dot{\psi }_m=q_m(\psi _m,\check{x}_m)\), which is exponentially input-to-state practically stable (exp-ISpS). Meanwhile, \(V_m(\psi _m)\) is an exp-ISpS Lyapunov function (LF) satisfying
where \(\Upsilon _{m,1}, \Upsilon _{m,2}\), and \(\pi _{m0}\) are class \(K_\infty\)-functions and \(r_{m,1}>0\) and \(r_{m,2}>0\) denote known constants.
Assumption 3
[4] The interconnected term \(g_{m,i}(\bar{y})\) has the form of \(\mid g_{m,i}(\bar{y})\mid \le \sum \nolimits _{l= 1}^{M} j_{m,l}\aleph _i(\bar{x}_{l,i})\), with \(\aleph _i(\bar{x}_{l,i})\ (i=1,\ldots ,n_m)\) and \(j_{m,l}\) denote the unknown continuous function and unknown constant, respectively.
Lemma 2
[31] Assume exp-ISpS LF \(V_m(\psi _m)\) satisfies the equation condition (2), then, for \(\forall \bar{r}_{m,1} \in (0,r_{m,1})\), the initial condition \(\psi _{m0}=\psi _m(t_0)\), \(c_0>0\), and the function \(\bar{\pi }_{m0}(\Vert y_m\Vert )\ge \pi _{m0}(\Vert y_m\Vert )\), there exist a finite time \(T_{m0}=T_{m0}(\bar{r}_{m,1},c_0,\psi _{m0})\), a non-negative function \(Q(t_0,t)(t\ge t_0)\), and a dynamic signal expressed as:
satisfying \(Q(t_0,t)=0\) for \(\forall t\ge t_0+T_{m0}\)
where \(t_0\) denotes the initial time and assume that \(\bar{\pi }_{m0}(\Vert x_{m,1}(t)\Vert )=\pi _{m0}(\Vert x_{m,1}\Vert )\).
Lemma 3
[33] For variable \(\varrho \in R\) and \(\forall \jmath >0\), one has
Lemma 4
[34] The compact set \(\Omega _{m_k}\) is defined in \(\Omega _{m_k}=\left\{ s_{m,k}\mid \mid s_{m,k}\mid <0.2554m_k\right\}\). Next, for the case of \(\forall s_{m,k} \notin \Omega _{{m,k}}\), the term \(1-16{\text{tanh}}^2(s_{m,k}/{m_k})\le 0\) holds.
Lemma 5
[35] Consider the unknown smooth nonlinearities \(f_{m,i}(\bar{z}_{m,i})\) on the compact set \(\Omega _X\), which can be approximated by RBFNNs
where \(W_{m,i}\) and \(S_{m,i}(\bar{z}_{m,i})=[\psi _{m,i1}(\bar{z}_{m,i}),\ldots ,\psi _{m,iK}(\bar{z}_{m,i})]^{\rm{T}}\) are, respectively, the weight vector and basic function vector, with \(\bar{z}_{m,i}\) is the input of the RBFNNs, \(K\ge 1\) is the number of neuron, \(\varepsilon _{m,i}\) is the approximation error. There exists a positive constant \(\varepsilon _{m,i}^*>0\) such that \(\Vert \varepsilon _{m,i}\Vert \le \varepsilon _{m,i}^*\). Meanwhile, the Gaussian function is expressed as:
where \(h_{m,i}=\left[ h_{m,i1},\ldots ,h_{m,iM}\right] ^{\rm{T}}\) and \(b_{m,i}\) denote the center and width of the basis function, respectively.
This paper aims at synthesizing a decentralized-based adaptive neural quantized control algorithm for the interconnected nonlinear time-delay system in (1) such that all signals of the resulting closed-loop system (CLS) are semi-globally uniformly ultimately bounded (SGUUB), the system output is confined to a small adjustable region in a finite time interval, and as well as the Zeno phenomenon is ruled out.
3 Main results
In this section, a decentralized-based adaptive self-triggered neural control scheme will be put forward for interconnected nonlinear time-delay systems.
3.1 State transformation
To confine the system output \(x_{m,1}\) to the range \((-\ell _m,\ell _m)\), the FTPF can be chosen from Definition 2.
where \(\ell _{m0}>0\) and \(\ell _{mT_f}>0\) are design parameters.
Furthermore, it can be obtained that \(\ell _m(0)=\ell _{m0}+\ell _{mT_f}\) from Definition 2 and (3). The error transformation function with the transformed error \(\partial _m\) is selected as:
The transformation is defined as:
and
where \(p_{m,1}=\frac{1}{2\ell _m}\left( \frac{1}{x_{m,1}/\ell _m+1}-\frac{1}{x_{m,1}/\ell _m-1}\right)\).
3.2 Controller design
In contrast to the advancement of integer-order calculus [9, 18, 26, 27], fractional-order calculus [36,37,38,39] possesses the property of the favorable filter and enhances the freedom of control design attributed to its distinctive historical memory characteristics. In a nutshell, we propose a FOF-based adaptive self-triggered control algorithm for interconnected nonlinear time-delay systems in (1), which can not only overcome the complicated “ amount of calculation ” but also effectively upgrade the filter performance of the existing results in [9, 18, 26, 27].
Now, the fractional-order filter (FOF) are constructed:
where \(D^{P_m}\) indicates the fractional operator with \(0<P_m<1\) and \(k_{m,li}\ (i=1,2,3)\) is the design parameter. The virtual control function \(\alpha _{m,{i - 1}}\) indicates the filter input, \(\zeta _{m,i}=\beta _{m,1}\) and \(D^{P_m}\zeta _{m,i}=I_{m,1}\) are the filter output.
Define the change of coordinate as follows:
where \(z_{m,i}\) denotes the error surface and \(\xi _{m,i}\) denotes the filter error. Additionally, the compensated error signal \(s _{m,i}\) with \(i=1,\ldots ,n_m\) is designed as follows:
where \(\nu _{m,i}\) is the error compensation signal, which can be specifically designed as:
where \(\nu _{m,i}(0)=0,\ p_{m,i}=1\) and \(c_{m,i}>0 \ (i=2,\ldots ,n_m)\) denotes the positive design parameters.
The virtual control functions and the adaptive laws are designed for each subsystem \((i=2,\ldots ,{n_m-1})\) as follows:
where \(c_{m,i}\) and \(\rho _{m,i}\) are positive design parameters. Let us define that \(\hat{\Theta }_{m,i}\) denotes the estimates of \(\Theta _{m,i}\). \(\tilde{\Theta }_{m,i}= \Theta _{m,i}-\hat{\Theta }_{m,i}\) being the parameter estimation error with \(\Theta _{m,i} = \max \nolimits _{ 1 \le i \le n_m} \left\{ {\left\| W_{m,i}^*\right\| }^2 \right\}\). Analogous to (16), \(\dot{\hat{\Theta }}_{m,n_m}\) can be expressed by replacing i with \(n_m\).
For simplicity, the corresponding abbreviations are considered: \(f_{m,i}\) indicates \(f_{m,i}(\bar{x}_{m,i})\), \(\Delta _{m,i}\) indicates \(\Delta _{m,i}(\check{x}_m,\psi _i)\), \(g_{m,i}\) indicates \(g_{m,i}(\bar{y})\), \(h_{m,i}(x_{m,i\tau })\) indicates \(h_{m,i}(\bar{x}_{m,i}(t-\tau _{m,i}(t)))\), and \(S_{m,i}^{\rm{T}} (\cdot )S_{m,i}(\cdot )\) indicates \(S_{m,i}^{\rm{T}}(\bar{z}_{m,i})S_{m,i}(\bar{z}_{m,i})\).
Proof
The specific control design procedures of this paper are expressed in detail as follows.
Step m, 1: From (7) and (9), the time derivative of \(z_{m,1}\) yields
It follows from (10)–(11) that
Now, we consider the following Lyapunov function:
where \(\mu _{m,1}\) and \(\varrho _{m0}\) are positive constants. The Lyapunov–Krasovskii functional is formulated as \(\bar{V}_{m,1}=[e^{\kappa _{m,1}\bar{\tau }_{m,1}}/2(1-\check{\tau }_{m,1})]\int _{t - \tau _{m,1}(t)}^t e^{-\kappa _{m,1}(t-s)}h_{m,1}^2(x_{m,1}(s))ds\) to tackle the time delay challenge, where \(\kappa _{m,1}>0\) denotes a constant.
Through calculation, one has
where \(\Psi _{m,1}=[(e^{\kappa _{m,1}\bar{\tau }_{m,1}})/2(1-\check{\tau }_{m,1})]h_{m,1}^2(x_{m,1}).\)
By utilizing Assumption 3 and Young’s inequality yields
where \(\bar{\varpi }_{m,l}=(1/2)Mj_{m,l}^2\).
In light of Assumption 1 and Lemma 3 holds
where \(\jmath _{m,11}>0\) denotes a constant and \(\bar{\varphi }_{m,11}(\Vert x_{m,1}\Vert )=\varphi _{m,11}(\Vert x_{m,1}\Vert )\tanh \left( (p_{m,1}\mid s_{m,1}\mid \varphi _{m,11}(x_{m,1}))/\jmath _{m,11}\right)\).
In view of Assumption 2 and Lemma 3, one obtains
where \(\jmath _{m,12}>0\) denotes a constant and \(\bar{\varphi }_{m,12}(x_{m,1},\lambda _m)=\varphi _{m,12}(\vartheta _{m,1}^{-1}(2\lambda _m))\tanh ((p_{m,1}s_{m,1}\varphi _{m,12}\left( \vartheta _{m,1}^{-1}(2\lambda _m)))/\jmath _{m,12}\right)\).
Calculating \(\dot{V}_{1}\), one obtains
where \(\varpi _{m,1}=\bar{\varphi }_{m,11}(\Vert x_{m,1}\Vert )+\bar{\varphi }_{m,12}(x_{m,1},\lambda _m)\) and \(\chi _{m,1}=0.2785\omega _{m,11}+0.2785\omega _{m,12}+r_{m,2}/\varrho _{m0}+\frac{1}{2}\varphi _{m,12}^2(\vartheta _{m,1}^{-1}(2\bar{D}_m(t_0,t)))\).
The hyperbolic tangent function \(16\tanh ^2(s_{m,1}/\sigma _{m,1})\) is utilized to solve the \(\Psi _{m,1},\ \sum \nolimits _{l= 1}^M \bar{\varpi }_{m,l}\aleph _i^2(\bar{x}_{l,1})\), and \((\bar{\pi }_{m0}(\Vert x_{m,1}(t)\Vert )/\varrho _{m0})\).
where \(G_{m,1}=\Psi _{m,1}+(\bar{\pi }_{m0}(\Vert x_{m,1}(t)\Vert ))/\varrho _{m0}+\sum \nolimits _{l= 1}^M \bar{\varpi }_{m,l}\aleph _i^2(\bar{x}_{m,1})\) and \(\sigma _{m,1}>0\) denotes a constant. \(\square\)
Remark 1
Since \(\lim \nolimits _{s_{m,1} \rightarrow 0} [G_{m,1}/(p_{m,1}s_{m,1})]=\infty\), the lumped terms \(G_{m,1}/(p_{m,1}s_{m,1})\) cannot be directly approximated by the RBFNNs. To tackle this obstacle, a hyperbolic tangent item \((16/p_ms_{m,1})\tanh ^2(s_{m,1}/\sigma _{m,1}))G_{m,1}\) is used to prevent the singularity of the potential challenge. Meanwhile, two different cases will be further elaborated to handle the term \((1-16\tanh ^2(s_{m,1}/\sigma _{m,1}))G_{m,1}\) in Sect. 3.3.
To further simplify the \(\dot{V}_1\) in (24), the RBFNNs is employed to approximate the unknown item \(F_{m,1}(\bar{z}_{m,1})=f_{m,1}+(16/p_ms_{m,1})\tanh ^2(s_{m,1}/\sigma _{m,1})G_{m,1}+\varpi _{m,1}+2p_{m,1}s_{m,1}\), where the RBFNNs input vector \(\bar{z}_{m,1}=[x_{m,1},s_{m,1},p_{m,1},\lambda _1(t)]^{\rm{T}}\).
Thus, one has
where \(\varepsilon _{m,1}\) is the approximation error with \(\mid \varepsilon _{m,1}\mid \le \varepsilon _{m,1}^*\).
Substituting (11), (12), (15), and (25) into (24) becomes
Step m, i. From (10) and (11), it deduces that
By utilizing Assumption 3 and Young’s inequality holds
Select the following Lyapunov function candidate:
where \(\kappa _{m,i}>0\) denotes a constant and \(\bar{V}_{m,i}=(e^{\kappa _{m,i}\bar{\tau }_{m,i}}/2(1-\check{\tau }_{m,i}))\int _{t - \tau _{m,i}(t)}^t e^{-\kappa _{m,i}(t-s)}h_{m,i}^2(x_{m,i}(s))ds.\)
Through calculation, one has
where \(\Psi _{m,i}=(e^{\kappa _{m,i}\bar{\tau }_{m,i}}/2(1-\check{\tau }_{m,i}))h_{m,i}^2(x_{m,i})\).
Similar to (21) and (22) yields
where \(\jmath _{m,i1}>0\) denotes a constant.
where \(\jmath _{m,i2}>0\) denotes a constant and \(\bar{\varphi }_{m,i2}(\bar{x}_{m,i},\lambda _m)=\varphi _{m,i2}(\vartheta _{m,i}^{-1}(2\lambda _m))\tanh ((s_{m,i}\varphi _{m,i2}(\vartheta _{m,i}^{-1}(2\lambda _m)))/\jmath _{m,i2})\).
The time derivative of \(V_i\) can be obtained
where \(\varpi _{m,i}=\varphi _{m,i1}(\Vert \bar{x}_{m,i}\Vert )\tanh \left( (\mid s_{m,i}\mid \varphi _{m,i1}(\bar{x}_{m,i}))/\jmath _{m,i1}\right) +\bar{\varphi }_{m,i2}(\bar{x}_{m,i},\lambda _m)\) and \(\chi _{m,i}=\frac{1}{2}\varphi _{m,i2}^2(\vartheta _{m,i}^{-1}(2\bar{D}_m(t_0,t)))+0.2785\jmath _{m,i1}+0.2785\jmath _{m,i2}.\)
By applying the hyperbolic tangent function, \(\dot{V}_i\) becomes
where \(G_{m,i}=\Psi _{m,i}+\sum \nolimits _{l= 1}^M \bar{\varpi }_{m,l}\aleph _i^2(\bar{x}_{m,i}).\)
The RBFNNs is utilized to approximate the unknown item \(F_{m,i}(\bar{z}_{m,i})=f_{m,i}+(16/s_{m,i})\tanh ^2(s_{m,i}/\sigma _{m,i})G_{m,i}+\varpi _{m,i}- \dot{\zeta } _{m,i}+2s_{m,i}\) in (33), where the RBFNNs input vector \(\bar{z}_{m,i}=[x_{m,i},s_{m,i},\lambda _i(t),\dot{\zeta } _{m,i}]^{\rm{T}}\). Therefore, one obtains
where \(\varepsilon _{m,i}\) is the approximation error with \(\mid \varepsilon _{m,i}\mid \le \varepsilon _{m,i}^*\).
Substituting (11), (13), (16), and (34) into (33) yields
Step \(m,{n_m}\). In industrial process, the existence of actuator saturation nonlinearity will affect the system stability. Therefore, the following saturation function is considered:
where \(u_m^{\rm{Max}}\) is the bound of \(u_m\) and \(\bar{u}_m\) is the input of the following saturation nonlinearity:
Meanwhile the quantized parameters satisfy \((1-\bar{\kappa }_{m})u_{m}^{\rm{min}}/(\bar{\kappa }_{m}+\hslash _{m})\ge u_m^{\rm{Max}}\). Referring to [40], \({\text{sat}}(\bar{u}_m)\) becomes
where \(\mid \bar{\varrho }(\bar{u}_m)\mid =\mid {\text{sat}}(\bar{u}_m))-\bar{b}_m(\bar{u}_m)\mid \le u_m^{\rm{Max}}(1-\tanh (1))=\bar{\underline{\varrho }}_{m}.\)
According to the mean-value theorem, \(\bar{b}_m(\bar{u}_m)\) can be expressed as:
where \(\bar{u}_{m}^{c_m}=c_m\bar{u}_m+(1-c_m)\bar{u}_{m0}\) and \(0<c_m<1\).
Let \(\bar{u}_{m0}=0\), (37) becomes
For the positive constants \(\underline{\zeta }_{m0}\) and \(\bar{\zeta }_{m0}\), the expression \(\underline{\zeta }_{m0}<\zeta _{m0}(\bar{u}_{m}^{c_m})<\bar{\zeta }_{m0}\) holds. Thus, (36) becomes
According to (9)–(11) and (39), \(\dot{s}_{m,n_m}\) can be expressed as:
Let us construct the Lyapunov function candidate as \(V_{n_m} = V_{n_m-1}+ \sum \limits _{m = 1}^M \left[ \frac{1}{2}s_{m,n_m}^2 + \frac{1}{2\mu _{m,n_m}}\tilde{\Theta }_{m,n_m}^2+\bar{V}_{m,n_m}\right].\)
By utilizing Young’s inequality, one has
Then, it can be deduced that
where
Define \(G_{m,n_m}=\Psi _{m,n_m}+\sum \nolimits _{l= 1}^M \bar{\varpi }_{m,l}\aleph _{n_m}^2(\bar{x}_{m,n_m})\) yields
The RBFNNs is employed to approximate the unknown function \(F_{m,n_m}(\bar{z}_{m,n_m})=f_{m,n_m}+(16/s_{m,n_m})\tanh ^2(s_{m,n_m}/\sigma _{m,n_m})G_{m,n_m}+\varpi _{m,n_m}- \dot{\zeta } _{m,n_m}+2.5s_{m,n_m}\) in (43), where the RBFNNs input vector \(\bar{z}_{m,n_m}=[x_{m,n_m},s_{m,n_m},\lambda _{n_m}(t),\dot{\zeta } _{m,n_m}]^{\rm{T}}\). Therefore, one obtains
where \(\varepsilon _{m,n_m}\) is the approximation error with \(\mid \varepsilon _{m,n_m}\mid \le \varepsilon _{m,n_m}^*\).
Meanwhile, the self-triggered scheme is developed as follows:
where \(t_k,t_{k+1}\in Z^+; 0<\Lambda _m<1\), \(E_m\) and \(L_m\) are positive constant. \(\Lambda _m\mid \bar{u}_m(t)\mid +E_m\) denotes control signal interval between two successive triggered instants; \(L_m\) and \(\mid \dot{v}_m(t)\mid\) are change rates of control signal interval.
When the conditions of (45) are satisfied, \(\bar{u}_m(t)=v_m(t_k)\) will be used to the control plant in (1). The next trigger point \(t_{k+1}\) will be obtained, and control signal \(\bar{u}_m(t)\) holds as a constant \(v_m(t_k )\) in the period interval \([t_k,t_{k+1})\).
\(v_m(t_k )\) is constructed as follows:
where \(\jmath _m> 0\) and \(\bar{E}_m>E_m/(1-\Lambda _m)\).
From (45), it has \(\mid \bar{u}_m(t_{k+1})-u_m(t)\mid \le \Lambda _m\mid \bar{u}_m(t)\mid +E_m\). There exist the continuous time-varying parameters \(\theta _{m,1}, \theta _{m,2}\) satisfying \(\theta _{m,1}(t_k)=\theta _{m,2}(t_k)=0, \theta _{m,1}(t_{k+1})=\theta _{m,2}(t_{k+1})=\pm 1\) and \(\mid \theta _{m,1}\mid \le 1\), \(\mid \theta _{m,2}\mid \le 1, \forall t\in [t_k,t_{k+1})\),
Therefore, \(\dot{V}_{n_m}\) can be expressed as:
On the basis of (46) and Lemma 3, it has \(s_{m,n_m}v_m<0\). In light of \(0< [(s_{m,n_m}v_m)/(1 + \theta _{m,1}\Lambda _m)]< [(s_{m,n_m}v_m)/(1 +\Lambda _m)]\) and \(- [(s_{m,n_m}\theta _{m,2}E_m)/(1 + \theta _{m,1}\Lambda _m)] \le \mid s_{m,n_m}E_m/1 -\Lambda _m\mid \le \mid {s_{m,n_m}}{\bar{E}_m} \mid,\) the following inequality holds
Since \(\tilde{\Theta }_{m,i}\hat{\Theta }_{m,i} \le - (\tilde{\Theta } _{m,i}^2/2) + (\Theta _{m,i}^2/2)\) and invoking (11), (14), (48) and (49) results in
where \(\bar{\chi }_m=\sum \nolimits _{k = 1}^{n_m}\chi _{m,k}+\sum \nolimits _{k = 1}^{n_m}(\rho _{m,k}/2)\Theta _{m,k}^2 +\sum \nolimits _{k = 1}^{n_m}(a_{m,k}^2/2)+\sum \nolimits _{k = 1}^{n_m}(\varepsilon _{m,k}^{*2}/2)+0.557\jmath _m(1-\bar{\kappa }_{m})\zeta _{m0}\).
From the analysis in [16, 41], the Lyapunov function is designed as \(V=\bar{\delta }_mV_{n_m}\), where \(\bar{\delta }_m>0\) refers to the cofactor of the mth diagonal element by instead of \(r_{m,l}\) with \(\bar{\varpi }_{m,l}\). Thus, (50) becomes
In light of [4], \(\sum \nolimits _{m = 1}^M\sum \nolimits _{k = 1}^{n_m}\sum \nolimits _{l= 1}^M\bar{\delta }_m\bar{\varpi }_{m,l}(\aleph _{k}^2(\bar{x}_{l,k})-\aleph _{k}^2(\bar{x}_{m,k}))=0\) holds. Therefore, (51) arrives at
To this end, the adaptive self-triggered neural control scheme was preliminary accomplished. The results are summarized in the following Theorem 1.
3.3 Stability analysis
Theorem 1
Let us consider the interconnected nonlinear time-delay system in (1), under Assumptions 1–3, the error compensation signals in (11), the virtual control functions in (12)–(14), the adaptive updating law in (15)–(16), then the decentralized-based adaptive quantized control algorithm guarantees the following features:
-
The system output was confined to a small adjustable region in finite time.
-
All the signals of the CLS remain SGUUB.
-
The Zeno phenomenon was ruled out.
Proof
Let us select the Lyapunov function \(\bar{V}_{\nu }=\sum \nolimits _{m = 1}^M\sum \nolimits _{i=1}^{n_m}\frac{1}{2}\nu _{m,i}^2\) to prove the boundedness of \(\nu _{m,i}\).
where \(p_l=\max \left\{ {p_{m,1},1}\right\}\).
From [27], one has \(\xi _{m,l+1}=\mid \zeta _{m,{l+1}}-\alpha _{m,l}\mid \le \Delta _{m,l}\ (l=1,\cdots ,n_m-1)\) and define \(D_1=\min \{(c_{m,1}-p_{m,1}^2/2),\ (c_{m,2}-1/2), \ldots ,\)
\((c_{m,n_m}-1/2)\}\) and \(D_2=\frac{1}{2}\Delta _{m,l}^2\). Thus, (53) yields
Therefore, the boundness of \(\nu _{m,i}\) can be obtained. This completes the proof. \(\square\)
Let \(A=\min \bar{\delta }_m\left\{ 2c_{m,k}, \rho _{m,k}\mu _{m,k}, \bar{r}_{m,1}, \kappa _{m,k}\right\}\) and \(\chi _{n_m}= \sum \nolimits _{m = 1}^M \bar{\delta }_m\bar{\chi }_{m}\), (52) is further simplified as:
Firstly, by using Lemma 1, let us discuss the following two cases about \(\sum \nolimits _{m = 1}^M(s_{m,n_m}\bar{\kappa }_{m} \bar{\varsigma }_m-(1-\bar{\kappa }_{m})s_{m,n_m}u_{m}^{\rm{min}})\).
Case I: \(u_{m}^{\rm{min}}\le u_m^{\rm{Max}}\), there exist two probabilities, i.e.,
(a): \(\mid u_m\mid \le u_{m}^{\rm{min}}\),based on Lemma 1 with \(\mid \bar{\varsigma }_m(t)\mid \le (1-\bar{\kappa }_{m})u_{m}^{\rm{min}}/\bar{\kappa }_{m}\), it can be obtained
(b): \(\mid u_m\mid \ge u_{m}^{\rm{min}}\), based on Lemma 1 and \(\mid \bar{\varsigma }_m(t)\mid \le (\bar{\kappa }_{m}+\hslash _{m})\mid u_m\mid /\bar{\kappa }_{m}\), one has
Since \((1-\bar{\kappa }_{m})u_{m}^{\rm{min}}/(\bar{\kappa }_{m}+\hslash _{m})\ge u_m^{\rm{Max}}\), then we can obtain \(\mid u_m\mid \le u_m^{\rm{Max}}\le (1-\bar{\kappa }_{m})u_{m}^{\rm{min}}/(\bar{\kappa }_{m}+\hslash _{m})\). By invoking (57), the similar results with the form in (56) can be derived.
Case II
\(u_{m}^{\rm{min}}\ge u_m^{\rm{Max}}\), then \(\mid u_m\mid \le u_{m}^{\rm{min}}\). The similar result can be derived by referring to (a) in Case I.
In a nutshell, by combining with the Case I and II, the inequality (56) can be obtained accordingly.
Next, let us discuss the sign of \(\sum \nolimits _{m = 1}^M\sum \nolimits _{k = 1}^{n_m}\bar{\delta }_m(1-16\tanh ^2(s_{m,k}/\sigma _{m,k}))G_{m,k}\), two cases exist as follows:
Case A
\(s_{m,k}\in \Omega _{m_k}=\left\{ s_{m,k}\mid \mid s_{m,k}\mid <0.2554m_k\right\}\) for \(1\le m \le M\) and \(1\le k\le n_m\). Thus, \(s_{m,k}\) is bounded. According to \(z_{m,k}=s_{m,k}+\nu _{m,k}\) and the boundedness of \(\nu _{m,k}\), \(z_{m,k}\) is bounded. Because of (16), \(\hat{\Theta }_{m,k}\) is bounded. Due to \(\Theta _{m,k}\) being constant, the estimation error \(\tilde{\Theta }_{m,k}\) is also bounded. Thus, the boundedness of \(\alpha _{m,k}\) and \(u_m\) are bounded, and \(\zeta _{m,k}\) is also bounded. According to (9), \(x_{m,k}\) and \(\partial _m\) are bounded. Since \(G_{m,k}\) is a non-negative function, \(\bar{\delta }_m\left( 1-16\tanh ^2\left( \frac{s_{m,k}}{\sigma _{m,k}}\right) \right) G_{m,k}\) is bounded and suppose \(G_{m0}\) is its upper bound. On the basis of (56), one obtains
where \(\hat{\chi }_{n_m}=\chi _{n_m}+G_{m0}\). Furthermore, it can be obtained that \(0\le V \le [V(0)-(\hat{\chi }_{n_m}/A)]e^{-At}+ (\hat{\chi }_{n_m}/A)\). Consequently, all signals of the CLS are regulated to a compact set \(\Xi _1\).
Case B
\(s_{m,k}\notin \Omega _{m_k}\) for \(1\le m \le M\) and \(1 \le k \le n_m\). In light of Lemma 4, one has \(\bar{\delta }_m(1-16{\text{tanh}}^2(s_{m,k}/{m_k}))G_{m,k}\le 0\). Thus, (56) yields
The same as Case A, all signals of the CLS are regulated to a compact set \(\Xi _2\).
To sum up the above, it can be concluded that all signals in the CLS remain SGUUB. According to the above discussions, the parameters selection guideline of the proposed control solution is given in Algorithm 1.
Remark 2
In contrast to the classical ETC utilized in References [10, 11, 26], which requires continuous monitoring of the measurement error to determine whether it needs to be triggered, the developed self-triggered control scheme avoids the use of an additional observer by obtaining the next trigger moment from the current one. In addition, although the self-triggered control approach can reduce the transmission cost, the stabilization performance of the control system may be weakened based on the control signal being discontinuous between two consecutive trigger moments. In particular, the interconnected nonlinear time-delay system has a relatively complicated structure. To this end, the FTPF is introduced to achieve the stabilization control goal, which also increases the challenges and difficulties of control design to a certain extent.
Remark 3
In this brief, the trial-and-error-based tactic for choosing the design parameters is utilized to achieve satisfactory control performance. By increasing the design parameters \(c_{m,i}, \mu _{m,k}, \bar{r}_{m,1}\), and \(\kappa _{m,k}\) or decreasing the design parameters \(\Lambda _m, E_m\), and \(\rho _{m,i}\) such that the stabilization error is as small as possible. Theoretically speaking, provided that the design parameters are larger, it may result in a larger amplitude of the control input. Therefore, it is essential to make a trade-off between control performance and control energy by choosing the design parameters appropriately.
Remark 4
Following the above discussion, we can derive that the control input \(\bar{u}_m\) is bounded. Hence, \((\Lambda _m\mid \bar{u}_m(t)\mid +E_m)/(\max \left\{ L_m,\mid \dot{v}_m(t)\mid \right\} )\) is also bounded. Given the circumstances of (45), it is available that the lower bound for inter-execution intervals satisfies \(\mathop t\nolimits ^ * =t_{k+1}-t_k> 0\), that is, excluding the Zeno-behavior.
4 Simulation example
In this section, two simulation examples are used to validate the effectiveness of the proposed adaptive self-triggered quantized control scheme.
Example 1
In the background of the chemical industry, two-stage chemical reactors are universally prevalent. However, it is well known that interconnected nonlinear systems consisting of two two-stage chemical reactors have significant application potential, which not only models more complex chemical reaction systems but also improves the overall conversion rate. To sum up, the nonlinear systems composed of two two-stage chemical systems have certain practical implications to some extent.
Let us consider two-stage chemical reactors with dynamic uncertainly and delayed recycle streams, as exhibited in Fig. 1. As in [42], the mathematical model of the reactors can be expressed as:
where \(x_{m,1}\) and \(x_{m,2}\ (m=1,2)\) are the compositions to be controlled; \(q(u_m)\) is the quantized signal; the representation and values of the system parameters are shown in Table 1.
To confirm the validity of Assumption 2 for \(\psi _m\)-dynamic, let us choose \(\check{V}_m(\psi _m)=\psi _m^2\), then
By selecting \(l_m=2.5\), results in
Define \(\Upsilon _{m,1}(\mid \psi _m\mid )=0.6\psi _m^2, \Upsilon _{m,2}(\mid \psi _m\mid )=0.5\psi _m^2, r_{m,1}=1.5, r_{m,2}=0.625\), and \(\pi _{m0}(\Vert x_{m,1}\Vert )=2.5x_{m,1}^4\), Assumption 2 is certified successful. On the basis of Lemma 1 and taking \(\bar{r}_{m,1}=1.2 \in (0,r_{m,1})\), a dynamic signal \(\lambda _m\) is represented as:
Following the control design in Sect. 3, the corresponding design parameters are identified in Table 2. To intuitively show the superiority of the proposed methodology, a comparison simulation between the proposed approach and the algorithm in [16] is exhibited in Figs. 2, 3, 4, 5, and 6. Specifically, the states \(x_{m,1}\) and \(x_{m,2}\ (m=1,2)\) are shown for each subsystem in Fig. 2. Figure 3 plots the profiles of system output \(x_{m,1}\) for each subsystem. From Figs. 2 and 3, it is evidence obtained that the system states are SGUUB and system output \(x_{m,1}\) can converge better to a predetermined range \((-\ell _m,\ell _m)\) than the method in [16] in finite time. Furthermore, the response curves of the adaptive parameters \(\hat{\Theta }_{m,1}\) and \(\hat{\Theta }_{m,2}\) are depicted for each subsystem in Fig. 4. Figure 5 illustrates the graphs of the control input \(v_m\) and quantized input \(q(u_m)\) for each subsystem. It can be seen that the quantization amplitude of the proposed control method is relatively small. Figure 6 expresses the trigger interval for each subsystem. To sum up, the proposed FTPP quantized control approach is superior to that investigated in [16] for achieving a better trade-off between the networked control and stability performance, and also excluding the Zeno phenomenon.
Example 2
Consider the following interconnected nonlinear time-delay plants:
where \(f_{m,1}=(x_{m,1}-x_{m,1}^3)/(1+x_{m,1}^2),\ f_{m,2}=-(x_{m,1}^2+2x_{m,2}){\text{sin}}(x_{m,1}).\ g_{1,1}=0.1{\text{sin}}(x_{1,1}+x_{2,1}),\ g_{1,2}=0.2{\text{sin}}(x_{2,1}),\ g_{2,1}=0.1{\text{sin}}(x_{1,1}),\ g_{2,2}=0.2{\text{sin}}(x_{1,1}-x_{2,1}),\ g_{3,1}=0.1{\text{sin}}(x_{1,1}+x_{3,1}),\ g_{3,2}=0.2{\text{sin}}(x_{2,1}-x_{3,1}).\ \Delta _{1,1}=\psi _1x_{1,1}{\text{sin}}(x_{1,1}),\ \Delta _{1,2}=\psi _1x_{1,1}x_{1,2},\ \Delta _{2,1}=\psi _2x_{2,1}{\text{sin}}(x_{2,1}),\ \Delta _{2,2}=\psi _2x_{2,1}{\text{sin}}(x_{2,2}),\ \Delta _{3,1}=\psi _3x_{3,1}{\text{sin}}(x_{3,1}),\ \Delta _{3,2}=\psi _3x_{3,1}{\text{sin}}(x_{3,2}).\ \tau _{m,1}=0.2+0.08{\text{sin}}(2t),\ \tau _{m,2}=0.3+0.12{\text{sin}}(2t)\).
We select the \(\psi _m\)-dynamic, the RBFNN basis function, virtual control signals, and adaptive laws of the numerical simulations like Example 1. The different parameters related to Example 1 are selected as follows:\(x_{m,1}(0)=0.2,\ x_{m,2}(0) =1.2,\ c_{m,1}= c_{m,2}=5\ (m=1,2,3)\). To further illustrate the stabilization performance investigated in this paper, the responses of the states \(x_{m,1}\) and \(x_{m,2}\ (m=1,2,3)\) are exhibited for each subsystem in Fig. 7. Figure 8 depicts the graphs of system output \(x_{m,1}\) for each subsystem. The investigated self-triggered control scheme can guarantee that the system states in the resulting CLS are SGUUB and system output \(x_{m,1}\) can regulated to a predetermined range \((-\ell _m,\ell _m)\) in finite time. In addition, the response curves of the adaptive parameters \(\hat{\Theta }_{m,1}\) and \(\hat{\Theta }_{m,2}\) are shown for each subsystem in Fig. 9. Figure 10 plots the graphs of the control input \(v_m\) and quantized input \(q(u_m)\) for each subsystem. Figure 11 exhibits the trigger interval for each subsystem. To sum up, the presented self-triggered control approach can achieve a better trade-off between networked control and stability performance, and the Zeno phenomenon is also ruled out.
5 Conclusion
Self-triggered-based adaptive neural decentralized control strategy for interconnected nonlinear time-delay systems involving finite-time prescribed performance has been developed. By utilizing the hyperbolic tangent function and RBFNNs, the potential singularity issue has been eliminated. Meanwhile, an improved FOF was incorporated into the controller design, which can avoid the tremendous “ amount of calculation ” and the adverse impact of filter error has been excluded. Besides, the effect of bandwidth limitation was considered in the self-triggered adaptive FTPP control scheme, achieving a trade-off between stability performance and networked control. The presented controller ensured all signals in CLS are SGUUB, and the system output fluctuated to a small adjustable range within a finite time interval. Last but not least, the effectiveness and superiority of the proposed scheme have been verified by two simulation examples. In the future, the author will work to consider the self-triggered adaptive secure control issue for interconnected nonlinear time-delay systems with cyberattacks. In addition, inspired by [43] in dealing with the tracking of small-amplitude signals, a straightforward and effective nonlinearity is implemented in the controller, resulting in better signal-to-noise ratio performance, which will be another topic for our future research.
Data availability statements
Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
References
Tong S, Li Y, Liu Y (2021) Observer-based adaptive neural networks control for large-scale interconnected systems with nonconstant control gains. IEEE Trans Neural Netw Learn Syst 32(4):1575–1585
Hamdy M, ElGhazaly G (2014) Adaptive neural decentralized control for strict feedback nonlinear interconnected systems via backstepping. Neural Comput Appl 24:259–269
Zhu S, Han Y (2022) Adaptive decentralized prescribed performance control for a class of large-scale nonlinear systems subject to nonsymmetric input saturations. Neural Comput Appl 34:11123–11140
Li X, Yang G (2017) Adaptive decentralized control for a class of interconnected nonlinear systems via backstepping approach and graph theory. Automatica 76:87–95
Swaroop D, Hedrick JK, Yip PP, Gerdes JC (2000) Dynamic surface control for a class of nonlinear systems. IEEE Trans Autom Control 45(10):1893–1899
Tabatabaei SM, Kamali S, Arefi MM, Cao J (2020) Prescribed performance adaptive DSC for a class of time-delayed switched nonlinear systems in nonstrict-feedback form: Application to a two-stage chemical reactor. J Process Control 89:85–94
Cheng T, Niu B, Zhang J, Wang D, Wang Z (2021) Time-/event-triggered adaptive neural asymptotic tracking control of nonlinear interconnected systems with unmodeled dynamics and prescribed performance. IEEE Trans Neural Netw Learn Syst. https://doi.org/10.1109/TNNLS.2021.3129228
Farrell JA, Polycarpou M, Sharma M, Dong W (2009) Command filtered backstepping. IEEE Trans Autom Control 54(6):1391–1395
Lu S, Chen M, Liu Y, Shao S (2022) Adaptive NN tracking control for uncertain MIMO nonlinear system with time-varying state constraints and disturbances. IEEE Trans Neural Netw Learn Syst. https://doi.org/10.1109/TNNLS.2022.3141052
Song S, Park JH, Zhang B, Song X (2022) Event-based adaptive fuzzy fixed-time secure control for nonlinear CPSs against unknown false data injection and backlash-like hysteresis. IEEE Trans Fuzzy Syst 30(6):1939–1951
Liu L, Liu YJ, Tong S, Gao Z (2022) Relative threshold-based event-triggered control for nonlinear constrained systems with application to aircraft wing rock motion. IEEE Trans Ind Inf 18(2):911–921
Li X, Wu H, Cao J (2023) Prescribed-time synchronization in networks of piecewise smooth systems via a nonlinear dynamic event-triggered control strategy. Math Comput Simul 203:647–668
Wang J, Zhang H, Ma K, Liu Z, Chen CP (2022) Neural adaptive self-triggered control for uncertain nonlinear systems with input hysteresis. IEEE Trans Neural Netw Learn Syst 33(11):6206–6214
Chen W, Wang J, Ma K, Wu W (2022) Adaptive self-triggered control for a nonlinear uncertain system based on neural observer. Int J Control 95(7):1922–1932
Hayakawa T, Ishii H, Tsumura K (2009) Adaptive quantized control for linear uncertain discrete-time systems. Automatica 45(3):692–700
Sun H, Hou L, Zong G, Yu X (2020) Adaptive decentralized neural network tracking control for uncertain interconnected nonlinear systems with input quantization and time delay. IEEE Trans Neural Netw Learn Syst 31(4):1401–1409
Liu W, Lim C, Shi P, Xu S (2017) Backstepping fuzzy adaptive control for a class of quantized nonlinear systems. IEEE Trans Fuzzy Syst 25(5):1090–1101
Sun W, Su S, Xia J, Zhuang G (2021) Command filter-based adaptive prescribed performance tracking control for stochastic uncertain nonlinear systems. IEEE Trans Syst Man Cybern Syst 51(10):6555–6563
Song X, Sun P, Song S, Stojanovic V (2022) Event-driven NN adaptive fixed-time control for nonlinear systems with guaranteed performance. J Franklin Inst 359(9):4138–4159
Tong D, Liu X, Chen Q, Zhou W, Liao K (2022) Observer-based adaptive finite-time prescribed performance NN control for nonstrict-feedback nonlinear systems. Neural Comput Appl 34:12789–12805
Wang H, Bai W, Zhao X, Liu PX (2022) Finite-time-prescribed performance-based adaptive fuzzy control for strict-feedback nonlinear systems with dynamic uncertainty and actuator faults. IEEE Trans Cybern 52(7):6959–6971
Yan H, Li Y (2017) Adaptive NN prescribed performance control for nonlinear systems with output dead zone. Neural Comput Appl 28:145–153
Wang T, Tong S, Li Y (2013) Adaptive neural network output feedback control of stochastic nonlinear systems with dynamical uncertainties. Neural Comput Appl 23:1481–1494
Zhang L, Zhu L, Hua C, Qian C (2022) Adaptive decentralized control for interconnected time-delay uncertain nonlinear systems with different unknown control directions and deferred full-state constraints. IEEE Trans Neural Netw Learn Syst. https://doi.org/10.1109/TNNLS.2022.3171518
Mao J, Huang S, Xiang Z (2017) Adaptive tracking control for a class of non-affine switched stochastic nonlinear systems with unmodeled dynamics. Neural Comput Appl 28:1069–1081
Li M, Li S, Ahn CK, Xiang Z (2022) Adaptive fuzzy event-triggered command-filtered control for nonlinear time-delay systems. IEEE Trans Fuzzy Syst 30(4):1025–1035
Yu J, Shi P, Zhao L (2018) Finite-time command filtered backstepping control for a class of nonlinear systems. Automatica 92:173–180
Liu Y, Liu X, Jing Y (2018) Adaptive neural networks finite-time tracking control for non-strict feedback systems via prescribed performance. Inf Sci 468:29–46
Podlubny I (1998) Fractional differential equations. Academic, New York
Zhou J, Wen C, Yang G (2014) Adaptive backstepping stabilization of nonlinear uncertain systems with quantized input signal. IEEE Trans Autom Control 59(2):460–464
Jiang ZP, Praly L (1998) Design of robust adaptive controllers for nonlinear systems with dynamic uncertainties. Automatica 34(7):825–840
Tong S, Li Y (2010) Robust adaptive fuzzy backstepping output feedback tracking control for nonlinear system with dynamic uncertainties. Sci China Inf Sci 53:307–324
Polycarpou MM (1996) Stable adaptive neural control scheme for nonlinear systems. IEEE Trans Autom Control 41(3):447–451
Wang M, Chen B, Shi P (2008) Adaptive neural control for a class of perturbed strict-feedback nonlinear time-delay systems. IEEE Trans Syst Man Cybern Part B 38(3):721–730
Wang LX (1993) Stable adaptive fuzzy control of nonlinear systems. IEEE Trans Fuzzy Syst 1(2):146–155
Song X, Sun P, Song S, Wu Q, Lu J (2023) Event-triggered fuzzy adaptive fixedtime output-feedback control for nonlinear systems with multiple objective constraints. Int J Fuzzy Syst. 25:275-288
Liu H, Pan Y, Cao J, Wang H, Zhou Y (2020) Adaptive neural network backstepping control of fractional-order nonlinear systems with actuator faults. IEEE Trans Neural Netw Learn Syst 31(12):5166–5177
Li Y, Chen Y, Podlubny I (2009) Mittag–Leffler stability of fractional order nonlinear dynamic systems. Automatica 45(8):1965–1969
Song S, Park JH, Zhang B, Song X (2022) Composite adaptive fuzzy finite-time quantized control for full state-constrained nonlinear systems and its application. IEEE Trans Syst Man Cybernet Syst 52(4):2479–2490
Zhou Q, Shi P, Tian Y, Wang M (2015) Approximation-based adaptive tracking control for MIMO nonlinear systems with input saturation. IEEE Trans Cybern 45(10):2119–2128
Knuth DE (1997) The art of computer programming. Addison-Wesley, Reading
Hua C, Liu PX, Guan X (2009) Backstepping control for nonlinear systems with time delays and applications to chemical reactor systems. IEEE Trans Ind Electron 56(9):3723–3732
Bucolo M, Buscarino A, Fortuna L, Gagliano S (2021) Can noise in the feedback improve the performance of a control system? J Phys Soc Jpn 90(7):075002
Acknowledgements
This work was supported in part by the National Natural Science Foundation of China under Grants 61976081 and 62203153, in part by the Natural Science Fund for Excellent Young Scholars of Henan Province under Grant 202300410127, in part by Key Scientific Research Projects of Higher Education Institutions in Henan Province under Grant 22A413001, in part by Top Young Talents in Central Plains under Grant Yuzutong (2021) 44, in part by Technology Innovative Teams in University of Henan Province under Grant 23IRTSTHN012, in part by the Natural Science Fund for Young Scholars of Henan Province under Grant 222300420151, and in part by the Serbian Ministry of Education, Science and Technological Development (No. 451-03-68/2022-14/200108).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declared that they have no conflicts of interest in this work. We declare that we do not have any commercial or associative interest that represents a conflict of interest in connection with the work submitted.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Song, X., Sun, P., Song, S. et al. Quantized neural adaptive finite-time preassigned performance control for interconnected nonlinear systems. Neural Comput & Applic 35, 15429–15446 (2023). https://doi.org/10.1007/s00521-023-08361-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00521-023-08361-y