This paper investigated H∞ performance of the switched nonlinear systems with time-varying delay, which contains mixed modes, important characteristics of switched delay systems. Most of the references ignored the existence of mixed modes in such systems, which is a fatal mistake. This paper will correct the mistake of the existing references. The mixed modes give the H∞ control for switched nonlinear delay systems a challenge. With this difficulty, we take measures to handle the effects of mixed modes on such systems and achieve the H∞ performance of such systems. Firstly, the H∞ control analysis is divided into two cases, and the sufficient conditions for the exponential decay of the Lyapunov-Krasovskii functionals are given with the integral interval approach. Secondly, the combination of average dwell-time approach and subsystem activation rate is utilized to achieve the H∞ performance for the mixed-mode based switched systems with time-varying delay. Finally, simulation examples show the efficiency of the switching control strategy.
National Key R&D Program of China2017YBF13009001. Introduction
In recent decades, switched delay systems have emerged as a hot subject of research because of their applications in many fields, such as network control system, power system, and dynamic optimization [1, 2]. Lyapunov-Krasovskii functions are always utilized to analyze the stability for time delay systems [1, 3]. Average dwell time method is a common method for solving the asymptotic stability of switched delay system [4, 5]. However, switched time-varying delay system is hard to control because of the interaction between switching characteristics and time delay characteristics. It should be pointed out that the mixed modes are widespread in various switched delay systems, where the state depends on the present subsystem and the previous subsystem at the same time. It imposes great difficulty on the H∞ control problem. Without loss of generality, most of the results are imposed constraints on the switching rules and time delays, which can avoid the mixed modes happening. Reference [6] is concerned with the robust tracking control for switched delay systems, but the strict switching constrained condition avoids the mixed modes happening. Reference [7] investigated the passivity for the switched delay systems under stochastic disturbance. Reference [8] investigated finite-time control for the switched delay systems via mode-based average dwell-time approach. It effectively avoided the mixed modes happening by the method that each mode had its respective dwell time. The above references have not solved the effects of mixed modes on such systems. Reference [9] proposed the concept for the mixed modes of the switched delay system. It effectively solved the stabilization problem of switched networked delay systems, which has aroused great concern. Reference [10] analyzed the effects of mixed modes on a class of switched delay systems and derived the stability condition for such system, but the proposed method is not applicable to nonlinear delay systems. Lyapunov functional approach is an effective strategy to the delay systems. Reference [11] proposed conditions to achieve the decay of solutions for the delay differential equation via the Lyapunov functional method. Reference [12] defined a novel Lyapunov function to guarantee the boundedness of solutions for a vector Linard equation. To the author’s knowledge, H∞ control issues on mixed-mode based switched delay systems have not been explored, which motivates us to further investigate this work.
On the other hand, the study on the switched systems between stable and unstable modes has been a hot subject in recent years. Reference [13] was concerned with the H∞ performance of the uncertain switched neutral systems with stable and unstable modes, which first solved the H∞ control problems on such systems. Reference [14] derived exponential stability conditions for the switched singular systems with stable and unstable modes, which first solved the exponential stability issue on such systems. Reference [15] derived the stability conditions of switched delay systems, which effectively eliminated effects of the unstable subsystems on such systems. However, [13–15] have not taken the mixed modes into account, which is a fatal mistake. If the switching between the stable subsystem and the unstable subsystem occurs over two times during the period of the time delay of the system, the mixed modes will be inspired. So mixed modes are widespread in the switched delay systems and they should not be ignored. If the mixed modes are ignored, the stability of such system will not be achieved and the transient performance of the system will be adversely affected. The combination of the mixed modes and the switched systems with stable modes and unstable modes makes the H∞ control for such systems a challenge, where the state depends not only on the stable modes but also on the unstable modes. With the difficulty, we take the mixed modes and the H∞ control strategy into the switched nonlinear delay systems with stabilizable subsystems and unstabilizable subsystems and derived the conditions for the H∞ performance of the mixed-mode based switched nonlinear delay systems.
In this paper, the H∞ control for the mixed-mode based switched nonlinear delay systems is investigated. This focused on solving the effect of the mixed modes on the switched system with the parameter constraints. The main contributions can be listed as follows. First, correcting the mistake of [13–15], the effects of mixed modes are first considered for the switched nonlinear delay systems with stabilizable subsystems and unstabilizable subsystems. It is more reasonable to the H∞ analysis of the switched nonlinear systems with time-varying delays because the mixed modes are widespread in such systems. The mixed modes make the switching between the stabilizable subsystems and the unstabilizable subsystems more complex and give the H∞ analysis of such systems a challenge. Second, due to the effects of mixed modes, the combination of the integral interval for the Lyapunov function and the estimation of quadrature is utilized to demonstrate the exponential decay of the Lyapunov functions. Third, sufficient criteria are given to achieve the H∞ performance for the mixed-mode based switched delay systems via dwell-time strategy and the subsystem activation rate approach.
This paper is organized as follows. The problem formulation is introduced in Section 2. The main results are stated in Section 3. Simulation examples are in Section 4. Conclusion is provided in Section 5.
Notation. The notations utilized in this paper are quite standard. Rn and Rn×n refer to, respectively, the n-dimensional Euclidean space and n×n real matrices. Q>0(≥0) denotes that Q is positive (semipositive) definite. For a square matrix Q, λmax(Q) and λmin(Q) are its maximum eigenvalue and minimum eigenvalue, respectively.
2. Problem Formulation
Consider the switched nonlinear delay systems:(1)x˙t=fσtxt+Aσtxt+Dσxt-τt+Bσtut+ωtyt=Cσtxtxt=ϕt,t∈-τ,0where x(t)∈Rn denotes the state vector, u(t)∈Rr denotes the control input, ω(t)∈Rn denotes the exogenous disturbance which satisfies L2[0,∞), y(t)∈Rq denotes the output, fi(x(t)):Rn→Rn denotes a nonlinear function, ϕ(t) denotes initial condition defined on [-τ,0], σ(t):[0,∞)→Ω=1,2…N, and τ(t) is the time-varying delay. 0<τ(t)<τ for a constant τ and 0<τ˙(t)<δ for a constant δ; Ai, Di, and Bi are the matrices of the system and Bi is invertible matrix. Define that ith subsystem is stabilizable subsystem and jth subsystem is unstabilizable subsystem, where i∈N_ and j∈Ω∖N_.
Assumption 1.
For any x(t)∈Rn and x^(t)∈Rn, nonlinear function fi(x(t)) meets the global Lipschitz condition:(2)fixt-fix^t≤Mixt-x^twhere Mi∈Rn×n is a given Lipschitz constant matrix.
The performance index is as follows:(3)∫0∞e-αtxTtQxtdt≤γ2∫0∞ωTtωtdtwhere Q represents a positive weighting matrix, γ>0 represents the disturbance attenuation level, and α is positive scalar.
The controller of system (1) is chosen as(4)ut=Kσtxt-BσtTBσtBσtT-1fσt0where Kσ(t) is the controller gain.
Remark 2.
In general, assuming that fσ(t)(0)=0, the controller of the system will be u(t)=Kσ(t)x(t). However, in order to handle system (1) which dissatisfied such characteristic, we modify the controller. In order to deal with fσ(t)(0)≠0, the controller should contain Bσ(t)T(Bσ(t)Bσ(t)T)-1fσ(t)(0). Then the condition of the H∞ performance will be given in Section 3.
Taking controller (4) into system switched (1), the closed-loop system can be shown as(5)x˙t=Aσt+BσtKσtxt+Dσtxt-τt+fσtx-fσt0+ωtxt=ϕt,t∈-τ,0
If σ(t)=i, the system will be represented as(6)x˙=Ai+BiKixt+Dixt-τt+fixt-fi0+ωt
3. Main Result
Considering switched system (5), if there exists unstabilizable system in the subsystems, the H∞ control of switched system will be complicated. Taking z(t)=x˙(t), one has(7)zt=Aσt+BσtKσt+Dσtxt-Dσt∫t-τttzsds+ωt
Consider Lyapunov-Krasovskii functional as follows:(8)Vσtt=xTtQσtxt+∫-τ0∫t+θtzTseχσtt-sRzsdsdθwhere Ql and R are the symmetric positive matrices, Vi(t) denotes the stabilizable subsystems, and Vj(t) corresponds to the unstabilizable subsystems of (1). The parameters are chosen as χi=-α and χj=β. According to average dwell-time strategy, the unstabilizable modes activation rate satisfies(9)Δu=TuT1,T2T1-T2<α-α∗α+βwhere α∗∈(0,α), Tu represents the dwell time of unstabilizable subsystems during [T1,T2), and Δu represents the unstabilizable subsystems activation rate.
Then (8) is rewritten as(10)Vσtt=xTtQσtxt+∫t-τts-t+τzTseχσtt-sRzsds
The system is assumed that it is switched from the lth mode to the kth mode at the instant tk and (t-τ)∈[tk-1,tk). When t∈[tk,tk+τ), one has(11)Vkt=xTtQkxt+∫t-τtks-t+τzTseχlt-sRzsds+∫tkts-t+τzTseχkt-sRzsds
When t∈[tk+τ,tk+1),(12)Vkt=xTtQkxt+∫t-τts-t+τzTseχkt-sRzsds
Remark 3.
When t∈[tk,tk+τ), Vk(t) depends not only on the kth mode but also on the lth mode, which is called mixed modes. If tk+1-tk≤τ, the mixed modes must be considered. It makes the H∞ analysis of switched delay systems a challenge. References [6–8] avoid the occurrence of mixed modes by constraining the switching point tk+1-tk>τ. This has strict constraints on switching rules and time delays. The essential difference of this paper is to analyze the case tk+1-tk≤τ and to overcome the effects of mixed modes on system (6). The result of [15] can be regarded as a special case of this paper, where Case 1 has not happened.
In order to eliminate the effects of mixed modes on H∞ analysis of systems (5), the parameter constraints of switched system (5) are given as follows.
Lemma 4.
For given constants α>0,β>0, if there exist symmetric positive definite matrices Ql and R, any matrices Ul,Tl with appropriate dimensions, and positive parameters q,q0, and τ, such that the conditions(13)Θi=η11iη12i-UiDiUiUi∗η22i-TiDiTiTi∗∗η3300∗∗∗-I0∗∗∗∗-γ2I<0(14)Θ¯j=η¯11jη¯12j-UjDjUjUj∗η¯22j-TjDjTjTj∗∗η¯3300∗∗∗-I0∗∗∗∗-γ2I<0(15)Hj,k=ϱ11j1-δkDjR∗0≥-q0Ihold, then when σ(t)=i∈N_, we have(16)Vit≤e-αt-t0Vit0-∫t0te-αt-sΓsdsWhen σ(t)=j∈Ω∖N_, we have(17)Vjt≤eβt-t0Vjt0-∫t0teβt-sΓsdswhere k=1,2,δ1=0,δ2=δ,
η11i=Ai+BiKi+DiTUiT+Ui(Ai+BiKi+Di)+(α+1)Qi+MiTMi
η12i=Qi-Ui+Ai+DiTTiT
η22i=-TiT-Ti+τR
η33=-[τ-1e-ατ-q(α-α∗)eβd]R
η¯11j=Aj+BjKj+DjTUjT+Uj(Aj+BjKj+Dj)+(1-β)Qj+MjTMj
η¯12j=Qj-Uj+Aj+DjTTjT
η¯22j=-TjT-Tj+τR
η¯33=-τ-1e-ατR
ϱ11j=Aj+BjKjTR+R(Aj+BjKj)-βR+MjTR+RMj
Γ(s)=xT(s)Qx(s)-γ2ωT(s)ω(s)
Proof.
The Lyapunov functionals (11) will be analyzed in two cases, tk+1-tk≤τ and tk+1-tk>τ.
Case 1. tk+1-tk≤τ.
In this case, the switched system covers [tk,tk+1). Differentiating functional (11), one can obtain(18)V˙it=2xTtQizt+τzTtRzt-∫t-τtkzTθeχjt-θRzθdθ-∫tktzTθeχit-θRzθdθ+χj∫t-τtks-t+τzTθeχjt-θRzθdθ+χi∫tkts-t+τzTθeχit-θRzθdθ
Suppose that the system is switched from the jth mode to the ith mode at instant tk, and in the ith mode for t∈[tk,tk+1), one has(19)V˙it+αVit=2xTtQizt+τzTtRzt+xTtαQixt-∫t-τtkzTθeβt-θRzθdθ-∫tktzTθe-αt-θRzθdθ+β∫t-τtks-t+τzTθeβt-θRzθdθ+α∫t-τtks-t+τzTθeβt-θRzθdθ
For tk∈[t-τ,t], the quadrature estimation inequality is established as follows:(20)∫t-τtks-t+τYsds≤qtk-t+ττ∫t-τts-t+τYsdswhere q=eq0(1+p)τ/λmin(R) and Y(s)=zT(s)eβ(t-s)Rz(s). In order to obtain inequality (20), we take the derivative of Y(s):(21)Y˙s=Aj+BjKjzs+1-τ˙jsDjzs-τjs+f˙jxsTeβt-sRzs+βeβt-szTsRzs+zTseβt-sRAj+BjKjzs+1-τ˙jsDjzs-τjs+f˙jxs≤Aj+BjKjzs+1-τ˙jsDjzs-τjs+MjzsTeβt-sRzs+βeβt-szTsRzs+zTseβt-sRAj+BjKjzs+1-τ˙jsDjzs-τjs+Mjzs=ηTseβt-sΛjηswhere Λj=ϱ11j1-τ˙sDjR∗0 and η(s)=[z(s)z(s-τj(s))]T.
Since 0≤τ˙j(s)=1-∑k=12θkδk, where 0≤θk≤1, we can get Y˙(s)≥-q0η(s)2. Since z(s)2≤(Y(s)/λmin(R)), we can obtain(22)Y˙s≥-q01+pλminRYs
Taking ζ∈[t-τ,tk], we can get ln(Y(t)/Y(ζ))≥-(q0(1+p)/λmin(R))(t-ζ)≥-(q0(1+p)/λmin(R))=ln(1/q). Then one has(23)Yζ≤qYt
For tk∈[t-τ,t], (23) can help us further establish the estimation inequality (20).
During [t-τ,tk), the unstabilizable subsystem is active. According to (20), we can get(24)α+β∫t-τtks-t+τzTθeβt-θRzθdθ≤α+βtk-t+ττqτ∫t-τtzTθeβτRzθdθ≤α-α∗qτ∫t-τtzTθeβτRzθdθ
We have fi(x(t))-fi(0)≤Mix(t) from Assumption 1. Then we can further obtain(25)fixt-fi0Tfixt-fi0≤xTtMiTMixt
For matrices Ui,Ti, taking (24), (25), and (26) into (19), one obtains(27)V˙it+αVit+Γt≤2xTtQizt+τzTtRzt+xTtαQixt-∫t-τtzTse-ατRzsds+α-α∗qτ∫t-τtzTθeβτRzθdθ+xTtMiTMixt-fixt-fi0Tfixt-fi0+xTtQixt-γ2ωTtωt≤2xTt,zTtQiUi0Ti×ztϑi+τzTtRzt+xTtαQixt-∫t-τttzTse-ατ-α-α∗qτeβτRzsds+xTtMiTMixt-fixt-fi0Tfixt-fi0+xTtQixt-γ2ωTtωtwhere ϑi=-z(t)+(Ai+BiKi+Di)x(t)-Di∫t-τ(t)tz(s)ds+fi(x(t))+ω(t).
Let ς(t,s)=[xT(t),zT(t),zT(s),(fi(x(t))-fi(0))T,ωT(t)]T; then we have(28)V˙it+αVit+Γt≤1τit×∫t-τittςTt,sΞiςt,sdswhere Ξi are obtained by performing the congruent transformation with the matrix diag{I,I,τi(t),I,I} to both sides of (13). Then we can get V˙i(t)+αVi(t)+Γ(t)≤0. We can further get Vi(t)≤e-α(t-t0)Vi(t0)-∫t0te-α(t-s)Γ(s)ds.
Case 2. tk+1-tk>τ.
In this case, two intervals, [tk,tk+τ) and [tk+τ,t), should be considered. When t∈[tk,tk+τ), the proof is similar to Case 1. When t∈[tk+τ,t), Vi(t) can be shown in (12). Notice that there is no need for the time interval decomposition of the integral, so condition (13) can satisfy it. We omit it.
Similarly, if the system satisfies condition (14), it is not hard to obtain Vj(t)≤eβ(t-t0)Vj(t0)-∫t0teβ(t-s)Γ(s)ds with the similar proof above.
The proof is completed.
Remark 5.
Case 1 is a typical case for switched systems with mixed modes, which is essentially different from [15]. In [15], only the case where tk+1-tk>τ is considered. This has strict constraints on switching rules and time delays. In this paper, conditions (13)–(15) are the added parameter constraints compared with [15], which can make the Lyapunov-Krasovskii functional exponentially decay whether it is Case 1 or Case 2. There is no need for the design of the switching rules and time delays to avoid Case 1 occurring. So it can reduce the constraints on switching rules and time delays.
Remark 6.
From Lemma 4, we can get that if conditions (13)–(15) hold, (16) and (17) will be guaranteed. So if ω(t)≡0, we have Vi(t)≤e-α(t-t0)Vi(t0) for σ(t)=i∈N_ and Vj(t)≤eβ(t-t0)Vj(t0) for σ(t)=j∈Ω∖N_. Noticing that α>0 and β>0, we can say that the exponential decay rate for the stable subsystem is -α and the exponential divergence rate for the unstable subsystem is β.
Lemma 4 achieves the exponential decay and divergence of Lyapunov-Krasovskii functional (10). Here we analyze the H∞ control of switched systems (5), which contains stabilizable subsystems and unstabilizable subsystems.
Next we will give the switching H∞ control conditions for the switched delay system (5).
Theorem 7.
For given constants α>0,β>0, if there exist symmetric positive definite matrices Ql and R, any matrices Ul,Tl, and positive parameters q,q0, and τ such that (13)–(15) are satisfied and if the switching signal meets the activation rate (9) and the average dwell-time condition(29)Nσt0,t≤N0+t-t0Tα,Tα>Tα∗=lnκεα∗>lnκεα,the H∞ performance index (3) will be achieved, where ε=e(α+β)τ/2 and κ=inf{ν∣Qi≤νQj}.
Proof.
We suppose that Vi(t) is activated during [t2k,t2k+1) and Vj(t) is activated during [t2k+1,t2k+2).
First, one will prove the stability of system (5) when ω(t)≡0.
Then one can find from Lemma 4 that(30)Vσtt≤e-αt-t2kVit2k,ift2k≤t<t2k+1eβt-t2k+1Vjt2k+1,ift2k+1≤t<t2k+2
Then one has(31)Vit≤κVjt,Vjt≤eα+βτκVit
If the system is switched from mode j to i, we define ν=κ. If the system switched from mode i to mode j, we define ν=κe(α+β)τ. Then we can get(32)Vσtiti≤νVσti-ti-,i=1,2,…,
and taking Nu(t0,t) as total switching times from the unstabilizable subsystems to the stabilizable subsystems during [t0,t), one can get(33)Vσtt≤eχσtkt-tkνVσtk-tk-≤κNσt0,teα+βτNut0,t×e-αt-t0-Tut0,t+βTut0,tVσt0t0
It can be seen that Nu(t0,t)≤(Nσ(t0,t)+1)/2. Because Nσ(t0,t)≤N0+((t-t0)/Tα), we can further get(34)Vσtt≤eNσt0,tlnκeα+βτNσt0,t+1/2×e-α1-Tut0,t/t-t0+βTut0,t/t-t0t-t0Vσt0t0≤eα+βτ/2eNσt0,tlnκ+α+βτ/2×e-α1-Δu+βΔut-t0Vσt0t0≤εN0+1κN0elnκε-α+α+βΔut-t0Vσt0t0≤εN0+1κN0e-2λt-t0Vσt0t0where λ=-1/2[(lnκε/Tα)-α+(α+β)Δu].
Referring to (8), we can obtain(35)axt2≤Vσtt,Vσt0t0≤bxt02where a=minl∈N_λmin(Ql) and b=maxl∈N_λmax(Ql)+(τ2/2)λmax(R).
Therefore, one has x(t)2≤(1/a)Vσ(t)(t)≤(bεN0+1κN0/a)e-2λ(t-t0)xt02. So we have x(t)≤(bεN0+1κN0/a)e-λ(t-t0)xt0. It can be seen that λ>0. So the stability of system (5) is guaranteed when ω(t)≡0.
Next, one will prove the system achieves the performance index (3) when ω(t)≠0.
We can find from Lemma 4 that(36)Vσtt≤e-αt-t2kVit2k-∫t2lte-αt-sΓsds,ift2k≤t<t2k+1eβt-t2k+1Vjt2k+1-∫t2l+1teβt-sΓsds,ift2k+1≤t<t2k+2
Then we can get(37)Vσtt≤eχσtkt-tkνVσtk-tk--ν∫tlteχσtt-sΓsds≤κNσt0,teα+βτNut0,t×e-αt-t0-Tut0,t+βTut0,tVσt0t0-∫t0teα+βτNus,t-αt-s-Tus,t+βTus,t+Nσs,tlnμΓsds
Because Nσ(t0,t)≤N0+((t-t0)/Tα), it yields(38)Vσtt≤eNσt0,tlnκeα+βτNσt0,t+1/2×e-α1-Tut0,t/t-t0+βTut0,t/t-t0t-t0Vσt0t0-∫t0teNσs,tlnκeα+βτNσs,t+1/2×e-α1-Tus,t/t-s+βTus,t/t-st-sΓsds≤eα+βτ/2eNσt0,tlnκ+α+βτ/2×e-α1-Δu+βΔut-t0Vσt0t0-∫t0teα+βτ/2eNσs,tlnκ+α+βτ/2×e-α1-Δu+βΔut-sΓsds≤εNσt0,t+1κNσt0,te-αt-t0Vσt0t0-∫t0tεNσs,t+1κNσt0,te-αt-sΓsdsThen(39)1εVσtt≤e-αt-t0+Nσt0,tlnκεVσt0t0-∫t0te-αt-s+Nσs,tlnκεΓsdsLet t0=0; and multiplying e-Nσ0,tlnκε on both sides of (39), we can obtain(40)∫0te-αt-s-Nσ0,slnκεxTsQxsds≤e-αt-t0-Nσ0,t0Vσt0t0+γ2∫0te-αt-s-Nσ0,slnκεωTsωsds
Because Nσ(0,s)≤N0+s/Tα, N0>0, and Tα>lnκε/α, we obtain(41)Nσ0,slnκε≤N0lnκε+sα
Under the zero initial condition, taking (41) into (40), we have(42)∫0te-αtxTsQxsds≤γ2∫0te-αt-sωTsωsds
Integrating (42), one has(43)∫0∞∫0te-αtxTsQxsdsdt=∫0∞∫s∞e-αtxTsQxsdtds=1α∫0∞e-αsxTsQxsds
So we can obtain ∫0∞e-αsxT(s)Qx(s)ds≤γ2∫0∞ωT(s)ω(s)ds.
The proof is completed.
Remark 8.
Lemma 4 is proposed to handle the effects of mixed modes, where the parameter constraints (13)–(15) are utilized to guarantee the decay of the Lyapunov-Krasovskii functionals. Theorem 7 is proposed to achieve the H∞ performance based on Lemma 4, which is different from [15]. In [15], the mixed modes are not taken into account, which cannot achieve the stability of the system merely with average dwell-time approach and unstable subsystems activation rate.
Remark 9.
Traditionally, as for the systems with time-varying delays, the variable delay terms are usually used in the Lyapunov-Krasovskii functional. In this paper, instead of the variable delay terms, in Lyapunov-Krasovskii functional (8), the upper bound of time-varying delay is used. Then the exponential decay of functional (8) can be guaranteed by Lemma 4. The merits of the Lyapunov-Krasovskii functional not containing variable delay terms can be summarized as two points. On one hand, the Lyapunov-Krasovskii functional (8) does not contain the variable delay terms, so the number of constaint conditions is reduced. On the other hand, the Lyapunov-Krasovskii functional (8) can be converted to single integral forms (9). Then it is convenient to be analyzed by dividing it into two segments based on the mixed modes.
4. Simulation Examples
In this section, numerical examples are introduced to demonstrate the theoretical results.
Example 1.
Consider the switched system (6), where A1=0110, A2=-322-3, B1=-0.20.410, B2=-0.6-0.20.2-0.4, D1=0.250.150.25-0.5, D2=-0.10.10.20, τ(t)=0.5+0.5sint, τ=1, γ=1, f1xt=0.1cos0.1x10.1cos0.1x2, f2xt=0.2cos0.1x10.2cos0.1x2, f1(0)=[0.10.1]T≠0, f2(0)=[0.20.2]T≠0, C1=C2=10, M1=0.1000.1, and M2=0.2000.2. Note that λ(A1)=1,-1 and λ(A2)=-1,-5; then A1 is unstable with open loop and A2 is stable. External disturbance is as follows:(45)ωt=0.1sint,0≤t<20,t>2
We select α=0.16 and β=1.5. By solving LMI (13)–(15), we can get Q1=0.91320.53700.53701.0028, Q2=4.0186-0.9375-0.93753.8104, R=1.27850.39050.39051.4072, K1=0, and K2=5.3184-1.79403.77049.2840.
We can get κ=1.2603 and Tα∗=16.8205 according to equation (9) and condition (29). Then Δu≤0.1206. Because the mixed modes are taken into account, this paper is different from the case of [15]. The simulation results are shown in Figures 1 and 2.
The switching signal.
Output of the switched systems.
From Figures 1 and 2, we can obtain that the output of the system is exponentially stable under the switching of the stabilizable and unstabilizable subsystems. It also has good transient performance. It should be pointed out that if the system is not satisfied with conditions (13)–(15), the asymptotical stability will not be guaranteed. This will be verified in Example 2.
Example 2.
We use the control method of [15] for the switched system (5), which does not take the mixed modes into account. It means that the system is not satisfied with conditions (13)–(15). So the control strategy is merely based on average dwell time and unstable subsystems activation rate.
We select that A1, A2, B1, B2, C1, C2, D1, D2, τ(t), f1(x(t)), f2(x(t)), M1, M2, ω(t), α, and β are the same as Example 1. We select K1=-1.250.512 and K2=0.5-0.20.20.8. It is obvious that the parameters are not satisfied with conditions (14) and (15). This means that the effects of mixed modes are not handled. We can further get κ=1.2603 and Tα∗=16.8205 according to equation (9) and condition (29). The switching signal is the same as Figure 1. Then we can obtain the contrast simulation result listed as Figure 3.
Contrast for the output of the switched systems of [15].
As Figure 3 shows, the output is not stable. So we can say that if the parameters of the switched systems are not satisfied with the constraints of the mixed modes, the exponential stability for such system will not be guaranteed. It further proves the importance of mixed modes on such systems.
5. Conclusion
This paper addressed the H∞ control for the mixed-mode based switched delay systems. The two cases of the H∞ control analysis of switched delay systems were considered, and the sufficient conditions for the exponential decay of the Lyapunov functionals were developed. Then the combination of the average dwell-time approach and the subsystem activation rate was utilized to achieve the H∞ performance for the switched system with both stabilizable subsystems and unstabilizable subsystems. Finally, numerical examples and contrast simulation were given to show the effectiveness of the theoretical results.
Data Availability
The data source of the system model comes from [13]: “Stability of Switched Nonlinear Time-Delay Systems with Stable and Unstable Subsystems.”
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
This work was supported by National Key R&D Program of China under Grant 2017YBF1300900.
DeM. S.IbeasA.Stability results for switched linear systems with constant discrete delays200820082854314510.1155/2008/543145ChengL.ChenG.GaoW.ZhangF.LiG.Adaptive time delay compensator (ATDC) design for wide-area power system stabilizer201456295729662-s2.0-8490830045810.1109/TSG.2014.2347401WangZ.SunJ.ZhangH.Stability analysis of T-S fuzzy control system with sampled-dropouts based on time-varying lyapunov function method201811210.1109/TSMC.2018.2822482ChenH. B.ShiP.LimC. C.Stability of neutral stochastic switched time delay systems: an average dwell time approach201727351253210.1002/rnc.3588ZamaniI.ShafieeM.On the stability issues of switched singular time-delay systems with slow switching based on average dwell-time201424459562410.1002/rnc.2903MR3160375Zbl1284.931712-s2.0-84892535303LiQ.-K.ZhaoJ.DimirovskiG. M.Robust tracking control for switched linear systems with time-varying delays20082644945710.1049/iet-cta:20070344MR2428460LianJ.ShiP.FengZ.Passivity and passification for a class of uncertain switched stochastic time-delay systems201343131310.1109/TSMCB.2012.21988112-s2.0-84878640620RenH.ZongG.HouL.YiY.Finite-time control of interconnected impulsive switched systems with time-varying delay201627614315710.1016/j.amc.2015.12.012MR34520052-s2.0-84952360559LiQ. K.GaoW. W.YangN.Mixed-modes-based switched delay system approach to the stabilization of networked control systems201366986703LiQ.-K.LinH.Effects of mixed-modes on the stability analysis of switched time-varying delay systems201661103038304410.1109/TAC.2015.2501361MR3554040KorkmazE.TuncC.Inequalities and exponential decay of certain differential equations of first order in time varying delay2017261157166MR3676333Zbl1382.34071TunçC.Stability to vector Liénard equation with constant deviating argument20137331245125110.1007/s11071-012-0704-8MR30837772-s2.0-84880925231WangS.ShiT.ZhangL.Extended finite-time H∞ control for uncertain switched linear neutral systems with time-varying delays201515237738710.1016/j.neucom.2014.10.0472-s2.0-84921022918ZamaniI.ShafieeM.IbeasA.Stability analysis of hybrid switched nonlinear singular time-delay systems with stable and unstable subsystems20144551128114410.1080/00207721.2012.745026MR3172873TianY.CaiY.SunY.Stability of switched nonlinear time-delay systems with stable and unstable subsystems201724586810.1016/j.nahs.2016.11.003MR3624522Zbl1377.93144