
In practice, network operators tend to choose sparse communication topologies to cut costs, and the concurrent use of a communication network by multiple users commonly results in feedback delays. Our goal was to obtain the optimal sparse feedback control matrix K. For this, we proposed a sparse optimal control (SOC) problem governed by the cyber-physical system with varying delay, to minimize ||K||0 subject to a maximum allowable compromise in system cost. A penalty method was utilized to transform the SOC problem into a form that was constrained solely by box constraints. A smoothing technique was used to approximate the nonsmooth element in the resulting problem, and an analysis of the errors introduced by this technique was subsequently conducted. The gradients of the objective function concerning the feedback control matrix were obtained by solving the state system and a variational system simultaneously forward in time. An optimization algorithm was devised to tackle the resulting problem, building on the piecewise quadratic approximation. Finally, we have presented of simulations.
Citation: Sida Lin, Dongyao Yang, Jinlong Yuan, Changzhi Wu, Tao Zhou, An Li, Chuanye Gu, Jun Xie, Kuikui Gao. A new computational method for sparse optimal control of cyber-physical systems with varying delay[J]. Electronic Research Archive, 2024, 32(12): 6553-6577. doi: 10.3934/era.2024306
[1] | Ramalingam Sakthivel, Palanisamy Selvaraj, Oh-Min Kwon, Seong-Gon Choi, Rathinasamy Sakthivel . Robust memory control design for semi-Markovian jump systems with cyber attacks. Electronic Research Archive, 2023, 31(12): 7496-7510. doi: 10.3934/era.2023378 |
[2] | Xiaoming Wang, Yunlong Bai, Zhiyong Li, Wenguang Zhao, Shixing Ding . Observer-based event triggering security load frequency control for power systems involving air conditioning loads. Electronic Research Archive, 2024, 32(11): 6258-6275. doi: 10.3934/era.2024291 |
[3] | Majed Alowaidi, Sunil Kumar Sharma, Abdullah AlEnizi, Shivam Bhardwaj . Integrating artificial intelligence in cyber security for cyber-physical systems. Electronic Research Archive, 2023, 31(4): 1876-1896. doi: 10.3934/era.2023097 |
[4] | Saeedreza Tofighi, Farshad Merrikh-Bayat, Farhad Bayat . Designing and tuning MIMO feedforward controllers using iterated LMI restriction. Electronic Research Archive, 2022, 30(7): 2465-2486. doi: 10.3934/era.2022126 |
[5] | Jye Ying Sia, Yong Kheng Goh, How Hui Liew, Yun Fah Chang . Constructing hidden differential equations using a data-driven approach with the alternating direction method of multipliers (ADMM). Electronic Research Archive, 2025, 33(2): 890-906. doi: 10.3934/era.2025040 |
[6] | Yang Song, Beiyan Yang, Jimin Wang . Stability analysis and security control of nonlinear singular semi-Markov jump systems. Electronic Research Archive, 2025, 33(1): 1-25. doi: 10.3934/era.2025001 |
[7] | Meng Hu, Xiaona Cui, Lingrui Zhang . Exponential stability of Thermoelastic system with boundary time-varying delay. Electronic Research Archive, 2023, 31(1): 1-16. doi: 10.3934/era.2023001 |
[8] | Xudong Hai, Chengxu Chen, Qingyun Wang, Xiaowei Ding, Zhaoying Ye, Yongguang Yu . Effect of time-varying delays' dynamic characteristics on the stability of Hopfield neural networks. Electronic Research Archive, 2025, 33(3): 1207-1230. doi: 10.3934/era.2025054 |
[9] | Yi Gong . Consensus control of multi-agent systems with delays. Electronic Research Archive, 2024, 32(8): 4887-4904. doi: 10.3934/era.2024224 |
[10] | Lichao Feng, Dongxue Li, Chunyan Zhang, Yanmei Yang . Note on control for hybrid stochastic systems by intermittent feedback rooted in discrete observations of state and mode with delays. Electronic Research Archive, 2024, 32(1): 17-40. doi: 10.3934/era.2024002 |
In practice, network operators tend to choose sparse communication topologies to cut costs, and the concurrent use of a communication network by multiple users commonly results in feedback delays. Our goal was to obtain the optimal sparse feedback control matrix K. For this, we proposed a sparse optimal control (SOC) problem governed by the cyber-physical system with varying delay, to minimize ||K||0 subject to a maximum allowable compromise in system cost. A penalty method was utilized to transform the SOC problem into a form that was constrained solely by box constraints. A smoothing technique was used to approximate the nonsmooth element in the resulting problem, and an analysis of the errors introduced by this technique was subsequently conducted. The gradients of the objective function concerning the feedback control matrix were obtained by solving the state system and a variational system simultaneously forward in time. An optimization algorithm was devised to tackle the resulting problem, building on the piecewise quadratic approximation. Finally, we have presented of simulations.
A cyber-physical system (CPS) is a sophisticated, multi-layered system that combines computing, networking, and the physical environment [1]. By leveraging the integrated collaboration features of computation, communication, and control (3C) technology, it is possible to achieve real-time monitoring, control, and information services for large-scale engineering systems [2]. The applications of a CPS are wide-ranging. In study [3], the existing research on insider threat detection in a CPS is thoroughly reviewed and discussed. In [4], the problem of observer-based adaptive resilient control for a class of nonlinear CPSs is studied, taking the sensors that are vulnerable to deception attacks into account. The study in [5] focuses on the challenge of security control for CPSs subjected to aperiodic denial-of-service attacks. To reduce the need for explicit communication, the semantic knowledge within a CPS is leveraged, especially the use of physical radio resources to transmit potential informative data [6]. As discussed in [7], CPSs are vulnerable to numerous attacks and their attack surface continues to expand. To summarize, there are two key features of a CPS [8]: (ⅰ) large-scale, intricate systems focused on physical, biological, and engineering domains; (ⅱ) a network core that includes communication networks and computing resources for monitoring, coordinating, and controlling these physical systems. A CPS closely integrates these two essential components, enabling analysis and design within a unified framework.
Over the past two decades, extensive research has been conducted on control theory related to the CPS. Traditional CPS control designs, however, often produce dense feedback matrices, with the optimal controller relying on all the information within the feedback matrix [9]. In extensive networks, implementation costs can be considerable, and the computational load required for communication between the controller and the dynamical system can be heavy [10]. Two idealistic assumptions are embedded in traditional CPS control designs: communication costs are infinite, and the communication network is solely reserved for control purposes [11]. In reality, however, network operators typically prefer sparse communication topologies to reduce costs, and the shared usage of the communication network by various users commonly leads to feedback delays [12]. To address this issue, we propose a CPS system with a static state feedback controller u(t)=−Kx(t−τ), where K∈Rm×n and τ represents the delay, which can be either constant or variable, and is introduced by the communication process between the state x and the computation of the input u [2]. The network control design presented in this paper aims to achieve a balance between two primary objectives: (ⅰ) system performance, represented by the traditional cost function J0(K), and (ⅱ) the sparsity of the communication network. Thus, in this paper, we seek to solve the following problem: for a given CPS system, identify a feedback matrix K that balances system performance with controller sparsity.
Sparsity refers to a situation where the majority of elements in a vector or matrix are zero. The sparsity of a vector or matrix is characterized by its l0-norm. Sparsity is crucial in large-scale optimization problems, such as sound field reconstruction [13]. Employing sparsity not only minimizes storage requirements but also cuts transmission costs through vector compression. It streamlines a complex problem by extracting and using only the essential information from large datasets. In the absence of consideration for network topology, maximizing the sparsity of the feedback matrix typically has several reasons: (ⅰ) A sparse feedback matrix means that the controller only focuses on a small subset of variables in the system. This can significantly reduce the computational load and improve the responsiveness of real-time control, especially in large-scale systems. (ⅱ) Sparse control strategies can decrease the number of required sensors and actuators, thus reducing the overall system cost and energy consumption. This is particularly important in resource-constrained environments. (ⅲ) Sparse matrices often lead to simpler and more understandable control decisions. This can help designers more easily analyze and comprehend control strategies, making debugging and optimization processes more effective. (ⅳ) Sparsity may enhance the system's tolerance to failures of certain components. If some sensors or actuators fail, the system can still maintain functionality through other effective connections. (ⅴ) In some cases, sparse control strategies can exhibit better robustness to noise and disturbances, as they rely only on key parts of the system, reducing sensitivity to the overall system state.
Currently, sparse optimization has been extensively applied in various fields such as blade tip timing [14], robotic machining systems [15], and perimeter control [16,17]. Sparse optimization models can be generally categorized into two types [18]: (ⅰ) l0-regularization optimization problems modifying the traditional objective function by incorporating the l0-norm into a new objective function, and (ⅱ) sparse constrained optimization problems including the l0-norm within the constraints. However, both types of problems are NP-hard. In earlier studies, methods for solving the l0-norm minimization problem are typically categorized into model transformation techniques and direct processing techniques. The common feature of the model transformation method is to approximate the l0-norm with the l1-norm [19]. In terms of algorithms, several methods have been studied, including the iterative hard-thresholding algorithm (IHTA) [20], fast iterative shrinkage-thresholding algorithm (FISTA) [21], augmented method (ALM) [22], and alternating direction method of multipliers (ADMM) [23].
IHTA has two advantages as follows [20]: (ⅰ) It is straightforward to implement, making it accessible for various applications; and (ⅱ) it effectively promotes sparsity in solutions, which is beneficial in many signal processing and statistical tasks. There exists two disadvantages for IHTA as follows [20]: (ⅰ) It can converge slowly, especially for large-scale problems; and (ⅱ) it may struggle with nonsmooth functions, limiting its applicability in some optimization scenarios. FISTA offers two key benefits [21]: (ⅰ) It significantly accelerates the convergence compared to IHT by using Nesterov's acceleration, making it suitable for large datasets; and (ⅱ) it can handle a variety of loss functions and regularization terms, providing versatility in applications. FISTA has two drawbacks, outlined below [21]: (ⅰ) The implementation is more complex than IHT, requiring careful tuning of parameters; and (ⅱ) it may require more memory for storing additional variables, which could be a concern for very large problems. ALM has two notable benefits, listed below [22]: (ⅰ) It is effective for problems with constraints, making it a good choice for constrained optimization; and (ⅱ) it generally exhibits robust convergence properties, especially for nonconvex problems. ALM presents two notable drawbacks, detailed below [22]: (ⅰ) The performance heavily depends on the choice of parameters, which can be challenging to optimize; and (ⅱ) the method can be computationally intensive, particularly for high-dimensional problems. ADMM offers two key advantages, as outlined below [23]: (ⅰ) It allows large problems to be decomposed into smaller subproblems, which can be solved more easily; and (ⅱ) it handles a wide range of objective functions and constraints, making it versatile for various applications. ADMM comes with two significant downsides, as outlined below [23]: (ⅰ) While it has good convergence properties, it can sometimes converge slowly compared to other methods; and (ⅱ) like ALM, the performance can be sensitive to the choice of parameters, requiring careful tuning.
Optimal sparse control theory is now well developed. In [24], it explores the optimal control problem involving sparse controls for a Timoshenko beam, including its numerical approximation using the finite element method and its numerical solution through nonsmooth techniques. In [25], the study is focused on sparse optimal control for continuous-time stochastic systems using a dynamic programming approach, analyzing the optimal control through the value function. In [26], the study aims to develop a sparse tube-based robust economic model predictive control scheme. In [27], a novel sparse control strategy for acidic wastewater sulfidation is presented to ensure the continuous and safe HSS process. In [28], the construction of an eigenfunction vector of the Koopman operator is based on the sparse control strategy. However, for the CPS system with varying delay, these methods in [24,25,26,27,28] are not sufficient to determine an optimal sparse control policy. In this paper, we first demonstrate the existence of the partial derivatives of the system state with respect to the elements of the feedback matrix, and then use this to show that the gradient of the cost function can be computed by solving the state system and a variational system forward in time. In this case, our optimal control policy would be more straightforward and efficient for real-world applications compared to the methods in [24,25] depending on the numerical approximation, which could introduce a gap between the real and the approximated control policy.
Many existing optimal sparse control theories for the CPS assume constant delays, which can oversimplify the dynamics of a CPS where delays are often variable and unpredictable. This can lead to suboptimal control strategies that do not account for the true nature of system behavior. Sparse control with feedback delay plays a vital role in the CPS. (ⅰ) Effective control strategy design is essential in the CPS. Sparse control can optimize input signals, particularly when there are limited sensors and actuators or when energy efficiency is a priority. It is also important to account for feedback delays in the control algorithms to prevent potential instability. (ⅱ) The CPS generally requires immediate responses, but feedback delays necessitate that control strategies be equipped to mitigate their effects. By simplifying control signals, sparse control can improve responsiveness while ensuring that performance remains stable despite these delays. (ⅲ) Additionally, sparse control can help minimize the need for computational power and communication bandwidth in the CPS environments. The optimal control of communication networks described previously only considers the system cost. However, network operators frequently opt for sparse communication topologies to lower costs. With this in mind, to determine the optimal feedback control matrix K, we propose a sparse optimal control (SOC) problem based on the CPS system with varying delay, aiming to minimize the sparsity of the controller subject to a maximum allowable compromise in system cost. A penalty method is employed to convert the SOC problem into a problem that is constrained only by box constraints. A smoothing technique is applied to approximate the nonsmooth component in the resulting problem. Subsequently, an analysis of the errors introduced by the smoothing technique is conducted. The gradients of the objective function with respect to the feedback control matrix are determined by solving both the state system and a variational system forward in time. Building upon the piecewise quadratic approximation [29], an optimization algorithm is developed to address the resulting problem. Finally, the paper provides the outcomes of the simulations.
The rest of the paper is organized as follows. We first describe the SOC problem based on the CPS system in Section 2. In Section 3, we develop an optimization algorithm to solve the SOC problem. Finally, a numerical example is given in Section 4.
Let In be the set of {1,2,...,n}.
We consider the following linear time-invariant (LTI) system [2]:
˙x(t)=Ax(t)+Bu(t), |
where x∈Rn is the state and u∈Rm is the control input. A∈Rn×n and B∈Rn×m. Assume that (A,B) is controllable.
Two idealistic hypotheses are involved in conventional CPS control designs: that communication costs are unlimited and that the communication network is exclusively allocated for control purposes. In practice, however, network operators often favor sparse communication topologies to minimize costs, and the shared use of the same communication network by multiple users frequently results in feedback delay.
After the sparsity level of matrix K∈Rm×n is achieved, the bandwidth c is equally redistributed among the remaining links. Assume that the communication network follows frequency division multiplexing. Then, delay τ can be defined by [2]:
τ(‖K‖0)=τt+τp=Z(‖K‖0,c,τp):=κ(‖K‖0/c)+τp, | (2.1) |
where ‖K‖0 denotes the number of nonzero elements in K, and κ:R→R is a positive function. Eq (2.1) implies that τ will change as ‖K‖0 changes. This change is captured by the function Z(⋅):R×R×R→R. The transmission of state xl for the computation of input ui is anticipated to encounter a delay denoted as τil (expressed in seconds), i∈Im,l∈In. This delay comprises two distinct components: τil=τpil+τtil, where τpil denotes the propagation delay, and τtil represents the transmission delay. The parameter τpil is characterized as the quotient of the link length divided by the speed of light, assumed to possess a uniform value denoted as τp across all pairs i,l. Our assumption posits an equal allocation of bandwidth for the communication link connecting any lth sensor to any ith actuator. Consequently, this implies that τtil maintains a uniform value across all i,l pairs [2]. Henceforth, we denote τil uniformly as τ across all pairs i,l. In practice, potential deviations of τil from the designated τ due to variations in traffic and uncertainties within the network are acknowledged.
This controller will be deployed in a distributed manner utilizing a communication network, as depicted in Figure 1. This figure presents the CPS represented by a closed-loop system architecture. The ith control input is written as ui(t)=−∑nl=1Kilxl(t−τ(‖K‖0)),i∈Im. Then accordingly the closed-loop system is written as:
˙x(t)=f(x(t),˜x(t),K)=Ax(t)−BKx(t−τ(‖K‖0)),x(t)=ν,t≤0, | (2.2) |
where ν∈Rn is a given vector, where each of its elements is assumed, without loss of generality, to be 0.5; and ˜x(t) denotes x(t−τ(‖K‖0)). Let x(⋅|K) be the solution of system (2.2) corresponding to the feedback matrix K∈Rm×n.
With reference to the delay mentioned above, we introduce the corresponding system cost as given below [30]:
J0(K)=(x(T|K))⊤Sx(T|K)+∫T0[(x(t|K))⊤Qx(t|K)+(u(t))⊤Wu(t)]dt, | (2.3) |
where T is the final time, the matrix W∈Rm×m is symmetric positive definite, the feedback controller u=−Kx(t−τ(‖K‖0)), and the matrices S∈Rn×n and Q∈Rn×n are symmetric positive semidefinite.
We now present the feedback optimal control problem as follows.
ProblemP1:minK∈Rm×nJ0(K)s.t.˙x(t)=f(x(t),˜x(t),K),x(t)=ν,t≤0, |
where J0(K) is given by (2.3).
Gradient-based optimization methods [31] can be used to solve Problem P1. Let K∗1∈Rm×n be the optimal feedback matrices for Problem P1. However, these matrices tend to be rather dense, and for large networks, the implementation cost will be expensive. Furthermore, the computation burden of the controller will be high because the state information is required to be transmitted through the communication network. Thus, we introduce the following Problem P2 given by
ProblemP2:minK∈Rm×n‖K‖0s.t.˙x(t)=f(x(t),˜x(t),K),x(t)=ν,t≤0,|J0(K)−J0(K∗1)|≤ε, | (2.4) |
where ‖K‖0 denotes the number of nonzero entries of the feedback matrix K∈Rm×n, J0(K) is given by (2.3), and J0(K∗1) is the benchmark optimal system cost obtained through solving Problem P1. ε is a small number that is used to ensure that the system cost is not greatly affected during the sparsity process of the feedback matrix K∈Rm×n.
Obviously, Problem P2 balances system performance and the sparse level of the the feedback matrix K∈Rm×n.
The feedback matrix K is decomposed as n column vectors, i.e., K=(K1,K2,…,Kn)∈Rm×n. Note that ‖Kl‖0,l∈In, regularization is NP-hard. Thus, it is difficult to solve. In the past two decade, many approximation methods, such as ‖Kl‖1 and ‖Kl‖qq (0<q<1) have been proposed. In [29], the l0-norm of the vector is approximated by a piecewise quadratic approximation (PQA) method. In this paper, we shall extend PQA to spare the feedback matrix K.
Remark 1. We shall illustrate that P(Kl),l∈In, performs better than other common approximations of ‖Kl‖0,l∈In, on [−e,e], e={1,1,…,1}∈Rm.
For l∈In, Figure 2 shows the approximation effects of ‖Kl‖1, ‖Kl‖1/21/2, ‖Kl‖1/31/3, ‖Kl‖1−‖Kl‖2, and P(Kl) for the one-dimensional case in [–1, 1] [29]. Obviously, for l∈In, P(Kl) is superior to ‖Kl‖1 for approximating the l0-norm when |Kli|≤1,i∈Im. For l∈In, when 0.38≤|Kli|≤1,i∈Im, P(Kl) gives a better approximation for ‖Kl‖0, and when 0.61≤|Kli|≤1,i∈Im, P(Kl) is better than ‖Kl‖1/31/3. Also, for l∈In, P(Kl) is superior to ‖Kl‖1−‖Kl‖2, which is identically equal to 0 and has a large gap with ‖Kl‖0 in [–1, 1].
On this basis, we use the piecewise quadratic function [29] to approximate ‖Kl‖0 over [−e,e].
P(Kl)=−(Kl)⊤Kl+2‖Kl‖1,Kl∈Rm,l∈In. |
If we choose f(Kl)=−‖Kl‖22 and g(Kl)=2‖Kl‖1, then
F(K)=n∑l=1P(Kl)=n∑l=1[f(Kl)+g(Kl)], |
where g is a proper closed convex and possibly nonsmooth function; f is a smooth nonconvex function of the type C1,1Lf(Rn), i.e., continuously differentiable with Lipschitz continuous gradient
‖∇f(Kl)−∇f(yl)‖≤Lf‖Kl−yl‖,Kl∈Rm,yl∈Rm,l∈In, |
with Lf>0 denoting the Lipschitz constant of ∇f.
Based on the piecewise quadratic approximation [29], Problem P2 can be approximated as given below:
ProblemP3:minK∈Rm×nF(K)s.t.˙x(t)=f(x(t),˜x(t),K),x(t)=ν,t≤0,|J0(K)−J0(K∗1)|≤ε, |
where J0(K), J0(K∗1), and ε are defined in Problem P2.
Since the objective function F(K) is nonsmooth, it is difficult to solve ProblemP3 by using gradient-based algorithms. To overcome this difficulty, we aim to find a smooth function for the objective function F(K) described in ProblemP3.
First, we introduce the following notation: for x,y,z∈Rm,
x=max{y,z}⇔xi=max{yi,zi},∀i∈Im. |
Lemma 1. If we define that p(Kli)=max{Kli,0} and q(Kli)=max{−Kli,0}, then the following properties are satisfied:
(1) Kli = p(Kli)−q(Kli), i∈Im,l∈In; and
(2) ‖Kl‖1=∑mi=1[p(Kli)+q(Kli)], l∈In.
Proof. (1) For l∈In, and i∈Im, we have
p(Kli)−q(Kli)=max{Kli,0}−max{−Kli,0}, |
Kli≥0⇒p(Kli)−q(Kli)=Kli−0=Kli, |
Kli<0⇒p(Kli)−q(Kli)=0−(−Kli)=Kli, |
which imply that Kli=p(Kli)−q(Kli),i∈Im,l∈In.
(2) For l∈In, and i∈Im, we get
p(Kli)+q(Kli)=max{Kli,0}+max{−Kli,0}, |
Kli≥0⇒p(Kli)+q(Kli)=Kli−0=|Kli|, |
Kli<0⇒p(Kli)+q(Kli)=0−(−Kli)=|Kli|, |
which prove that ‖Kl‖1=m∑i=1|Kli|=m∑i=1[p(Kli)+q(Kli)].
Based on Lemma 1, we obtain
F(K)=n∑l=1[f(Kl)+g(Kl)]=n∑l=1(−‖Kl‖22+2‖Kl‖1)=n∑l=1(−‖Kl‖22+2m∑i=1[p(Kli)+q(Kli)])=n∑l=1(−‖Kl‖22+2m∑i=1[max{Kli,0}+max{−Kli,0}]). |
Then ProblemP3 is equivalent to the following problem:
ProblemP4:minK∈Rm×nF1(K)s.t.˙x(t)=f(x(t),˜x(t),K),x(t)=ν,t≤0,|J0(K)−J0(K∗1)|≤ε, |
where
F1(K)=n∑l=1[f(Kl)+g(Kl)]=n∑l=1(−‖Kl‖22+2‖Kl‖1)=n∑l=1(−‖Kl‖22+2m∑i=1[p(Kli)+q(Kli)])=n∑l=1(−‖Kl‖22+2m∑i=1[max{Kli,0}+max{−Kli,0}]). |
Clearly, the function max{x,0} is nondifferentiable with respect to x at x=0. Thus, in order to smooth it, a smooth function ϕ(x,σ) [32] is introduced as follows:
ϕ(x,σ)=2σ2√x2+4σ2−x, |
where σ>0 is an adjustable smooth parameter.
Lemma 2. For any x∈R and σ>0, the smooth function ϕ(x,σ) has the following properties:
(1) limσ→0+ϕ(x,σ)=max{x,0},
(2) ϕ(x,σ)>0,
(3) 0<ϕ′(x,σ)=12(x√x2+4σ2+1)<1,
(4) 0<ϕ(x,σ)−max{x,0}≤σ.
Lemma 2 shows that the function ϕ(x,σ) is an effective smooth approximation for the function max{x,0}, and the approximation level can be controlled artificially by adjusting the value of smooth parameter σ.
Therefore, ProblemP4 can be approximated by the following problem:
ProblemP5:minK∈Rm×nF2(K)s.t.˙x(t)=f(x(t),˜x(t),K),x(t)=ν,t≤0,|J0(K)−J0(K∗1)|≤ε, |
where
F2(K)=n∑l=1(−‖Kl‖22+2m∑i=1[ϕ(Kli,σ)+ϕ(−Kli,σ)]). |
Note that the function F2(K) is continuously differentiable. Thus, ProblemP5 is a constrained optimal parameter selection problem, which can be solved efficiently by using any gradient-based algorithm.
Theorem 1 presents that the optimal solution of ProblemP5 is also the optimal solution of ProblemP4 as long as σ→0 and an error estimation between the solutions of ProblemP4 and ProblemP5 is given.
Lemma 3. For any σ>0, one has
0<F2(K)−F1(K)≤4mnσ. |
Proof. By using Lemma 2, one has
0<ϕ(x,σ)−max{x,0}≤σ, |
and
F2(K)−F1(K)=2n∑l=1m∑i=1(ϕ(Kli,σ)+ϕ(−Kli,σ)−max{Kli,0}−max{−Kli,0}). |
Thus,
0<F2(K)−F1(K)≤4mnσ. |
This completes the proof of Lemma 3.
Theorem 1. Let K4 and K5 be the optimal solutions of ProblemP4 and ProblemP5, respectively. Then, we have
0<F2(K5)−F1(K4)≤4mnσ. |
Proof. By using Lemma 3, one has
0<F2(K4)−F1(K4)≤4mnσ, |
0<F2(K5)−F1(K5)≤4mnσ. |
Note that K4 is the optimal solution of ProblemP4. Then, we have
F1(K5)≥F1(K4), |
which indicates that
F2(K5)−F1(K5)≤F2(K5)−F1(K4). |
Note that K5 is the optimal solution of ProblemP5. Then, we have
F2(K4)≥F2(K5), |
which indicates that
F2(K5)−F1(K4)≤F2(K4)−F1(K4). |
Then, we have 0<F2(K5)−F1(K5)≤F2(K5)−F1(K4)≤F2(K4)−F1(K4)≤4mnσ, which implies that
0<F2(K5)−F1(K4)≤4mnσ. |
This completes the proof of Theorem 1.
Remark 2. Theorem 1 shows that the optimal solution of Problem P5 is an approximate optimal solution of Problem P4, as long as the adjustable parameter σ is sufficiently small.
The inequality constraint |J0(K)−J0(K∗1)|≤ε is equivalent to
max{|J0(K)−J0(K∗1)|−ε,0}=0, |
and equivalent to
max{J0(K)−J0(K∗1)−ε,0}=0, | (3.1) |
and
max{−J0(K)+J0(K∗1)−ε,0}=0. | (3.2) |
Then, by using the idea of the penalty function method described by [31], the equality constraints (3.1) and (3.2) are appended to the objective function of ProblemP5 to form an augmented objective function. Thus, a penalty problem can be defined as follows:
ProblemP6:minK∈Rm×nG0(K)s.t.˙x(t)=f(x(t),˜x(t),K),x(t)=ν,t≤0, |
where
G0(K)=F2(K)+γ1max{J0(K)−J0(K∗1)−ε,0}+γ2max{−J0(K)+J0(K∗1)−ε,0}=n∑l=1(−‖Kl‖22+2m∑i=1[ϕ(Kli,σ)+ϕ(−Kli,σ)])+γ1max{J0(K)−J0(K∗1)−ε,0}+γ2max{−J0(K)+J0(K∗1)−ε,0}, |
with γ1 and γ2 being the penalty parameters. It is important to note that violations of the equality constraints in Eqs (3.1) and (3.2) are addressed through the integral term G0(K) in ProblemP6. It can be demonstrated that by selecting sufficiently high values for γ1 and γ2, any minimizer of G0(K) in ProblemP6 within the region defined by γ1>γ∗ and γ2>γ∗ (where γ∗ denotes the threshold for the penalty parameters) will also satisfy the feasibility conditions of ProblemP5. Therefore, a feasible solution to ProblemP5 can be effectively found by minimizing G0(K) in ProblemP6 with appropriately chosen penalty values for γ1 and γ2.
By adopting the function ϕ(x,σ) to approximate the function max{x,0} again, ProblemP6 can be written as the following problem:
ProblemP7:minK∈Rm×nG(K)s.t.˙x(t)=f(x(t),˜x(t),K),x(t)=ν,t≤0, |
where
G(K)=n∑l=1(−‖Kl‖22+m∑i=1[ϕ(Kli,σ)+ϕ(−Kli,σ)])+γ1ϕ(J0(K)−J0(K∗1)−ε,σ)+γ2ϕ(−J0(K)+J0(K∗1)−ε,σ). |
Note that the function G(K) is continuously differentiable. Thus, ProblemP7 is an unconstrained optimal parameter selection problem, which can be solved efficiently by using any gradient-based algorithm.
Theorem 2 shows that the optimal solution of ProblemP7 is also the optimal solution of ProblemP6 as long as σ→0 and an error estimation between the solutions of ProblemP5 and ProblemP7 is given.
Definition 1. A control input K7 is σ-feasible to Problem P7, if the control input K7 satisfies the following inequality constraint:
|J0(K7)−J0(K∗1)|−ε≤σ. |
Lemma 4 and Theorem 2 are the variations of Lemma 3 and Theorem 1, respectively.
Lemma 4. For any σ>0, one has
0<G0(K)−G(K)≤(γ1+γ2)σ, |
where 0<γ1 and γ2<1 are the penalty parameters.
Proof. By using Lemma 2, we have
0<ϕ(x,σ)−max{x,0}≤σ, |
and
G0(K)−G(K)=γ1(ϕ(J0(K)−J0(K∗1)−ε,σ)−max{J0(K)−J0(K∗1)−ε,0})+γ2(ϕ(−J0(K)+J0(K∗1)−ε,σ)−max{J0(K)−J0(K∗1)−ε,0}). |
Thus, we have
0<G0(K)−G(K)≤(γ1+γ2)σ. |
This completes the proof of Lemma 4.
Theorem 2. Let K6 and K7 be the optimal solutions of ProblemP6 and ProblemP7, respectively. Then, we have
0<G(K7)−G0(K6)≤(γ1+γ2)σ. |
Proof. By using Lemma 4, one has
0<G0(K6)−G(K6)≤(γ1+γ2)σ, |
0<G0(K7)−G(K7)≤(γ1+γ2)σ. |
Note that K6 is the optimal solution of ProblemP6. Then, we have
G0(K7)≥G0(K6), |
which indicates that
G(K7)−G0(K7)≤G(K7)−G0(K6). |
Note that K7 is the optimal solution of ProblemP7. Then, we have
G(K6)≥G(K7), |
which indicates that
G(K7)−G0(K6)≤G(K6)−G0(K6). |
Then, 0<G(K7)−G0(K7)≤G(K7)−G0(K6)≤G(K6)−G0(K6)≤(γ1+γ2)σ, which implies that
0<G(K7)−G0(K6)≤(γ1+γ2)σ. |
This completes the proof of Theorem 2.
Remark 3. Theorem 2 shows that the optimal solution of ProblemP7 is an approximate optimal solution of ProblemP6, as long as the adjustable parameter σ is sufficiently small.
Then, we can obtain the following theorem.
Theorem 3. Let K6 and K7 be the optimal solutions of ProblemP6 and ProblemP7, respectively. If K6 is feasible to ProblemP5 and K7 is σ−feasible to ProblemP5, then we obtain
−(γ1+γ2)σ<F2(K7)−F2(K6)≤(γ1+γ2)σ. |
Proof. Note that K6 is feasible to ProblemP5. Then, one has
max{J0(K)−J0(K∗1)−ε,0}=0, |
and
max{−J0(K)+J0(K∗1)−ε,0}=0. |
Then, we have J0(K)−J0(K∗1)−ε≤0 and −J0(K)+J0(K∗1)−ε≤0. Note that K7 is σ-feasible to ProblemP5 in Definition 1 and the smooth function ϕ(x,σ) is strictly monotone increasing (the first derivative is positive, see Lemma 1(3)). Thus, we obtain
ϕmax(J0(K)−J0(K∗1)−ε,σ)=ϕ(σ,σ)=12(√5+1)σ, |
and
ϕmax(−J0(K)+J0(K∗1)−ε,σ)=ϕ(σ,σ)=12(√5+1)σ. |
By using Lemma 2(2) that ϕ(x,σ)>0, we obtain
0<γ1ϕ(J0(K)−J0(K∗1)−ε,σ)+γ2ϕ(−J0(K)+J0(K∗1)−ε,σ)≤12(√5+1)(γ1+γ2)σ. |
Based on Theorem 2, we have
0<G(K7)−G0(K6)=F2(K7)−F2(K6)+γ1(ϕ(J0(K)−J0(K∗1)−ε,σ)−max{J0(K)−J0(K∗1)−ε,0})+γ2(ϕ(−J0(K)+J0(K∗1)−ε,σ)−max{−J0(K)+J0(K∗1)−ε,0})≤(γ1+γ2)σ. |
Thus, we get
−12(√5+1)(γ1+γ2)σ≤F2(K7)−F2(K6)≤(γ1+γ2)σ. |
This completes the proof of Theorem 3.
Remark 4. If γ1 and γ2 are greater than the threshold value γ∗, the optimal solution of ProblemP6 is the exact optimal solution of ProblemP7 [31]. Furthermore, Theorem 3 provides an error estimation between the solutions of ProblemP5 and ProblemP7 as long as γ1>γ∗ and γ2>γ∗. Thus, the approximate optimal solution of ProblemP5 can be achieved by solving ProblemP7. Note that we have proved that the optimal solution of ProblemP5 is also the optimal solution of ProblemP4 as long as the adjustable smooth parameter σ is sufficiently small, and ProblemP4 and ProblemP3 are equivalent. Thus, the approximate optimal solution of ProblemP3 can be achieved by solving ProblemP7.
Note that the cost function G(K) in Problem P7 is continuously differentiable. Thus, ProblemP7 is an unconstrained optimal parameter selection problem, which can be solved efficiently by using any gradient-based algorithm. Since the functionals depend implicitly on the feedback matrix K via system (2.2), it is important to derive an effective computational procedure for calculating the gradient of the cost function G(K) in Problem P7.
In this subsection, we investigate the gradients of x(⋅|K) with respect to the feedback matrix. Define
Θ:=[−1−Kli,1−Kli]. |
Then, 0∈Θ and ϵ∈Θ⇔K+ϵEli∈Θ. For each ϵ∈Θ, define
φϵ(t):=xϵ(t)−x(t),t≤T, |
and
θϵ:=xϵ(t−τ(F2(K)))−x(t−τ(F2(K))),t≤T. |
Clearly,
φϵ(t−τ(F2(K)))=θϵ(t),t≤T. |
Then we have the following lemmas.
Lemma 5. There exists a positive real number L1>0 such that for all ϵ∈Θ, we have
|xϵ(t)|≤L1,t∈[−τ,T]. |
Lemma 6. There exists a positive real number L2>0 such that for all ϵ∈Θ, we have
|φϵ(t)|≤L2|ϵ|,|θϵ+∂˜x∂τ∂τ(F2(K))∂F2∂F2(K)∂Kliϵ|≤L2|ϵ|,i∈Im,l∈In,t∈[0,T]. |
The partial derivatives of the system state with respect to the feedback matrix are given in Theorem 4.
Theorem 4. Let t∈(0,T] be a fixed time point. Then x(t|⋅) is differentiable with respect to Kli on [−1, 1]. Moreover, for each Kli, we have
∂x(t|K)∂Kli=Λli(t|K),t∈[0,T],i∈Im,l∈In, |
where Λli(⋅|K),i∈Im,l∈In, is the solution of the following auxiliary system:
˙Λli(t)=∂f(x(t),˜x(t),K)∂xΛli(t)+∂f(x(t),˜x(t),K)∂˜x[Λli(t−τ(F2(K)))+∂˜x∂τ∂τ(F2(K))∂F2∂F2(K)∂Kli]+∂f(x(t),˜x(t),K)∂Kli,Λli(t)=0,t≤0. |
Proof. Let Kli∈[−1,1],i∈Im,l∈In, be arbitrary but fixed. For each ϵ∈Θ, define the following functions:
ˉfϵ(s,η)=f(x(s)+ηφϵ,˜x(s)+ηθϵ(s),Kli+ηϵEli),η∈[0,1],Δϵ1=∫10{∂˜fϵ(s,η)∂x−∂˜fϵ(s,0)∂x}φϵ(s)dη,s∈[0,t],Δϵ2=∫10{∂˜fϵ(s,η)∂˜x−∂˜fϵ(s,0)∂˜x}[θϵ(s)+∂˜x(s)∂τ∂τ(F2(K))∂F2∂F2(K)∂Kliϵ]dη,s∈[0,t],Δϵ3=∫10ϵ{∂˜fϵ(s,η)∂Kli−∂˜fϵ(s,0)∂Kli}dη,s∈[0,t], |
where Eli denotes the matrix in which the elements in row i and column l are 1 and the rest are 0.
Based on Lemmas 5 and 6 and the continuous differentiability of the functions f, we can obtain that there exist two constants M1>0 and M2>0 such that
‖∂ˉfϵ(s,0)∂x‖≤M1,‖∂ˉfϵ(s,0)∂˜x‖≤M2, |
where ‖⋅‖ denotes the natural matrix norm on Rn×n. In addition, by Lemmas 5 and 6, the following limits exist uniformly with respect to η∈[0,1]:
limϵ→0{x(s)+ηφϵ(s)}=x(s), |
limϵ→0{˜x(s)+ηθϵ(s)}=˜x(s). |
Thus, for each δ>0, there exists an ϵ1>0 such that for all ϵ satisfying |ϵ|<ϵ1,
‖∂ˉfϵ(s,η)∂x−∂ˉfϵ(s,0)∂x‖<δ,η∈[0,1], |
‖∂ˉfϵ(s,η)∂˜x−∂ˉfϵ(s,0)∂˜x‖<δ,η∈[0,1], |
and
‖∂ˉfϵ(s,η)∂Kli−∂ˉfϵ(s,0)∂Kli‖<δ,η∈[0,1]. |
Thus, it follows from Lemma 6 that
|Δϵ1(s)|≤L2δ|ϵ|,|Δϵ2(s)|≤L2δ|ϵ|,|Δϵ3(s)|≤δ|ϵ|. | (3.3) |
Now, let δ>0 be arbitrary but fixed and choose ϵ∈Θ such that 0<|ϵ|<ϵ1. Then, by the chain rule, we have
∂ˉfϵ(s,η)∂η=∂ˉfϵ(s,η)∂x⋅∂x∂η+∂ˉfϵ(s,η)∂˜x[∂˜x∂η+∂˜x∂τ∂τ(F2(K))∂F2∂F2(K)∂Kli∂Kli∂η]+∂ˉfϵ(s,η)∂Kli⋅∂Kli∂η=∂ˉfϵ(s,η)∂x⋅φϵ(s)+∂ˉfϵ(s,η)∂˜x⋅[θϵ(s)+∂˜x∂τ∂τ(F2(K))∂F2∂F2(K)∂Kliϵ]+∂ˉfϵ(s,η)∂Kli⋅ϵ. |
Recall that t∈(0,T] is a fixed time point. Then, by the fundamental theorem of calculus, we get
φϵ(t)=xϵ(t)−x(t)=∫t0ˉfϵ(s,1)−ˉfϵ(s,0)ds=∫t0(∫10∂ˉfϵ(s,η)∂ηdη)ds. |
Thus, the chain rule follows:
φϵ(t)=∫t0(∫10∂ˉfϵ(s,η)∂x⋅φϵ(s)dη)ds+∫t0{∫10∂ˉfϵ(s,η)∂˜x[θϵ(s)+∂˜x∂τ∂τ(F2(K))∂F2∂F2(K)∂Kliϵ]dη}ds+∫t0(∫10∂ˉfϵ(s,η)∂Kli⋅ϵdη)ds. | (3.4) |
Note that we have
∫10∂ˉfϵ(s,η)∂xφϵ(s)dη=Δϵ1(s)+∂ˉfϵ(s,0)∂xφϵ(s), | (3.5) |
∫10∂ˉfϵ(s,η)∂˜xθϵ(s)dη=Δϵ2(s)+∂ˉfϵ(s,0)∂˜x[θϵ(s)+∂˜x∂τ∂τ(F2(K))∂F2∂F2(K)∂Kliϵ], | (3.6) |
∫10ϵ∂ˉfϵ(s,η)∂Klidη=Δϵ3(s)+ϵ∂ˉfϵ(s,0)∂Kli. | (3.7) |
Substituting (3.5)–(3.7) into (3.4) gives
φϵ(t)=∫t0(Δϵ1(s)+∂ˉfϵ(s,0)∂xφϵ(s))ds+∫t0{Δϵ2(s)+∂ˉfϵ(s,0)∂˜x[θϵ(s)+∂˜x∂τ∂τ(F2(K))∂F2∂F2(K)∂Kliϵ]}ds+∫t0(Δϵ3(s)+ϵ∂ˉfϵ(s,0)∂Kli)ds. | (3.8) |
Note that
Λli(t|K)=∫t0{∂f(x(s),˜x(s),K)∂xΛli(s)+∂f(x(s),˜x(s),K)∂˜x[Λli(s−τ(F2(K)))+∂˜x∂τ∂τ(F2(K))∂F2∂F2(K)∂Kli]+∂f(x(s),˜x(s),K)∂Kli}ds. | (3.9) |
Then multiplying (3.8) by ϵ−1, subtracting (3.9), taking the norm of both sides, and finally applying (3.3) yields
|ϵ−1φϵ(t)−Λli(t|K)|≤(L2+L2+1)δT+M1∫t0|ϵ−1φϵ(s)−Λli(s)|ds+M2∫t0|ϵ−1φϵ(s−τ(F2(K)))−Λli(s−τ(F2(K)))|ds. | (3.10) |
Also, we know that φϵ(t)=0,t≤0, Λli(s)=0,s≤0, and then the last integral term on the right-hand side of (3.10) can be simplified as follows:
∫t−τ−τM2|ϵ−1φϵ(s)−Λli(s|K)|ds≤∫t0M2|ϵ−1φϵ(s)−Λli(s|K)|ds. |
Thus, (3.10) becomes
|ϵ−1φϵ(s)−Λli(s|K)|≤(2L2+1)δT+(M1+M2)∫t0|ϵ−1φϵ(s)−Λli(s|K)|ds. |
By the Gronwall-Bellman lemma, it follows that
|ϵ−1φϵ(s)−Λli(s|K)|≤(2L2+1)δTexp[(M1+M2)T], |
which holds whenever 0<|ϵ|<ϵ1. Since δ is arbitrarily chosen, we conclude that ϵ−1φϵ(t)→Λli(t|K) as ϵ→0.
Then, we get
limϵ→0φϵ(t)ϵ=Λli(t|K). |
Because of
limϵ→0φϵ(t)ϵ=limϵ→0xϵ(t)−x(t)ϵ=Λli(t|K),i∈Im,l∈In, |
we obtain
∂x(t|K)∂Kli=Λli(t|K),i∈Im,l∈In, |
thereby completing the proof.
We now present the following algorithm for computing the cost function of ProblemP7 and its gradient at a given controller K.
Based on Theorem 4 and the chain rule, we have:
Theorem 5. The gradients of the cost function G(K) with respect to Kli,i∈Im,l∈In, is given by
∂G(K)∂Kli=n∑l=1{−2Kli+m∑i=1[2σ2(√(Kli)2+4σ2−Kli)2(2Kli√(Kli)2+4σ2−1)+2σ2(√(−Kli)2+4σ2+Kli)2(2Kli√(Kli)2+4σ2+1)]}+γ12σ2(√J0(K)−J0(K∗1)−ε+4σ2−J0(K)+J0(K∗1)+ε)2[∂J0(K)/∂Kli√J0(K)−J0(K∗1)−ε+4σ2−∂J0(K)∂Kli]+γ22σ2(√(−J0(K)+J0(K∗1)−ε)+4σ2+J0(K)−J0(K∗1)+ε)2[−∂J0(K)/∂Kli√J0(K)−J0(K∗1)−ε+4σ2+∂J0(K)∂Kli], |
where
∂J0(K)∂Kli=2(x(T|K))⊤SΛli(t|K)+∫T0[(x(t|K))⊤QΛli(t|K)+(u(t))⊤Wu(t)]dt,i∈Im,l∈In. |
Then, ProblemP7 can be solved efficiently by using any gradient-based algorithm based on Theorem 5.
In this section, all computational experiments are carried out in MATLAB R2021a on a computer with a 3.70 GHz Intel Core i9-10900K CPU243 and 32.0GB RAM. The Euler method is used to solve system (2.2) with a step size of 1/10, and the initial time and terminal time are 0 and 1, respectively. We consider the 10th-order.
We use the gradient-based methods described in [31] for solving Problem P1 to obtain the optimal dense feedback matrix denoted by K∗1. Then, through the application of Algorithm 1, we solve Problem P7 to obtain the optimal sparse feedback matrix K∗2.
To indicate the sparse level of the feedback matrix K, we define the following indicator:
r=NumberofnonzeroelementsinKNumberofelementsinK, |
which represents the proportion of nonzero elements in the matrix. Obviously, a smaller value of r means a better sparse level of K.
Based on empirical data, we consider a 10-th order LTI system with a state matrix
A=[−8−100−1−9−2−8−4−4−6−10−4−9−7−5−6−4−3−10−4−6−4−7−3−4−4−8−1−9−6−9−70−4−6−5−7−5−1−2−1−5−90−2−5−5−10−7−5−5−10−2−6−8−5−3−30−4−60−9−9−4−5−10−9−9−3−3−4−2−3−6−6−7−10−5−9−7−8−6−9−4−10−2−7−90−5−9−300−1−4−8−2]. |
Assume that B=R=Q=E10, S=0∈R10×10, and ε=0.1. The cost function is given by [30]:
J0(K)=∫10[(x(t|K))⊤x(t|K)+u(t)⊤u(t)]dt, |
where u=−Kx(t−τ(‖K‖0)).
Based on empirical data, we consider another 10th-order LTI system with a state matrix
A=[3−100−1−9−2−8−4−4−6−10−46−7−56−42−10−4−67−7−3−4−4−8−1−9−6−9−70−4−6−5−7−5−1−2−1−5−90−2−1−5−10−7−5−5−10−2−6−8−5−3−30−4−60−94−4−5−10−9−9−3−3−4−2−39−6−72−53−7−8−6−9−41−2−7−90−5−9−300−1−4−8−2]. |
Assume that B=R=Q=E10, S=0∈R10×10, ε=0.1. The cost function is given by [30]:
J0(K)=∫10[(x(t|K))⊤x(t|K)+u(t)⊤u(t)]dt, |
where u=−Kx(t−τ(‖K‖0)).
The distributions of nonzero components in the feedback matrices K∗1 and K∗2 and their corresponding state at each moment are displayed in Figures 3 and 4, where nz means the number of nonzero elements. Their corresponding optimal cost and sparsity levels are given in Tables 1 and 2. Figures 3 and 4 indicate that the feedback matrix K∗1 exhibits a high degree of density, while the feedback matrix K∗2 displays a notable level of sparsity. Furthermore, from Tables 1 and 2, it follows that the value of the cost function, specifically J0(K∗2), slightly exceeds that of J0(K∗1).
the optimal feedback matrix | the optimal cost function | r |
K∗1 | J0(K∗1)=2.59 | 0.88 |
K∗2 | J0(K∗2)=2.68 | 0.06 |
the optimal feedback matrix | the optimal cost function | r |
K∗1 | J0(K∗1)=2.41 | 0.93 |
K∗2 | J0(K∗2)=2.57 | 0.08 |
We can see that the number of zero components in K∗2 increases rapidly with only a small increase in cost. Therefore, we can conclude that Algorithm 1 proposed in this paper can produce a better quality solution which balances the system performance and sparsity.
Algorithm 1 The gradients calculation of the cost function in Problem P7 |
1: Step 1: Obtain x(t|K), Λli(t),i∈Im,l∈In, by solving the enlarged time-delay system consisting of the original system (2.2) and the auxiliary system in Theorem 4. |
2: Step 2: Use the state values x(t|K) to compute G(K) in ProblemP7. |
3: Step 3: Use x(t|K), G(K) and Λli(t) to compute ∂G(K)∂Kli,i∈Im,l∈In. |
In practical scenarios, network operators often choose sparse communication topologies to minimize costs, but the concurrent use of the network by multiple users often leads to feedback delays. Our goal is to determine the optimal sparse feedback control matrix K. To achieve this, we formulate a SOC problem tailored to the CPS with variable delays, with the aim of minimizing ||K||0 while adhering to a specified maximum compromise in system costs. We employ a penalty method to transform the SOC problem into one governed solely by box constraints. To handle the nonsmooth aspects of the resulting problem, we utilize a smoothing technique and subsequently analyze the errors it may introduce. The gradients of the objective function concerning the feedback control matrix are computed by simultaneously solving the state system and the associated variational system over time. An optimization algorithm is designed to tackle the resulting challenge, utilizing a piecewise quadratic approximation. The paper concludes with a discussion of the simulation results.
The innovation of the paper is stated as follows: (ⅰ) The innovative use of a penalty method transforms the problem into one constrained by box limits, simplifying the optimization process while still enforcing necessary control constraints; (ⅱ) the application of smoothing techniques to approximate nonsmooth components enables the derivation of gradients efficiently, facilitating the use of gradient-based optimization algorithms in sparse settings; (ⅲ) the method incorporates simultaneous solving of the state and variational systems, enhancing accuracy in gradient calculations and improving the overall performance of the control strategy; and (ⅳ) the development of an optimization algorithm based on piecewise quadratic approximation offers a computationally efficient way to navigate the optimization landscape, making the approach feasible for real-time applications.
The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.
This work was supported in part by the National Key Research and Development Program of China under Grant 2022YFB3304600, in part by the Nature Science Foundation of Liaoning Province of China under Grant 2024-MS-015, in part by the Fundamental Research Funds for the Central Universities under Grant 3132024196, and Grant DUT22LAB305, in part by the National Natural Science Foundation of China under Grant 12271307 and Grant 12161076, in part by the China Postdoctoral Science Foundation under Grant 2019M661073, and in part by the Xinghai Project of Dalian Maritime University.
The authors declare there is no conflicts of interest.
[1] |
M. Alowaidi, S. K. Sharma, A. AlEnizi, S. Bhardwaj, Integrating artificial intelligence in cyber security for cyber-physical systems, Electron. Res. Arch., 31 (2023), 1876–1896. http://doi.org/10.3934/era.2023097 doi: 10.3934/era.2023097
![]() |
[2] |
N. Negi, A. Chakrabortty, Sparsity-promoting optimal control of cyber-physical systems over shared communication networks, Automatica, 122 (2020), 109217. http://doi.org/10.1016/j.automatica.2020.109217 doi: 10.1016/j.automatica.2020.109217
![]() |
[3] |
M. N. Al-Mhiqani, T. Alsboui, T. Al-Shehari, K. H. Abdulkareem, R. Ahmad, M. A. Mohammed, Insider threat detection in cyber-physical systems: a systematic literature review, Comput. Electr. Eng., 119 (2024), 109489. http://doi.org/10.1016/j.compeleceng.2024.109489 doi: 10.1016/j.compeleceng.2024.109489
![]() |
[4] |
S. Liu, X. Wang, B. Niu, X. Song, H. Wang, X. Zhao, Adaptive resilient output feedback control against unknown deception attacks for nonlinear cyber-physical systems, IEEE Trans. Circuits Syst. II: Express Briefs, 71 (2024). 3855–3859. http://doi.org/10.1109/TCSII.2024.3372413 doi: 10.1109/TCSII.2024.3372413
![]() |
[5] |
M. Zhao, W. Qin, J. Yang, G. Lu, Security control for cyber-physical systems under aperiodic denial-of-service attacks: A memory-event-triggered active approach, Neurocomputing, 600 (2024), 128159. http://doi.org/10.1016/j.neucom.2024.128159 doi: 10.1016/j.neucom.2024.128159
![]() |
[6] |
P. E. G. Silva, P. H. J. Nardelli, A. S. de Sena, H. Siljak, N. Nevaranta, N. Marchetti, et al., Semantic-functional communications in cyber-physical systems, IEEE Network, 38 (2024), 241–249. http://doi.org/10.1109/MNET.2023.3329192 doi: 10.1109/MNET.2023.3329192
![]() |
[7] |
L. M. Castiglione, E. C. Lupu, Which attacks lead to hazards combining safety and security analysis for cyber-physical systems, IEEE Trans. Dependable Secure Comput., 21 (2024), 2526–2540. http://doi.org/10.1109/TDSC.2023.3309778 doi: 10.1109/TDSC.2023.3309778
![]() |
[8] |
S. Das, P. Dey, D. Chatterjee, Almost sure detection of the presence of malicious components in cyber-physical systems, Automatica, 167 (2024), 111789. http://doi.org/10.1016/j.automatica.2024.111789 doi: 10.1016/j.automatica.2024.111789
![]() |
[9] |
C. Fioravanti, S. Panzieri, G. Oliva, Negativizability: a useful property for distributed state estimation and control in cyber-physical systems, Automatica, 157 (2023), 111240. http://doi.org/10.1016/j.automatica.2023.111240 doi: 10.1016/j.automatica.2023.111240
![]() |
[10] |
H. N. AlEisa, F. Alrowais, R. Allafi, N. S. Almalki, R. Faqih, R. Marzouk, et al., Transforming transportation: safe and secure vehicular communication and anomaly detection with intelligent cyber-physical system and deep learning, IEEE Trans. Consum. Electron., 70 (2024), 1736–1746. http://doi.org/10.1109/TCE.2023.3325827 doi: 10.1109/TCE.2023.3325827
![]() |
[11] |
L. Khoshnevisan, X. Liu, Resilient neural network-based control of nonlinear heterogeneous multi-agent systems: a cyber-physical system approach, Nonlinear Dyn., 111 (2023), 19171–19185. http://doi.org/10.1007/s11071-023-08840-w doi: 10.1007/s11071-023-08840-w
![]() |
[12] |
L. Chen, Y. Li, S. Tong, Robust adaptive control for nonlinear cyber‐physical systems with FDI attacks via attack estimation, Int. J. Robust Nonlinear Control, 33 (2023), 9299–9316. http://doi.org/10.1002/rnc.6851 doi: 10.1002/rnc.6851
![]() |
[13] |
G. Routray, R. M. Hegde, Sparsity-driven loudspeaker gain optimization for sound field reconstruction with spherical microphone array, Digital Signal Process., 154 (2024), 104688. http://doi.org/10.1016/j.dsp.2024.104688 doi: 10.1016/j.dsp.2024.104688
![]() |
[14] |
K. Zhou, Y. Wang, B. Qiao, J. Liu, M. Liu, Z. Yang, et al., Non-convex sparse regularization via convex optimization for blade tip timing, Mech. Syst. Signal Process., 222 (2025), 111764. http://doi.org/10.1016/j.ymssp.2024.111764 doi: 10.1016/j.ymssp.2024.111764
![]() |
[15] |
T. Zhang, F. Peng, X. Tang, R. Yan, R. Deng, S. Zhao, A sparse knowledge embedded configuration optimization method for robotic machining system toward improving machining quality, Rob. Comput-Integr. Manuf., 90 (2024), 102818. http://doi.org/10.1016/j.rcim.2024.102818 doi: 10.1016/j.rcim.2024.102818
![]() |
[16] |
J. Yuan, C. Wu, K. L. Teo, J. Xie, S. Wang, Computational method for feedback perimeter control of multiregion urban traffic networks with state-dependent delays, Transp. Res. Part C: Emerging Technol., 153 (2023), 104231. https://doi.org/10.1016/j.trc.2023.104231 doi: 10.1016/j.trc.2023.104231
![]() |
[17] |
J. Yuan, C. Wu, K. L. Teo, S. Zhao, L. Meng, Perimeter control with state-dependent delays: optimal control model and computational method, IEEE Trans. Intell. Transp. Syst., 23 (2022), 20614–20627. https://doi.org/10.1109/TITS.2022.3179729 doi: 10.1109/TITS.2022.3179729
![]() |
[18] | C. Zhao, Z. Luo, N. Xiu, Some advances in theory and algorithms for sparse optimization, Oper. Res. Trans., 24 (2020), 1–24. |
[19] |
S. Dai, Variable selection in convex quantile regression: L1-norm or L0-norm regularization?, Eur. J. Oper. Res., 305 (2023), 338–355. http://doi.org/10.1016/j.ejor.2022.05.041 doi: 10.1016/j.ejor.2022.05.041
![]() |
[20] |
B. Hong, H. Qian, Z. Wang, Iterative hard thresholding algorithm-based detector for compressed OFDM-IM systems, IEEE Commun. Lett., 26 (2022), 2205–2209. http://doi.org/10.1109/LCOMM.2022.3187451 doi: 10.1109/LCOMM.2022.3187451
![]() |
[21] |
H. Liu, T. Wang, Z. Liu, Some modified fast iterative shrinkage thresholding algorithms with a new adaptive non-monotone stepsize strategy for nonsmooth and convex minimization problems, Comput. Optim. Appl., 83 (2022), 651–691. http://doi.org/10.1007/s10589-022-00396-6 doi: 10.1007/s10589-022-00396-6
![]() |
[22] |
B. Liu, K. Gong, L. Zhang, Convergence analysis of the augmented Lagrangian method for lp-norm cone optimization problems with p≥2, Numer. Algorithms, 2024 (2024). http://doi.org/10.1007/s11075-024-01912-x doi: 10.1007/s11075-024-01912-x
![]() |
[23] |
D. Han, D. Sun, L. Zhang, Linear rate convergence of the alternating direction method of multipliers for convex composite programming, Math. Oper. Res., 43 (2018), 622–637. https://doi.org/10.1287/moor.2017.0875 doi: 10.1287/moor.2017.0875
![]() |
[24] |
E. Hernández, P. Merino, Sparse optimal control of timoshenko's beam using a locking-free finite element approximation, Optim. Control Appl. Methods, 45 (2024), 1007–1029. https://doi.org/10.1002/oca.3085 doi: 10.1002/oca.3085
![]() |
[25] |
K. Ito, T. Ikeda, K. Kashima, Sparse optimal stochastic control, Automatica, 125 (2021), 109438. http://doi.org/10.1016/j.automatica.2020.109438 doi: 10.1016/j.automatica.2020.109438
![]() |
[26] |
F. Lejarza, M. Baldea, Economic model predictive control for robust optimal operation of sparse storage networks, Automatica, 125 (2021), 109346. http://doi.org/10.1016/j.automatica.2020.109346 doi: 10.1016/j.automatica.2020.109346
![]() |
[27] |
M. Liu, H. Zhu, F. Zhang, J. Wang, C. Zhou, Y. Lv, Model-based sparse optimal control of the hydrogen sulfide synthesis process for acidic wastewater sulfidation, J. Water Process Eng., 65 (2024), 105836. http://doi.org/10.1016/j.jwpe.2024.105836 doi: 10.1016/j.jwpe.2024.105836
![]() |
[28] |
J. Yuan, C. Wu, Z. Liu, S. Zhao, C. Yu, K. L. Teo, et al., Koopman modeling for optimal control of the perimeter of multi-region urban traffic networks, Appl. Math. Modell., 138 (2025), 115742. https://doi.org/10.1016/j.apm.2024.115742 doi: 10.1016/j.apm.2024.115742
![]() |
[29] |
Q. Li, Y. Bai, C. Yu, Y. Yuan, A new piecewise quadratic approximation approach for L0 norm minimization problem, Sci. China Math. 62 (2019), 185–204. http://doi.org/10.1007/s11425-017-9315-9 doi: 10.1007/s11425-017-9315-9
![]() |
[30] |
A. Modi, M. K. S. Faradonbeh, A. Tewari, G. Michailidis, Joint learning of linear time-invariant dynamical systems, Automatica, 164 (2024), 111635. http://doi.org/10.1016/j.automatica.2024.111635 doi: 10.1016/j.automatica.2024.111635
![]() |
[31] | K. L. Teo, B. Li, C. Yu, V. Rehbock, Applied and Computational Optimal Control: a Control Parametrization Approach, Springer, Cham, 2021. http://doi.org/10.1007/978-3-030-69913-0 |
[32] |
X. Wu, K. Zhang, M. Chen, A gradient-based algorithm for non-smooth constrained optimization problems governed by discrete-time nonlinear equations with application to long-term hydrothermal optimal scheduling control, J. Comput. Appl. Math., 412 (2022), 114335. https://doi.org/10.1016/j.cam.2022.114335 doi: 10.1016/j.cam.2022.114335
![]() |
the optimal feedback matrix | the optimal cost function | r |
K∗1 | J0(K∗1)=2.59 | 0.88 |
K∗2 | J0(K∗2)=2.68 | 0.06 |
the optimal feedback matrix | the optimal cost function | r |
K∗1 | J0(K∗1)=2.41 | 0.93 |
K∗2 | J0(K∗2)=2.57 | 0.08 |