Distributed optimization with closed convex set for multi-agent networks over directed graphs

In this paper, a distributed projection algorithm based on the subgradient method is presented to solve the distributed optimization problem with a constrained set over a directed multi-agent network, where the designed protocol is scaled by the left eigenvector associated with the weighted adjacency matrix. By using the property of the projection operation and nonnegative almost supermartingales, we give the convergence analysis of our algorithm and show that the optimal solution is the ultimate consensus state of all agents to be reached. A numerical simulation for a specific optimization problem is given to verify the effectiveness of our algorithm.     

Designing distributed consensus protocols for second-order nonlinear multi-agents with unknown control directions under directed graphs

This paper focuses on the leaderless and leader-following consensus problems of second-order nonlinear multi-agents under directed graphs. Both leaderless and leader-following consensus protocols are proposed for multi-agents with unknown control directions based on the Nussbaum-type gains. For the leaderless case, the proposed protocol can guarantee that the consensus errors asymptotically converge to zero. Moreover, for the leader-following case, the Lyapunov stability analysis shows that the consensus tracking errors can be made arbitrarily small by tuning the control parameters. It should also be noted that these proposed protocols do not require any information about the global communication topology and work with only the relative information of neighboring agents. Illustrative examples are given to show the effectiveness of the proposed control protocols.     

Consensus control for multi-agent systems with distributed parameter models via iterative learning algorithm

This paper deals with the problem of iterative learning control algorithm for a class of multi-agent systems with distributed parameter models. And the considered distributed parameter models are governed by the parabolic or hyperbolic partial differential equations. Based on the framework of network topologies, a consensus-based iterative learning control protocol is proposed by using the nearest neighbor knowledge. When the iterative learning control law is applied to the systems, the consensus errors between any two agents on L² space are bounded, and furthermore, the consensus errors on L² space can converge to zero as the iteration index tends to infinity in the absence of initial errors. Simulation examples illustrate the effectiveness of the proposed method.     

Primal-dual stochastic distributed algorithm for constrained convex optimization

This paper investigates distributed convex optimization problems over an undirected and connected network, where each node’s variable lies in a private constrained convex set, and overall nodes aim at collectively minimizing the sum of all local objective functions. Motivated by a variety of applications in machine learning problems with large-scale training sets distributed to multiple autonomous nodes, each local objective function is further designed as the average of moderate number of local instantaneous functions. Each local objective function and constrained set cannot be shared with others. A primal-dual stochastic algorithm is presented to address the distributed convex optimization problems, where each node updates its state by resorting to unbiased stochastic averaging gradients and projects on its private constrained set. At each iteration, for each node the gradient of one local instantaneous function selected randomly is evaluated and the average of the most recent stochastic gradients is used to approximate the true local gradient. In the constrained case, we show that with strong-convexity of the local instantaneous function and Lipschitz continuity of its gradient, the algorithm converges to the global optimization solution almost surely. In the unconstrained case, an explicit linear convergence rate of the algorithm is provided. Numerical experiments are presented to demonstrate correctness of the theoretical results.     

Backstepping-based adaptive dynamic programming for missile-target guidance systems with state and input constraints

In this paper, a novel backstepping-based adaptive dynamic programming (ADP) method is developed to solve the problem of intercepting a maneuver target in the presence of full-state and input constraints. To address state constraints, a barrier Lyapunov function is introduced to every backstepping procedure. An auxiliary design system is employed to compensate the input constraints. Then, an adaptive backstepping feedforward control strategy is designed, by which the tracking problem for strict-feedback systems can be reduced to an equivalence optimal regulation problem for affine nonlinear systems. Secondly, an adaptive optimal controller is developed by using ADP technique, in which a critic network is constructed to approximate the solution of the associated Hamilton–Jacobi–Bellman (HJB) equation. Therefore, the whole control scheme consists of an adaptive feedforward controller and an optimal feedback controller. By utilizing Lyapunov's direct method, all signals in the closed-loop system are guaranteed to be uniformly ultimately bounded (UUB). Finally, the effectiveness of the proposed strategy is demonstrated by using a simple nonlinear system and a nonlinear two-dimensional missile-target interception system.     

An adaptive penalty-like continuous-time algorithm to constrained distributed convex optimization

This paper considers a nonsmooth constrained distributed convex optimization over multi-agent systems. Each agent in the multi-agent system only has access to the information of its objective function and constraint, and cooperatively minimizes the global objective function, which is composed of the sum of local objective functions. A novel continuous-time algorithm is proposed to solve the distributed optimization problem and effectively characterize the appropriate gain of the penalty function. It should be noted that the proposed algorithm is based on an adaptive strategy to avoid introducing the primal-dual variables and estimating the related exact penalty parameters. Additional, it is demonstrated that the state solution of the proposed algorithm achieves consensus and converges to an optimal solution of the optimization problem. Finally, numerical simulations are given and the proposed algorithm is applied to solve the optimal placement problem and energy consumption problem.     

An accelerated exact distributed first-order algorithm for optimization over directed networks

Distributed optimization over networked agents has emerged as an advanced paradigm to address large-scale control, optimization, and signal-processing problems. In the last few years, the distributed first-order gradient methods have witnessed significant progress and enrichment due to the simplicity of using only the first derivatives of local functions. An exact first-order algorithm is developed in this work for distributed optimization over general directed networks with only row-stochastic weighted matrices. It employs the rescaling gradient method to address unbalanced information diffusion among agents, where the weights on the received information can be arbitrarily assigned. Moreover, uncoordinated step-sizes are employed to magnify the autonomy of agents, and an error compensation term and a heavy-ball momentum are incorporated to accelerate convergency. A linear convergence rate is rigorously proven for strongly-convex objective functions with Lipschitz continuous gradients. Explicit upper bounds of step-size and momentum parameter are provided. Finally, simulations illustrate the performance of the proposed algorithm.     

Neural network adaptive control design for robot manipulators under velocity constraints

This paper studies the neural adaptive control design for robotic systems with uncertain dynamics under the existence of velocity constraints and input saturation. The control objective is achieved by choosing a control Lyapunov function using joint error variables that are restricted to linear growth and furthermore by introducing a secant type barrier Lyapunov function for constraining the joint rate variables. The former is exploited to bind the forward propagation of the position errors, and the latter is utilized to impose hard bounds on the velocity. Effective input saturation is expressed, and neural networks are employed to tackle the uncertainty problem in the system dynamics. Feasibility conditions are formulated, and the optimal design parameters are obtained by solving the constrained optimization problem. We prove that under the proposed method, semi-global uniform ultimate boundedness of the closed-loop system can be guaranteed. Tracking errors meanwhile converge to small neighborhoods of the origin, and violations of predefined velocity constraints are avoided. Finally, numerical simulations are performed to verify the effectiveness of the theoretical developments.     

Disturbance observer-based robust missile autopilot design with full-state constraints via adaptive dynamic programming

This paper aims to develop a robust optimal control method for longitudinal dynamics of missile systems with full-state constraints suffering from mismatched disturbances by using adaptive dynamic programming (ADP) technique. First, the constrained states are mapped by smooth functions, thus, the considered systems become nonlinear systems without state constraints subject to unknown approximation error. In order to estimate the unknown disturbances, a nonlinear disturbance observer (NDO) is designed. Based on the output of disturbance observer, an integral sliding mode controller (ISMC) is derived to counteract the effects of disturbances and unknown approximation error, thus ensuring the stability of nonlinear systems. Subsequently, the ADP technique is utilized to learn an adaptive optimal controller for the nominal systems, in which a critic network is constructed with a novel weight update law. By utilizing the Lyapunov's method, the stability of the closed-loop system and the convergence of the estimation weight for critic network are guaranteed. Finally, the feasibility and effectiveness of the proposed controller are demonstrated by using longitudinal dynamics of a missile.     

Barrier Lyapunov function-based decentralized adaptive neural control for uncertain high-order stochastic nonlinear interconnected systems with output constraints

Decentralized adaptive neural backstepping control scheme is developed for uncertain high-order stochastic nonlinear systems with unknown interconnected nonlinearity and output constraints. For the control of high-order nonlinear interconnected systems, it is assumed that nonlinear system functions are unknown. It is for the first time to control stochastic nonlinear high-order systems with output constraints. Firstly, by constructing barrier Lyapunov functions, output constraints are handled. Secondly, at each recursive step, only one adaptive parameter is updated to overcome over-parameterization problems, and RBF neural networks are used to identify unknown nonlinear functions so that the difficulties caused by completely unknown system functions and stochastic disturbances are tackled. Finally, based on the Lyapunov stability method, the decentralized adaptive control scheme via neural networks approximator is proposed, ultimately reducing the number of learning parameters. It is shown that the designed controller can guarantee all the signals of the resulting closed-loop system to be semi-globally uniformly ultimately bounded (SGUUB), and the tracking errors for each subsystem are driven to a small neighborhood of zero. The simulation studies are performed to verify the effectiveness of the proposed control strategy.     

Distributed consensus-based multi-agent convex optimization via gradient tracking technique

This paper considers solving a class of optimization problems over a network of agents, in which the cost function is expressed as the sum of individual objectives of the agents. The underlying communication graph is assumed to be undirected and connected. A distributed algorithm in which agents employ time-varying and heterogeneous step-sizes is proposed by combining consensus of multi-agent systems with gradient tracking technique. The algorithm not only drives the agents’ iterates to a global and consensual minimizer but also finds the optimal value of the cost function. When the individual objectives are convex and smooth, we prove that the algorithm converges at a rate of $O(1/\sqrt{t})$ if the homogeneous step-size does not exceed some upper bound, and it accelerates to O(1/t) if the homogeneous step-size is sufficiently small. When at least one of the individual objectives is strongly convex and all are smooth, we prove that the algorithm converges at a linear rate of O(λ^t) with 0?<?λ?<?1 even though the step-sizes are time-varying and heterogeneous. Two numerical examples are provided to demonstrate the efficiency of the proposed algorithm and to validate the theoretical findings.     

Decentralized adaptive neural control for interconnected stochastic nonlinear delay-time systems with asymmetric saturation actuators and output constraints

This paper investigates the problem of decentralized adaptive backstepping control for a class of large-scale stochastic nonlinear time-delay systems with asymmetric saturation actuators and output constraints. Firstly, the Gaussian error function is employed to represent a continuous differentiable asymmetric saturation nonlinearity, and barrier Lyapunov functions are designed to ensure that the output parameters are restricted. Secondly, the appropriate Lyapunov–Krasovskii functional and the property of hyperbolic tangent functions are used to deal with the unknown unmatched time-delay interactions, and the neural networks are employed to approximate the unknown nonlinearities. At last, based on Lyapunov stability theory, a decentralized adaptive neural control method is proposed, and the designed controller decreases the number of learning parameters. It is shown that the designed controller can ensure that all the closed-loop signals are 4-Moment (or 2 Moment) semi-globally uniformly ultimately bounded (SGUUB) and the tracking error converges to a small neighborhood of the origin. Two examples are provided to show the effectiveness of the proposed method.     

Experimental validation of a continuous-time MCSI algorithm with bounded adaptive gains

Model reference adaptive control algorithms with minimal controller synthesis have proven to be an effective solution to tame the behaviour of linear systems subject to unknown or time-varying parameters, unmodelled dynamics and disturbances. However, a major drawback of the technique is that the adaptive control gains might exhibit an unbounded behaviour when facing bounded disturbances. Recently, a minimal controller synthesis algorithm with an integral part and either parameter projection or σ-modification strategies was proposed to guarantee boundedness of the adaptive gains. In this article, these controllers are experimentally validated for the first time by using an electro-mechanical system subject to significant rapidly varying disturbances and parametric uncertainty. Experimental results confirm the effectiveness of the modified minimal controller synthesis methods to keep the adaptive control gains bounded while providing, at the same time, tracking performances similar to that of the original algorithm.     

Two-stage continuous-time triggered algorithms for constrained distributed optimization over directed graphs

This paper proposes two-stage continuous-time triggered algorithms for solving distributed optimization problems with inequality constraints over directed graphs. The inequality constraints are penalized by adopting log-barrier penalty method. The first stage of the proposed algorithms is capable of finding the optimal point of each local optimization problem in finite time. In the second stage of the proposed algorithms, zero-gradient-sum algorithms with time-triggered and event-triggered communication strategies are considered in order to reduce communication costs. Then, with the help of LaSalle’s invariance principle, it is proved that the state solution of each agent reaches consensus at the optimal point of the considered penalty distributed optimization problem, and Zeno behavior is also excluded. Finally, numerical examples are given to illustrate the effectiveness of the proposed algorithms.     

Adaptive distributed convex optimization for multi-agent and its application in flocking behavior

This paper studies adaptive optimization problem of continuous-time multi-agent systems. Multi-agents with second-order dynamics are considered. Each agent is equipped with a time-varying cost function which is known only to an individual agent. The objective is to make multi-agents velocities minimize the sum of local functions by local interaction. First, a distributed adaptive algorithm is presented, in which each agent depends only on its own velocity and neighbors velocities. It is indicated that all agents can track the optimal velocity. Then we apply the distributed adaptive algorithm to flocking of multi-agents. It is proved that all agents can track the optimal trajectory. The agents will maintain connectivity and avoid the inter-agent collision. Finally, two simulations are included to illustrate the results.     

Online adaptive policy iteration based fault-tolerant control algorithm for continuous-time nonlinear tracking systems with actuator failures

In this paper, a novel tracking control scheme for continuous-time nonlinear affine systems with actuator faults is proposed by using a policy iteration (PI) based adaptive control algorithm. According to the controlled system and desired reference trajectory, a novel augmented tracking system is constructed and the tracking control problem is converted to the stabilizing issue of the corresponding error dynamic system. PI algorithm, generally used in optimal control and intelligence technique fields, is an important reinforcement learning method to solve the performance function by critic neural network (NN) approximation, which satisfies the Lyapunov equation. For the augmented tracking error system with actuator faults, an online PI based fault-tolerant control law is proposed, where a new tuning law of the adaptive parameter is designed to tolerate four common kinds of actuator faults. The stability of the tracking error dynamic with actuator faults is guaranteed by using Lyapunov theory, and the tracking errors satisfy uniformly bounded as the adaptive parameters get converged. Finally, the designed fault-tolerant feedback control algorithm for nonlinear tracking system with actuator faults is applied in two cases to track the desired reference trajectory, and the simulation results demonstrate the effectiveness and applicability of the proposed method.     

P-type iterative learning control algorithm for a class of linear singular impulsive systems

This paper is focused on the iterative learning control problem for linear singular impulsive systems. For the purpose of tracking the desired output trajectory, a P-type iterative learning control algorithm is investigated for such system. Based on the fundamental property of singular impulsive systems and the restricted equivalent transformation theory of singular systems, the convergence conditions of the tracking errors for the system are obtained in the sense of λ norm. Finally, the validation of the algorithm is confirmed by a numerical example.     

Parameter identification of time-delayed nonlinear systems: An integrated method with adaptive noise correction

This paper proposes a novel method called the adaptive-noise-correction integrated parameter identification (ANCPI) for time-delayed nonlinear systems. Compared with the existing de-noising techniques, the significance of the proposed method is the use of the system itself to correct the noise-polluted components so that the accuracy of parameter identification is enhanced. To achieve the goal of adaptive noise correction, this study starts from the case of periodic response and then parameterizes the noise correction as the coefficient correction of harmonic basis. In this way, the parameter identification integrated with noise correction can be performed as the parameter optimization of the error function. For the convenience of application, a user-friendly program package is further provided and a detailed tutorial is presented in the supplementary material.     

Robust distributed tracking control for linear multi-agent systems based on distributed intermediate estimator

This paper considers the distributed tracking control problem for linear multi-agent systems with disturbances and a leader whose control input is nonzero and not available to any follower. Based on the relative output measurements of neighboring agents, a novel distributed observer-based tracking protocol is proposed, where the distributed intermediate estimators are constructed to estimate the leader’s unknown control input and the states of the tracking error system simultaneously, then a distributed tracking protocol is designed based on the derived estimates. It is proved that the states of the tracking error system are uniformly ultimately bounded and an explicit tracking error bound is obtained. A simulation example of aircrafts verifies the effectiveness of the proposed method.     

Bipartite output consensus for heterogeneous linear multi-agent systems with fully distributed protocol

This paper addresses the problem of bipartite output consensus of heterogeneous multi-agent systems over signed graphs. First, under the assumption that the sub-graph describing the communication topology among the agents is connected, a fully distributed protocol is provided to make the heterogeneous agents achieve bipartite output consensus. Then for the case that the topology graph has a directed spanning tree, a novel adaptive consensus protocol is designed, which also avoids using any global information. Each of these two protocols consists of a solution pair of the regulation equation and a homogeneous compensator. Numerical simulations show the effectiveness of the proposed approach.