ABSTRACT This paper considers the problem of predefined time event‐triggered control for high‐order multi‐input and multi‐output (MIMO) nonlinear systems. Neural networks (NNs) are employed to approximate the unknown nonlinear functions. A filter with predefined time stability is constructed to reduce the computational complexity. To reduce the controller's execution time and conserve communication resources, an event‐triggered mechanism (ETM) with a relative threshold strategy is constructed. Based on the backstepping recursive framework, a neural network adaptive predefined time event‐triggered control scheme is proposed. Using Lyapunov stability theory, it is proved that both the tracking errors and the filtering errors in the closed‐loop system can reach a small neighborhood around zero within the predefined time. Finally, a simulation example is provided to validate the stability of the system and the convergence of the tracking errors.
ABSTRACT This paper addresses the control challenges for discrete‐time singularly perturbed Markov jump cyber‐physical systems (CPSs) under deception attacks, actuator failures, and limited network bandwidth. A novel asynchronous sliding mode control (SMC) strategy is proposed, and it integrates finite‐time convergence with actuator fault tolerance. First, a novel probabilistic dynamic event‐triggered mechanism is designed. By using probabilistic decision‐making and dynamic auxiliary variables, it overcomes the conservatism of traditional triggers in stochastic networks and significantly reduces communication overhead. Second, deception attacks and actuator failures are considered at the same time. Deception attacks are described by Bernoulli distributions, and actuator failures are modeled by multiplicative uncertainties. This makes the system model closer to real working conditions. Third, an asynchronous control framework is established, and the co‐design of controller gains and event‐triggered weighting matrices is achieved. Controller modes and system modes are not synchronized, and they are linked by a conditional probability matrix. Moreover, sufficient conditions for singular perturbation finite‐time boundedness (SPFTB) with performance are given for both the reaching phase and the sliding motion phase. These conditions are presented as linear matrix inequalities (LMIs) and can be solved directly by MATLAB. Finally, numerical simulations on petroleum catalytic cracking processes verify the effectiveness and practicality of the proposed method.
ABSTRACT Existing adaptive Kalman filters for unknown and time‐varying noise statistics often exhibit numerical instability arising from the coupling between state and noise estimation. To overcome these issues, this paper proposes a Stable Adaptive Kalman Filter with Decay‐Factor Recursive Covariance Estimation (SAKF‐DFRCE). The proposed framework presents a non‐augmented, decoupled estimation strategy by constructing noise‐sensitive measurement difference sequences. This allows for the estimation of process and measurement noise covariance matrices via a linear least‐squares formulation, independent of the state estimation loop. A recursive update mechanism incorporating a dynamic decay factor facilitates robust adaptation to non‐stationary noise typical of maneuvering targets. Theoretical analysis establishes the strong consistency of the noise estimators and the asymptotic stability of the filter. The proposed SAKF‐DFRCE achieves estimation accuracy comparable to the theoretical benchmark of a standard Kalman filter while exhibiting reduced computational runtime compared to existing adaptive methods, making it suitable for high‐reliability, resource‐constrained aerospace applications. Monte Carlo simulations, including nonlinear radar tracking scenarios, validate the effectiveness of the proposed method.
ABSTRACT With the large‐scale integration of wind power, its inherent intermittency and uncertainty pose significant challenges to power system operation. Existing probabilistic forecasting methods often struggle with capturing spatiotemporal correlations among multiple turbines as well as generalizing to unseen turbines or future time steps. To address these issues, we propose a novel multi‐turbine collaborative probabilistic forecasting framework based on the flow matching enhanced neural process. By formulating wind power forecasting as a conditional stochastic process, our method naturally enables cross‐turbine knowledge sharing through a Transformer‐based encoder. The introduction of flow matching allows efficient single‐ or few‐step sampling while avoiding the usage of variational lower bounds, and hence is more effective. Experiments on a real‐world dataset comprising 134 turbines over 245 days demonstrate that the proposed framework achieves superior performance, significantly outperforming existing neural process variants and deep learning baselines in terms of RMSE, MAE, and continuous ranked probability score.
ABSTRACT Most existing H∞ adaptive methods cannot address non‐zero mean drift and bounded covariance at the same time. Conventional maximum likelihood schemes also tend to violate the γ‐robustness constraint. To overcome these issues, this paper proposes bias‐covariance joint H‐infinity filtering (BCJHF) algorithms. Bias‐covariance joint estimators are first introduced to allow for online estimation of both the mean and covariance. These estimators are then incorporated into the H∞ gain design to establish the BCJHF framework. The system is rigorously proven to maintain the γ‐suboptimal performance bound, preserving robustness during adaptive estimation. To further validate the proposed algorithms, the univariate nonstationary growth model is used as a simulation case.
ABSTRACT This paper investigates the optimal control problem of quadrotor unmanned aerial vehicles (UAVs) under uncertain dynamics and external environmental disturbances. First, the quadrotor UAV model with disturbances is considered, and a transformation function incorporating the prescribed time and accuracy via an auxiliary function is constructed to reformulate the problem as an optimal control problem. Subsequently, the control input and external disturbance are modeled as adversarial agents, and their interaction is analyzed using zero‐sum game theory, which results in the derivation of a modified Hamilton–Jacobi–Isaacs (HJI) equation. To implement the derived control policy, an online actor‐critic reinforcement learning algorithm is employed to efficiently approximate the solution of the otherwise intractable HJI equation in real time. Finally, numerical simulations are conducted to verify that prescribed‐time convergence, fast response, and robust flight performance under strong environmental disturbances are achieved by the proposed method.
ABSTRACT This study addresses the tracking control issue for unknown nonlinear systems, focusing on predefined‐time prescribed performance. The core contribution lies in establishing a unified control framework that integrates deferred prescribed performance with predefined‐time stability, yielding a dual‐time guarantee that transforms the convergence process from asymptotic behavior to a scheduled task with a user‐defined time. The deferred prescribed performance method ensures that the system error converges from any initial state, while the predefined‐time controller guarantees rigorous convergence within the prescribed duration. By employing a smooth shifting function and dynamic scaling, the constrained issue is reformulated into an unconstrained form, thus removing the feasibility condition on initial state restrictions. It further compensates for system uncertainties through neural network adaptation. The Lyapunov‐based analysis verifies the boundedness of all closed‐loop signals, predefined‐time convergence of the tracking error, and preservation of the prescribed transient and steady‐state performance. Compared to existing methods, the proposed strategy significantly improves practicality and robustness while maintaining guaranteed convergence speed and control accuracy. Finally, simulation studies are carried out to validate the effectiveness of the proposed control strategy.
ABSTRACT This paper addresses output‐tracking control for complex urban water distribution networks with multiple time‐varying delays and develops a feedback synthesis framework based on linear matrix inequalities (LMI). Guided by the network's interconnection structure, the physical system is abstracted as a discrete‐time multi‐delay model. An augmented Lyapunov–Krasovskii functional with free‐weighting matrices is constructed to derive LMI sufficient conditions that guarantee asymptotic stability and provide upper bounds on the mixed tracking performance. These conditions in turn enable a convex controller‐synthesis procedure. Numerical simulations demonstrate fast, robust, and energy‐efficient reference tracking in the presence of multiple delays and exogenous disturbances. The results furnish a computable, implementation‐ready design tool for pressure regulation and operational scheduling in urban water‐supply systems, with strong prospects for engineering deployment.
This paper investigates the adaptive dynamics learning and control problem for a class of nonlinear strict-feedback systems subject to output constraints. First, an output constraint transformation framework is introduced to equivalently convert the constrained system into an unconstrained form. Based on this framework, an adaptive fixed-time filtered learning control strategy is developed, which guarantees closed-loop stability while strictly enforcing the prescribed output constraints. Next, under the persistent excitation condition, a radial basis function neural network (RBFNN) is employed to accurately approximate the unknown nonlinear dynamics. Furthermore, the learned dynamic information is incorporated into the control law design to construct a learning-based controller, thereby effectively improving the transient performance and robustness of the system. Rigorous Lyapunov analysis is conducted to prove the uniform boundedness of all closed-loop signals and the strict satisfaction of the output constraints. Finally, simulation results are presented to demonstrate the effectiveness and superiority of the proposed adaptive constrained dynamics learning and control approach.
The paper solves the minimax robust Kalman estimation problem for a system under linearly correlated noise, uncertain noise variances, multiplicative noise, multi-step random measurement delays, and missing measurements. The system noise variance is uncertain but bounded above, and a set of Bernoulli distributed random variables with known probability is used to describe the multi-step random measurement delays and missing measurements from sensor to estimator. A novel model transformation method is proposed by using Hadamard product, and then the original system is converted into one only with uncertain fictitious noise variance. The robust time-varying Kalman predictor, filter and smoother are designed in a unified form based on the minimax robust estimation principle. A new robustness proof method, including Ger & scirc;gorin circle theorem, matrix elementary transformation, Hadamard product theorem and generalized Lyapunov equation method, is presented to demonstrate the robustness such that the actual estimation error variance is guaranteed to have minimal upper bound for all admissible uncertainties. The computational complexity is analyzed for the robust time-varying estimator in each step, and the robust steady-state estimator is designed. The convergence in a realization between the time-varying and steady state estimator is proved. A simulation example verifies the correctness and effectiveness of the proposed results.
ABSTRACT This paper designs a novel adaptive neural networks (NNs) event‐triggered fault‐tolerant control scheme, which can effectively address the impacts of sensor faults and unknown external disturbances on non‐strict feedback multi‐agent systems (MASs). First, the radial basis function (RBF) NNs are employed to reconstruct the sensor faults for mitigating its adverse effects on the system. Second, compared with the conventional disturbance scaling method, the external disturbance observer is constructed to accurately eliminate the adverse effects induced by external disturbances. Third, a novel dynamic event‐triggered switching threshold strategy (DETSTS) is proposed to efficiently balance the diverse requirements of trigger conditions in different environments. The designed controller ensures that all signals in the closed‐loop system are semi‐globally, uniformly, ultimately bounded (SGUUB). Finally, simulation results verify the effectiveness of the proposed strategy.
This paper investigates robust exponential stability and control for a class of discrete-time switched stochastic Hopfield neural networks with impulsive effects. First, a unified analysis framework is established for discrete-time switched stochastic impulsive systems. By combining Lyapunov theory with the average dwell-time approach, sufficient conditions in terms of linear matrix inequalities (LMIs) are derived to guarantee robust exponential stability and prescribed disturbance attenuation. Then, these results are extended to switched stochastic Hopfield neural networks with impulsive effects by incorporating sector conditions on the activation functions, leading to tractable state-feedback controller design criteria. The developed method explicitly captures the coupled influence of switching, stochastic disturbances, and impulsive behavior. Finally, the effectiveness of the proposed method is verified through numerical examples of photovoltaic energy storage charging stations and networked unmanned vehicle systems.
This article proposes a prescribed-time (PT) neural network composite learning tracking control scheme for a class of strict-feedback nonlinear systems with functional uncertainties. Specifically, a PT composite learning framework is constructed to improve the estimation performance of the unknown nonlinear functions under the weaker interval excitation (IE) condition. To guarantee PT tracking performance, a PT controller with time-varying bounded gains is designed to ensure that the tracking error converges to an arbitrarily small neighborhood of zero within the prescribed time, independent of initial conditions. Furthermore, a PT dynamic surface filter is developed to overcome the explosion of complexity inherent in traditional backstepping designs. Finally, numerical simulations and real-time experiments on a permanent magnet synchronous motor (PMSM) platform are conducted to validate the effectiveness and practical applicability of the proposed control scheme.
This paper proposes an adaptive fixed-time prescribed performance trajectory tracking control strategy based on finite-time command filtering for a class of quadrotor systems affected by external disturbances. Firstly, a nonlinear mapping and an error transformation function are introduced to establish an equivalent error model. Secondly, based on the backstepping method, fixed-time control, and adaptive control, a command-filter-based prescribed performance trajectory tracking controller for quadrotor systems with external disturbances is designed. This control strategy incorporates a command filter and fuzzy logic systems into the control framework. The command filter effectively avoids the repeated differentiation problem of the virtual control laws, significantly reducing computational complexity. The fuzzy logic systems are used to approximate the disturbance terms and unknown nonlinear terms in the model, enhancing the systems' robustness. In addition, the combination of prescribed performance control and fixed-time control ensures that the system tracking error converges within the boundaries defined by the prescribed performance function within a fixed-time, and the convergence time is independent of the systems' initial conditions. Finally, simulations verify the effectiveness and superiority of the proposed control method.
Iterative learning control applies to applications in which the same finite-duration task is repeated, with each instance termed a trial. The objective is to track a specified reference trajectory over a finite duration, termed the trial length. In some applications, such as multi-agent systems, tracking at each instant or point over the trial length is not required; only at selected points is it required, known as point-to-point iterative learning control. This article develops a new point-to-point design in which the points requiring tracking vary from trial to trial, and the solution minimizes energy, which is relevant to systems with a limited power budget. Also, an algorithm is developed to improve computational efficiency by sharing the burden among the agents forming the system. A numerical case study highlights the benefits of the new design.
The adaptive neural network (NN) output feedback optimal saturation control scheme is investigated for a single-phase photovoltaic (PV) grid-connected power system with partially unavailable states. The unavailable states are estimated by a state observer. The output feedback control scheme combines the adaptive dynamic programming (ADP) approach with the dynamic surface control (DSC) technique based on the backstepping design framework, in which the DSC technique can simplify the computation. By constructing an observer-critic-actor architecture, NNs are utilized via reinforcement learning (RL) to approximate the solution of the Hamilton-Jacobi-Bellman (HJB) equation, such that the difficulty of solving the HJB equation is overcome. By integrating the hyperbolic tangent function with a first-order auxiliary system, the adverse influence of the saturated links is removed. All the variables of the closed-loop PV power system are proved to be semi-globally uniformly ultimately bounded (SGUUB) by the Lyapunov stability theory. The simulation and comparative results show the feasibility and superiority of the presented control scheme.
This paper proposes an adaptive fuzzy fault-tolerant tracking strategy for uncertain nonlinear systems subject to full-state constraints and potential faults in sensors and actuators. To address the problems caused by full-state constraints, a barrier function is employed as an analytical tool to ensure that the system's state remains within a safe operational range. Meanwhile, the combined use of parameter separation and multi-adaptive law techniques effectively minimizes the influence of faults on system performance. Additionally, a finite-time command filter with error-compensation signals is embedded to prevent explosion of complexity. To further alleviate the communication load, an event-triggered fault-tolerant controller is designed, effectively avoiding Zeno behavior. The design of this controller ensures that all signals in the system remain bounded and enables the output signal to track the reference signal within a predetermined error range. In the end, the validity of the proposed strategy is confirmed by simulation results.
In this paper, an adaptive resilient control scheme based on reinforcement learning (RL) is proposed for the control of nonlinear multi-agent systems (MASs) under false data injection (FDI) attacks. When the system sensors are subject to unknown FDI attacks, traditional controller design methods are challenged by the inability to directly access all the original state information. To address this challenge, a novel coordinate-error construction and Nussbaum-type functions are incorporated into the control design, effectively attenuating the adverse impact of FDI attacks. Meanwhile, an adaptive event-triggered mechanism (AETM) is developed to substantially reduce network communication burden. The controller is synthesized by integrating a critic function with an actor-critic neural network-based RL algorithm, enabling accurate online estimation of lumped uncertainties. Using Lyapunov stability theory, it is rigorously proven that the closed-loop system is stable and that all signals remain bounded. Simulation results further corroborate the effectiveness and robustness of the proposed scheme, demonstrating improved control performance under FDI attacks with reduced communication load.
In steer-by-wire (SBW) vehicles, the steering feel feedback system is essential for providing realistic torque feedback to the driver. However, its performance is often degraded by the simultaneous presence of system uncertainties and time-varying delays caused by communication and computation. While existing studies have addressed either uncertainty or delay separately, their combined effect remains underexplored. This article proposes a novel integrated control framework that simultaneously handles both challenges. First, a barrier function-based adaptive super-twisting sliding mode control (STSMC) is employed to ensure robustness against system uncertainties. To further mitigate communication-induced delays, a relative threshold event-triggered mechanism is introduced, which significantly reduces unnecessary data transmission compared with conventional fixed-threshold methods. Second, to compensate for residual time-varying delays that cannot be eliminated by event-triggering alone, a Grey-Markov time delay prediction model is developed. This model combines grey system theory for trend extraction and Markov chains for stochastic fluctuation correction, enabling accurate real-time delay estimation. The predicted delay is then integrated into a fuzzy adaptive Smith predictor, which dynamically adjusts compensation parameters to enhance system robustness and response speed. Simulation results under high-speed and low-speed steering conditions demonstrate that the proposed strategy reduces the maximum absolute error by more than 50% and the root mean square error by more than 10% compared with conventional Markov-based methods, while also reducing communication frequency.
In this article, we investigate the joint estimation problem of the states and the in-domain Partial Differential Equation (PDE) parameters along with the unknown time-varying input for coupled wave PDE and infinite-dimensional Ordinary Differential Equation (ODE) systems. A finite set of measurements described by the sampled-in-space state of the ODE system is available to recover the distributed states along with the unknown parameters and input. Through the proposition of an appropriate Lyapunov functional the convergence proof is shown. The proposed Lyapunov functional considers lifted state solutions belonging to more regularized Hilbert spaces. Sufficient conditions on the observer's and adaptation laws' gains, along with the measurements' maximum sampling interval, are derived to guarantee the practical convergence of state estimation errors (uniform ultimate boundedness of the state estimation errors). The proposed observer is applied to characterize the spatiotemporal hemodynamic response in the brain from voxel-wise functional magnetic resonance imaging (fMRI) measurements. Numerical simulations are provided to demonstrate the efficiency of the theoretical findings.
