ABSTRACT Heating, ventilation and air conditioning (HVAC) chilled water systems offer significant potential for energy saving and reinforcement learning (RL) methods have been extensively studied and validated for optimizing HVAC energy consumption. However, RL's low sample efficiency and reliance on randomized exploration limit its practical application. To enhance the robustness and stability of RL‐based energy optimization methods, an RL optimization approach based on a mechanism‐data hybrid‐driven model is proposed, derived from the MBPO (Model‐based policy optimization) scheme. Firstly, a novel end‐to‐end HVAC chilled water system model is developed to serve as the foundation for the hybrid model design. Second, a hybrid‐driven RL environment model framework is introduced, combining a mechanistic model with a probabilistic neural network. The mechanistic component provides generalization capabilities, while the data‐driven component offers adaptability. Third, improvements to MBPO are proposed, including double policy optimization and adaptive branch rollout, further to enhance dynamic environmental adaptability and model utilization efficiency. Finally, comparative and ablation experiments conducted using both simulation environments and measured data demonstrate that the proposed method achieves higher learning efficiency and improved robustness.

ABSTRACT This study presents the development of a robust second‐order sliding mode controller (SOSMC) that incorporates position error constraints into the control design. To achieve this, a barrier Lyapunov function (BLF) is employed, enabling the enforcement of predefined bounds on tracking errors throughout the system's evolution. This approach ensures that errors remain within previously known limits, thereby improving safety and reliability in practical applications, particularly robotic systems. The proposed control scheme guarantees the existence of a sliding mode and achieves exponential convergence of the tracking errors, even in the presence of bounded disturbances and model uncertainties. Hence, the main contribution of this study is the integration of the BLF into the SOSMC framework, which not only maintains robustness but also addresses the critical issue of constraint satisfaction. This is often overlooked in traditional sliding mode designs. The effectiveness and improved performance of the proposed controller are validated through simulation studies conducted on a three‐degree‐of‐freedom robotic manipulator and through experiments on a six‐degree‐of‐freedom robotic manipulator. Comparative results demonstrate that, unlike a conventional SOSMC without error constraints, the proposed controller successfully maintains position errors within the specified limits while preserving fast convergence and robustness. These findings highlight the significant benefits of incorporating barrier Lyapunov functions in sliding mode control strategies for systems with strict performance and safety requirements.

Abstract The roller kiln with multi‐temperature zones used for cathode material sintering is an interconnected system with time delay in energy transfer and precise control of the temperature for the preparation of cathode materials for lithium‐ion batteries. However, the interconnection between the temperature zones, the time delay of the temperature state, and the disturbance of the external environment make it difficult to control the sintering process. For this reason, this paper develops a distributed control of the temperature of roller kiln based on adaptive dynamic programming (ADP). Firstly, the heat transfer mechanism of sintering process is discussed; the law of energy conservation provides the physical basis for the sintering process on which the autocorrelation function method identifies the time delay. Then, the cost function is constructed by combining the Lyapunov–Krasovskii function containing the time delay, the temperature state, the heating power of the silicon carbon rod and the perturbation. Subsequently, Hamilton–Jacobi–Isaac equation with the optimal cost function and the optimal distributed control strategy are approximated by neural network of ADP. Finally, the stability of the closed‐loop system is proved by Lyapunov functional analysis, and the effectiveness of the proposed method is verified by the simulation results of roller kiln temperature control.

ABSTRACT Robotic manipulators are complex mechanical systems that exhibit highly nonlinear dynamics and subject to various forms of disturbances such as friction, external forces and other unmodelled dynamics. Mathematical models representing robot dynamics are extensively used for design, simulation and control purposes, and can be derived through analytical and experimental methods. Dynamic behaviours predicted by mathematical models often deviate from the observed dynamics of robot manipulators because of external disturbances, parametric uncertainties, and unmodelled dynamics. The observer‐based control that eliminates the need for highly accurate system modelling is an appealing approach for robot control. This paper introduces an extended state observer‐based structure that can be either utilized as a stand‐alone controller or implemented within a model‐based adaptive control scheme. The proposed control scheme allows the implementation of extended state observers independently of the availability and quality of the dynamic model. The stability of the proposed controllers in presence of model uncertainties and generalized disturbances is investigated through the Lyapunov analysis. The experiments performed on a six‐DoF industrial robot validate the theoretical stability results. Evaluation of performances of the proposed controllers in various operating conditions are presented in a comparative manner. Experimental results show that the extended state observer‐based controller outperforms the adaptive controller in trajectory tracking performances.

ABSTRACT Nonlinear filtering algorithm is the key technology for dealing with complex systems in sensor data processing. To improve the filtering accuracy of the nonlinear filtering algorithm in the non‐Gaussian case, an improved version of the Gaussian sum algorithm, the Gaussian sum adaptive sparse grid quadrature filter (GSASQF), is proposed. The proposed algorithm overcomes the challenges by introducing the Gaussian sum principle, which converts the non‐Gaussian state and noise in the system into the form of weighted sum of Gaussian components. Based on the Bayesian filtering framework, a three‐level sparse grid sampling rule is introduced, with the sparse grid orthogonal filtering algorithm serving as the sub‐filter. By determining the sampling point parameters, the filtering process for each combination of Gaussian components is implemented, thereby ensuring the filtering accuracy of each group. In addition, in combination with the ideal of data‐driven, the weight of each Gaussian component combination is adaptively updated inversely by the values of the sensor measurement, which improves the global filtering accuracy of nonlinear system under non‐Gaussian noise. The combination of these three improvements enables high‐precision filtering of non‐Gaussian non‐linear systems. Theoretical analysis and simulation confirm that the proposed GSASQF algorithm provides advantages in filtering accuracy for nonlinear non‐Gaussian filtering problems.

ABSTRACT This article proposes a data‐driven model predictive control (MPC) method for multirotor collision avoidance, considering uncertainties and the unknown dynamics caused by a payload. To address this challenge, sparse identification of nonlinear dynamics (SINDy) is employed to derive the governing equations of the multirotor system. SINDy is capable of discovering the equations of target systems from limited data, under the assumption that a few dominant functions primarily characterize the system's behavior. In addition, a data collection framework that combines a baseline controller with MPC is proposed to generate diverse trajectories for model identification. A candidate function library, informed by prior knowledge of multirotor dynamics, along with a normalization technique, is utilized to enhance the accuracy of the SINDy‐based model. Using data‐driven model from SINDy, MPC is used to achieve accurate trajectory tracking while satisfying state and input constraints, including those for obstacle avoidance. Simulation results demonstrate that SINDy can successfully identify the governing equations of the multirotor system, accounting for mass parameter uncertainties and aerodynamic effects. Furthermore, the results confirm that the proposed method outperforms conventional MPC, which suffers from parameter uncertainty and an unknown aerodynamic model, in both obstacle avoidance and trajectory tracking performance.

ABSTRACTWe deduce a sufficient condition for the exponential (integral) turnpike property for infinite‐dimensional generalized linear‐quadratic optimal control problems in terms of structural properties of the control system, such as exponential stabilizability and detectability. The proof relies on the analysis of the exponential convergence of solutions to the differential Riccati equations to the algebraic counterpart, and on a necessary condition for exponential stabilizability in terms of a closed range test.

ABSTRACT The challenges of communication interruptions have caused unmanned vehicles to lose the real‐time transmission of essential data. The traditional control architecture, designed for continuous measurement, is inadequate to address the constraints posed by the fragility of facilities in harsh maritime conditions. Therefore, it is necessary to enhance the transient stability performance in non‐ideal conditions and environments. This work proposes a novel intermittent predefined‐time control algorithm. This approach eliminates the need for frequent measurement and recalculation of control laws at every moment, and extends the usage scenarios of intermittent control from first‐order systems to second‐order systems. During the holding interval, the control input remains as the last control signal in the driving interval, provided that the validated stability conditions are satisfied. The designed controller also ensures stability with zero control inputs during the holding interval. Furthermore, the proposed method aims to optimise energy consumption in control and measurement. Using the Lyapunov theorem and mathematical induction, sufficient conditions for achieving predefined‐time convergence in intermittent control are derived and verified with unmanned vehicles. The proposed method demonstrates a 5.4% reduction in tracking error compared to the fixed‐time control method, and a 7.9% reduction in tracking error compared to the event‐triggered predefined‐time control method. This approach reduces the frequency of control law recalculations, measurements and energy consumption.

Abstract This study concentrates on achieving quasi‐bipartite synchronization within signed coupled inertial neural networks featuring time‐varying delays. The pinning controller technique addresses this problem within a structure incorporating cooperative and competitive interaction among the nodes. With the structurally balanced networks, some linear matrix inequality based sufficient conditions are derived for both reduced and non‐reduced order methods with the help of Lyapunov–Krasovskii functional to achieve the quasi‐bipartite pinning synchronization criterion. Further, the error bound is derived analytically for the leader–follower representation of the signed coupled inertial neural network model. At last, a numerical simulation result is provided to verify the correctness of the established theoretical results.

ABSTRACT This article investigates the finite‐time event‐triggered controller design with minimum learning parameters (MLP) for nonlinear systems using neural networks in the presence of uncertainty. Specifically, firstly, the neural networks are devised to compensate online for the uncertain nonlinear functions. Then, a finite‐time prescribed performance function is employed in the controller design to achieve that the tracking error converges to within a prescribed region at any setting time. At the same time, the transient responses (e.g., maximum overshoot and convergence speed) can be enhanced for the tracking error. After that, unlike ordinary dynamic event‐triggered strategy, the developed dynamic event‐triggered methodology can further increase the triggering interval, which leads to the network bandwidth can be effectively saved. Moreover, one can prove that all the closed‐loop signals remain bounded and the Zeno phenomenon can be excluded. Finally, the advantages of the proposed strategy can be illustrated by two examples.