ISA transactions最新文献

This paper presents a novel reinforcement learning-based framework for pitch control in wind turbines, addressing the challenges posed by high nonlinearity and uncertainties in wind dynamics. Traditional pitch control methods, such as proportional-integral controllers, often struggle with adaptability and load mitigation under fluctuating wind conditions. To enhance control efficiency, the proposed framework employs a two-stage approach: policy transfer and policy refinement. In the policy transfer stage, a proportional-integral controller is used to estimate an initial control policy. This policy is then refined using the double deep Q-Network algorithm. Furthermore, a novel reward function is introduced as an improved version of a recent formulation in the literature, ensuring better alignment with practical operational scenarios. Simulations demonstrate that the proposed controller outperforms the industry-standard ROSCO controller, achieving a 15.87% reduction in power fluctuations, while preserving comparable structural load levels, with slight reductions in some cases. These results highlight the potential of reinforcement learning-based control to enhance wind turbine power regulation without compromising structural safety in terms of loads and vibrations.

This study proposes a robust fixed-time safety control scheme for a quadrotor unmanned aerial vehicle (QUAV) operating in environments with obstacles and external disturbances. Firstly, a robust high-order time-varying control barrier function (R-HO-TV-CBF) is designed as an obstacle avoidance constraint. This method achieves robustness without relying about any prior assumptions on the bounds of the disturbance and its derivative, effectively handles dynamic obstacle avoidance, suppresses oscillation and ensures optimal trajectory tracking. Subsequently, a nonsingular fixed-time sliding mode control (NFxSMC) method is developed to guarantee fixed-time convergence, independent of the initial state. Finally, simulation results demonstrate the superior performance of the proposed scheme in obstacle avoidance, disturbance rejection and error convergence.

Distributed convex optimization with time-varying or time-invariant cost functions remains one of the central challenges in multi-agent systems (MASs). Achieving efficient distributed optimization within a fixed/predefined-time continues to face difficulties such as restrictive assumptions and high computational complexity. This paper proposes innovative distributed optimization frameworks to address the aforementioned limitations. For time-invariant optimization problems, an estimator-based two-stage distributed protocol is introduced, which achieves both inter-agent consensus and convergence to the global optimum within a fixed/predefined-time. Notably, this protocol only requires strong convexity of the global cost function, thereby relaxing the constraints on local functions. For time-varying scenarios, an enhanced zero-gradient-sum (ZGS) framework is developed by integrating a zeroing neural network (ZNN) with sliding mode control. This framework not only eliminates dependence on initial conditions but also implicitly computes the Hessian inverse through ZNN dynamics, effectively avoiding the O(n³) computational burden associated with explicit matrix inversion. Numerical simulations validate the superior convergence speed and broad applicability of our method, attesting to its great potential for distributed optimization.

For heterogeneous linear multi-agent systems (MASs) in which not all followers have access to the information of multiple leaders, this paper studies the distributed output formation-containment problem. By converting the research question into the cooperative output regulation problem, a fully distributed control framework via a fully adaptive observer approach is developed. Since a part of the followers cannot directly access the states of leaders, a new observer is designed for reconstructing the system matrices and state information. Specifically, by introducing adaptive gains in place of fixed coupling coefficients in the observer design, the global topological information is no longer needed. Moreover, the conventional regulator equations cannot be directly applied for controller synthesis due to the time-varying estimated system matrices of the leaders. To overcome this challenge, the adaptive regulator equations are developed to dynamically compute the controller gains. Subsequently, both state and output feedback control strategies are presented to achieve the desired output formation-containment control objective. The validation of the proposed method is performed in the simulation results with practical and comparative examples.

This paper is concerned with a prescribed-time observer (PTO) for a class of linear systems by using a novel dynamic event-triggered mechanism (DETM) with bounded triggering conditions. Traditional PTOs often require real-time updates of the measurement output signal, leading to high communication costs. Event-triggered approaches can address this issue. However, event-triggered approaches hardly achieve prescribed-time convergence. In order to achieve fast state estimation while maintaining low communication costs, a dynamic event-triggered prescribed-time observer (DETPTO) is developed by using a parametric Lyapunov equation (PLE), in which the estimation of all system states can also be achieved within a prescribed-time T. To achieve this goal, a DETM is proposed by integrating both estimation error and measurement error, ensuring that the observer updates the measurement output only when necessary rather than continuously. Moreover, it is proven that the proposed DETPTO can avoid the Zeno behavior. Finally, numerical simulations validate the effectiveness of the proposed methods.

Under time-varying operating conditions, vibration signals from rotating machinery typically contain dense and non-proportional frequency components, which make it difficult for current time-frequency analysis (TFA) techniques to achieve high-precision and highly concentrated time-frequency representations (TFRs). Accordingly, the high-dimensional optimized extraction chirplet transform (HOECT) is proposed. In the HOECT, a window parameter adaptive optimization criterion is first designed to dynamically match local characteristics of the signal; second, the component matching CT (CMCT) is improved based on the proposed criterion, and its results are mapped into the time-frequency-chirprate domain to separate frequency components; finally, based on estimated local frequency and chirprate, the smoothing extraction operator (SEO) is constructed in the frequency-chirprate domain, which significantly enhances the energy concentration. In addition, a novel connected domain analysis (CDA) algorithm is proposed to effectively suppress strong noise interference and highlight key features of signal components. Simulation results illustrate that the HOECT can accurately characterize dense and non-proportional frequency components with high energy concentration. The superiority of the HOECT in operational state characterization and fault feature identification is further verified through two vibration signals of planetary gearboxes. Additionally, the whale sound signal results indicate that the proposed method exhibits strong robustness and applicability in analyzing complex non-stationary signals.

The unbalanced distribution of tyre normal forces induced by lateral load transfer poses a significant threat to vehicle handling stability under extreme conditions. However, the mitigation of such imbalance is inherently constrained by the suspension support capability. This work proposes a novel off-policy safe reinforcement learning-based cooperative (OSRC) controller that coordinates direct yaw control (DYC) and active suspension system (ASS) to optimize the distribution of tyre normal forces and enhance the handling stability of the four-wheel independent drive electric vehicle (FWID-EV) under extreme conditions. First, a DYC-ASS cooperative architecture is proposed to optimize the distribution of tyre normal forces in the FWID-EV system. Subsequently, a combined tyre-slip-angle and phase-portrait (CTP) method is proposed to characterize the relationship between the tyre force saturation and the lateral stability region (LSR). Based on a two-player game, the OSRC controller is proposed to achieve the cooperation between the DYC and the ASS and enhance the handling stability of the FWID-EV under extreme conditions. Finally, the effectiveness of the proposed OSRC controller is examined via the rapid control prototyping tests.

This article introduces a novel distributed optimization algorithm for time-varying objectives that achieves prescribed-time convergence. The method is developed based on zero-gradient-sum (ZGS) principles and employs a sliding-mode control framework. Specifically, the controller is designed to satisfy the ZGS condition within a prescribed time horizon, ensuring exact convergence by the user-specified deadline. A key contribution of this research is the introduction of adaptive parameters with time-varying scaling functions. This innovation addresses a fundamental limitation in existing methods by eliminating their dependency on global network information. Consequently, the algorithm achieves truly distributed control without requiring knowledge of Laplacian matrix eigenvalues. The algorithm offers significant advantages including complete freedom from initial condition constraints and local minimization requirements, full independence from global topological information, and rigorous prescribed-time convergence guarantees. Theoretical analysis establishes the convergence properties through an integrated approach that combines optimization theory with Lyapunov stability analysis. Numerical simulations demonstrate the algorithm's effectiveness and superior performance in handling time-varying optimization problems compared to existing approaches.

Distributed generation units (DGUs) in grid-connected microgrids are typically controlled as current sources. Nevertheless, when the microgrids are isolated, some DGUs must switch their control law to operate as voltage sources. Switching between controllers implies unwanted transients at the DGUs output power due to mismatches between controller's outputs, challenging both voltage and frequency transients and potentially leading to instability. This work proposes a bumpless-transfer scheme based on sliding-mode reference conditioning for the DGUs to achieve a smooth transition between grid-connected and island modes in electrical microgrids. By guaranteeing the establishment of a sliding regime, the suggested bumpless algorithm reduces to zero the mismatch between the DGU controllers' outputs when the switch is performed. In this manner, jumps in the control action and the associated unwanted transients are avoided regardless of when the control law switches. Results show good and robust performance in the face of both islanding and reconnection processes.

Active power control (APC) is an effective way to address the instability problem caused by high wind energy penetration in power systems. This study presents a coordinated APC scheme to reduce the pitch system's loads while ensuring accurate active power tracking performance. Firstly, a pitch activation limitation parameter updated by wind speed is designed to expand the rotor speed regulation range. Subsequently, a neural network (NN)-based controller with a segmented weight updating mechanism is designed to achieve smooth and stable pitch angle variations, effectively reducing the pitch system's loads. Furthermore, we develop a robust differentiator to estimate derivatives of the rotor speed and wind speed thereby avoiding the use of additional sensors. Finally, simulations on the OpenFAST platform demonstrate the effectiveness of our method.