This study argues a dynamic optimal control approach for nonlinear systems with state and input delays. First, a feedback system identical to the delayed system is specified based on parallel control, enabling the dynamic optimal control issue to be formulated in light of the L2 gain condition. The virtual controller developed using the backstepping integral technique makes the dynamic control feasible, with its design highlighting inertial-like features, so that the response speed of controller can be altered. Then, the optimal control of the physical system is further expanded to that of the augmented system composed of it, accomplishing the control goal through virtual and actual interaction. In the implementation, the critic-actor structure facilitates the control challenge to be settled online. Additionally, the stability of the system and the convergence of neural network weights are discussed, and the research conclusions are supported by simulation examples.

The transient-state acceleration control of aircraft engines has consistently been a focal point in the field of aircraft engine control research. To address the controller design problem of transient-state acceleration in aircraft engines based on switched linear systems, a data-driven optimization control algorithm is proposed. First, the optimization control problem for the transient-state acceleration process of aircraft engines is mathematically described. Subsequently, the Bellman optimal criterion in optimal control is utilized to derive the invariance of a value function at switching instants, providing an initial value for the optimization of the next subsystem. The optimization problem is then transformed into solving a Lyapunov equation using the principles of dynamic programming, and an iterative adaptive dynamic programming (ADP) algorithm is introduced to obtain optimal feedback control gains, which makes the solution of Lyapunov equation approach the optimal solution infinitely. Finally, simulation validation is conducted on a switched linear system for the transient-state acceleration of a specific aircraft engine. The effectiveness and superiority of the proposed algorithm are verified by comparing to traditional proportion integral differential (PID) control algorithms and active disturbance rejection control (ADRC).

Human-machine interaction can incorporate several components which when suitably designed, synthesized, and integrated can benefit the human decision maker in performing various tasks and activities. These interactions can be designed through multiple modalities such as textural, graphic, audio, or any tangible interfaces depending on the specific nature of the decision-making task which needs to be performed by human. The control and decision-making architecture of such human-machine interaction can be defined in variety of ways. In this paper, we present an overview of an architecture which has been recently proposed in the literature based on the notion of open quantum systems. We present a brief interpretation of some of the main modelling and solution components of this framework. The main implication of this framework is that it presents the state of the decision maker and the interacting environment in a probabilistic framework, where the time evolution of this ensemble is estimated within the interpretation and solutions of the open quantum system.

Boilers are significant contributors to carbon emissions and pollutants across various industrial sectors. Accurately modeling furnace temperature is critical for optimizing combustion and enhancing operational efficiency. However, modeling poses significant challenges due to the interplay between rapidly changing dynamic processes and slowly varying static data. To address the coupling and redundancy inherent in these heterogeneous features, a hybrid framework for boiler temperature modeling (HFBTM) is proposed in this paper. The framework utilizes a multi-layer dense network to extract static features and a selective state space model (S3M) to capture dynamic features. These features are effectively combined through a hybrid feature fusion module using weighted integration, generating accurate temperature predictions across multiple future time steps. Compared with traditional single dynamic models, HFBTM mitigates information redundancy, reduces error propagation, and serves as an end-to-end furnace temperature prediction model that integrates both static and dynamic features. Experimental results demonstrate that the HFBTM framework delivers superior prediction performance across forecasting tasks of varying lengths. Compared with the existing methods, the proposed framework achieves higher accuracy and meets the requirements for precise modeling of boiler systems.

Overhead cranes play a critical role in manufacturing, shipping, and construction industries. To improve operational efficiency and safety, effective anti-swing control is essential for crane automation. Traditional anti-swing algorithms often struggle with the non-linearity of system and are incompatible with the existing velocity control interface. In this paper, we propose a novel anti-swing control method for overhead cranes based on embodied intelligence. We implement a conventional anti-swing control algorithm based on trajectory planning and PID controllers to generate demonstration data in simulated environment. Using the collected demonstration data, we apply imitation learning to train an embodied agent in performing anti-swing control. Action chunking with transformer (ACT) algorithm is utilized to enhance the ability of agent to model the mapping between observations and action sequences. In simulation experiments, our proposed method outperforms conventional anti-swing control algorithms in suppressing the maximum transient of payload and eliminating residual swing under similar efficiency.

A neurodynamic method (NdM) for convex optimization is proposed in this paper with an equality constraint. The method utilizes a neurodynamic system (NdS) that converges to the optimal solution of a convex optimization problem in a fixed time. Due to its mathematical simplicity, it can also be combined with reinforcement learning (RL) to solve a class of nonconvex optimization problems. To maintain the mathematical simplicity of NdS, zero-sum initial constraints are introduced to reduce the number of auxiliary multipliers. First, the initial sum of the state variables must satisfy the equality constraint. Second, the sum of their derivatives is designed to remain zero. In order to apply the proposed convex optimization algorithm to nonconvex optimization with mixed constraints, the virtual actions in RL are redefined to avoid the use of NdS inequality constrained multipliers. The proposed NdM plays an effective search tool in constrained nonconvex optimization algorithms. Numerical examples demonstrate the effectiveness of the proposed algorithm.