As the paradigm of cloud computing burgeons, task scheduling emerges as a critical mechanism in enhancing resource utilization and quality of service. However, scheduling tasks modeled by Directed Acyclic Graphs (DAGs) poses significant challenges due to their inherent parallelism and dependency constraints. Deep reinforcement learning (DRL) offers a promising approach to address these challenges, yet existing approaches such as the Deep Deterministic Policy Gradient (DDPG) algorithm are hindered by issues including unstable Actor network training. This study introduces an advanced DDPG-based scheduling algorithm tailored for the unique complexities of DAG cloud computing task scheduling. We fortified the Actor network by incorporating a supervised learning approach to adjust its loss function, thereby stabilizing training outcomes. Empirical analyses across several critical metrics—task completion time, server energy consumption, and standard deviation of resource utilization—demonstrate that our refined model substantially outperforms traditional scheduling methods and unmodified DRL algorithms. The findings affirm that the enhanced DDPG algorithm not only expedites scheduling effectiveness but also reduces energy consumption while maintaining service quality, showcasing the extensive potential and pragmatic value of applying DRL in cloud computing task scheduling. Our improved approach is poised to contribute a fresh perspective to future research and application in DAG task scheduling.

In this paper, we introduce a novel approach combine the high-gain disturbance observer and super-twisting sliding mode control strategy. In this method, the disturbance observer is cleverly used to generate accurate estimates of the disturbance in real time, and then these estimates are incorporated into the control law to achieve effective compensation for the disturbance effect. By using a single parameter, the observer can be stabilized while the parameter tuning effort is reduced. In addition, interference estimation is combined with a super-twisting sliding mode controller to reduce the chattering effect caused by discontinuity. In order to ensure the stability of the system, Lyapunov stability analysis is further carried out to ensure that the tracking error is kept within the bounded range. To comprehensively evaluate the efficacy of the proposed control methods, this paper adopts a combined approach, encompassing both simulation and experimental methods. The research data show that the tracking accuracy and stability of the quadrotor UAV controller are significantly improved when the disturbance observer is introduced into the controller.

This paper studies the optimal trajectory tracking control strategy of robotic manipulators in the presence of external disturbances. Super-twisting algorithm (STA) is used as switching controller to effectively weaken chattering.  The choice of the barrier function as the gain of STA avoids the estimation of the disturbance upper bound and does not require the design of the low-pass filter. The stability of the closed-loop system is proved by Lyapunov theory. Finally, the effectiveness of the proposed method is verified by simulation experiments.