This article proposes a computational task scheduling optimization model that integrates reinforcement learning (RL) and metaheuristic algorithms, aiming to address the scheduling challenges brought by complex applications in cloud edge collaborative environments. In this architecture, there are complex dependencies between tasks, and some tasks involve sensitive data processing, making the design of efficient scheduling strategies particularly difficult. To model such dependencies, the task structure is abstracted as a directed acyclic graph (DAG). Traditional heuristic algorithms often exhibit poor adaptability and insufficient flexibility when facing heterogeneous computing clusters and diverse workflow structures. Therefore, this paper proposes a Q-learning based multi task multi-objective particle swarm optimization model (QM2PSO), which introduces the online learning and decision-making capabilities of RL into the particle swarm optimization algorithm (PSO). By dynamically adjusting the inertia weight and acceleration factor of the Q-learning algorithm, the search ability of particle swarm optimization in multi-objective optimization space is effectively improved, accelerating convergence to the Pareto front. The results indicate that QM2PSO outperforms traditional methods in key indicators such as task completion time, resource utilization, and task success rate, demonstrating stronger adaptability and optimization performance.
Wan et al. (Sun,) studied this question.