DBO-DDQN: a cloud-edge collaborative task scheduling strategy integrating dung beetle optimization and temporal difference learning

Abstract To tackle the challenges of scheduling computationally intensive tasks in edge computing environments, this paper proposes a novel hybrid scheduling method called “Dual-Stage Dung Beetle Optimization and Value Approximation Learning Guided by Clustering” (DBO-DDQN). First, a lightweight task clustering method is designed based on task size and time urgency to reduce the computational complexity of task scheduling. Then, the foraging behavior of a dung beetle population is simulated and the foraging process is improved to optimize the matching between task clusters and computing nodes. Finally, a novel temporal difference learning method is proposed to further optimize the matching strategy between tasks within clusters and computing nodes. Experimental results show that, compared to cat swarm optimization (CSO), red-tailed hawk algorithm (RTH), deep Q networks (DQN), DDQN, and Dueling DQN, the proposed algorithm improves task success rates by 15%, 13%, 23%, 19%, and 22%, and enhances weighted multi-objective scheduling performance by 49%, 67%, 33%, 26%, and 22%, respectively, while maintaining low latency and energy consumption. These results highlight the algorithm’s effectiveness and superiority in managing large-scale, computation intensive task scheduling within edge computing environments.