Autoencoder-aided Bi-level Evolutionary Algorithm to Solve Energy-Minimized Task Scheduling Problems of Cloud Computing

This paper considers an energy-minimized deadline-constrained task scheduling problem in cloud computing environment. To find its feasible schedules in a short time, an autoencoder-aided bi-level evolutionary algorithm is proposed. In it, a customized autoencoder is developed to learn compact latent representations of task sequences, capturing complex dependencies and structural features. Coarse-grained evolutionary operators explore promising regions in a low-dimensional latent space, while fine-grained operators refine solutions in the original solution space. Various numerical experiments are performed to compare the proposed method with several classic heuristics and some recently-developed methods. The results show that our proposed approach achieves superior energy efficiency and takes less computational burden compared with its peers.