Learning-Based Grey Wolf Optimizer for Resource Optimization Design in Cloud-Edge Collaborative Environments

Cloud-edge collaborative computing integrates the powerful computing capacity of cloud centers with the low-latency response of edge nodes, posing substantial challenges to resource scheduling under highly dynamic task loads. Balancing system operating costs and processing capacity constitutes a critical issue that urgently needs to be addressed. To tackle this problem, this paper models the cloud-edge collaborative system as a multi-server queuing system. It establishes a Cost-Performance Ratio (CPR) objective that jointly considers system capacity, operating costs, and delay penalties. To address the rigid, linearly decreasing convergence factor of the standard Grey Wolf Optimizer (GWO), a Q-learning-based Grey Wolf Optimizer (QGWO) is proposed that integrates an adaptive convergence factor, hierarchical weights, stagnation-duration jumping, periodic replacement of inferior wolves, and a greedy-strategy-based position acceptance mechanism. To evaluate online reconfiguration capability, a dynamic step-changing task-arrival-rate schedule is used. Extensive simulations show that our proposed QGWO achieves a cumulative normalized regret 27% lower than that of the standard GWO, providing an effective scheme for on-demand dynamic resource scheduling in cloud-edge environments.