Optimizing Hybrid Renewable Energy Systems for Rural Electrification: A Multi-Criteria Decision-Making Approach with Enhanced Whale Optimization Algorithm

Abstract

Cloud computing offers flexible and economical infrastructure, yet dynamic resource allocation is a significant problem owing to volatile workloads, disparate resources, and varied application needs. Static or rule-based solutions cannot be flexible, and machine learning methods commonly miss complex dependencies. Reinforcement Learning (RL) provides flexibility but is not scalable with large state spaces, and current graph-based solutions are heavily dependent on hand-crafted relations. To overcome these constraints, we suggest an end-to-end approach that combines RL and Graph Neural Networks (GNNs) for dynamic resource allocation. The cloud system is represented as a graph, where virtual machines, servers, and network devices are nodes and their dependencies are edges. The GNN captures complex representations of these dependencies, which are utilized by an RL agent to perform well-informed allocations. We compare our proposed GNN-RL framework with static, threshold-based, and independent RL methods using CloudSim. Our results show dramatic improvements in resource utilization, task execution time, and energy consumption. Our work identifies the feasibility of integrating GNN-RL to offer a scalable and flexible solution for resource management in clouds with potential to open up more intelligent and efficient cloud computing platforms.