A Comprehensive Review of Energy Optimization Strategies for Sustainable Green Computing
TL;DR
This comprehensive literature review indicates a lack of research combining high fidelity energy modelling with adaptive control within a single closed - loop framework, and proposes a hybrid architecture combining an LSTM network for multi horizon energy demand forecasting with a Deep Reinforcement Learning agent for real time resource optimization.
Abstract
Rapid growth of cloud platforms and large-scale data centers has placed extreme pressure on global power grids and has heavily amplified carbon footprints globally. This literature review delivers a step by evaluation of energy-conservation strategy for green computing, organized across five complementary domains: (1) containerized workload management lever-aging Kubernetes combined with nature-inspired solvers such as Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), and Dynamic Voltage and Frequency Scaling (DVFS); (2) data-driven energy forecasting using Long Short-Term Memory (LSTM), Transformer, and ensemble neural architectures; (3) thermodynamic optimization of cooling subsystems via metaheuristic search, multi-agent reinforcement learning, and intelligent thermal control; (4) power management at edge data centers through demand aggregation, workload migration, and online Lyapunov-based decision-making; and (5) clean energy sourcing via combined photovoltaic–wind–hydrogen storage topologies. The contribution of this review is synthesized from ten IEEE papers the survey identifies recurring methodological patterns, outstanding technical constraints, and pathways aiming for a cohesive structure for sustainability, autonomously managed infrastructure, self-managing cloud and cloud frameworks. Our comprehensive literature review indicates a lack of research combining high fidelity energy modelling with adaptive control within a single closed - loop framework. To mitigate this limitation, we propose a hybrid architecture combining an LSTM network for multi horizon energy demand forecasting with a Deep Reinforcement Learning agent for real time resource optimization. Overall, the examined works report energy reductions of 9.2% to 45%, quantifiable Power Usage Effectiveness (PUE) improvements, and significant CO2 cuts, validating the impactful potential of algorithmic and systems level innovations for environmental sustainability.