Abstract: The increasing deployment of AI applications at the edge has created an urgent need for intelligent energy management systems that can dynamically optimize power consumption while maintaining acceptable performance levels. Traditional approaches often treat energy consumption as a secondary concern, leading to suboptimal resource utilization and reduced operational sustainability. This research introduces a comprehensive energy-aware policy optimization framework utilizing Proximal Policy Optimization (PPO) for edge artificial intelligence applications, addressing the critical challenge of balancing computational performance with energy efficiency in resource-constrained environments. Our proposed framework integrates real-time energy monitoring, adaptive policy learning and intelligent resource allocation to create a holistic solution for edge AI deployment. The methodology employs a multi-objective optimization approach that considers both immediate energy costs and long-term performance implications, utilizing advanced reinforcement learning techniques to learn optimal policies from environmental feedback. Through extensive experimentation on real-world datasets including environmental sensor networks and mobile edge computing scenarios, we demonstrate significant improvements in energy efficiency while maintaining or enhancing computational performance. The results show up to 34.6% reduction in energy consumption compared to baseline methods with improved stability and adaptability across diverse operational conditions. This research contributes to the growing field of sustainable AI by providing practical solutions for energy-conscious edge computing deployment, particularly relevant for IoT applications, autonomous systems and smart city infrastructure where energy efficiency directly impacts operational viability and environmental sustainability.
Keywords: Edge Computing, Energy Optimization, Proximal Policy Optimization, Reinforcement Learning, Sustainable AI, Resource Management, IoT Applications, Power Efficiency
Copyright Form