Energy management in smart grids: An Edge-Cloud Continuum approach with Deep Q-learning

This paper introduces JEMADAR-AI, an approach to energy management within smart grids, leveraging an edge-cloud continuum architecture coupled with Deep Q-Learning to optimize the operation of smart home devices. The main hypothesis of this work is that combining advanced machine learning models with edge- cloud computing can significantly improve energy efficiency and cost savings in smart grids. The proposed system utilizes SARIMA for seasonal trends and LSTM for long-term dependency models to forecast energy consumption and production, enabling proactive decision-making to balance supply and demand in real-time. JEMADAR-AI employs a deep reinforcement learning algorithm (Deep Q-Learning) to optimize appliance operations, dynamically adjusting energy usage based on predicted demand and supply fluctuations. This ensures that household energy consumption aligns with production capabilities, particularly during periods of renewable energy generation. The architecture combines the high processing power of cloud computing for long-term forecasting with the low-latency responsiveness of edge computing for real-time appliance control. This Edge-Cloud Continuum approach provides an efficient solution for managing energy in distributed smart grids. The experimental results, obtained using Gridlab-D and Omnet++ simulations, demonstrate that JEMADAR-AI improves decision-making speed by 32.25% and reduces household energy bills by 22.11% compared to traditional cloud-based systems.