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.

中文翻译：

---

智能电网中的能源管理：具有深度 Q 学习的边缘-云连续体方法


本文介绍了 JEMADAR-AI，这是一种在智能电网中进行能源管理的方法，它利用边缘-云连续架构和深度 Q-Learning 来优化智能家居设备的运行。这项工作的主要假设是，将先进的机器学习模型与边缘云计算相结合，可以显著提高智能电网的能源效率并节省成本。所提出的系统利用 SARIMA 进行季节性趋势，利用 LSTM 进行长期依赖模型来预测能源消耗和生产，从而实现主动决策以实时平衡供需。JEMADAR-AI 采用深度强化学习算法 （Deep Q-Learning） 来优化设备运行，根据预测的需求和供应波动动态调整能源使用。这确保了家庭能源消耗与生产能力保持一致，尤其是在可再生能源发电期间。该架构结合了用于长期预测的云计算的高处理能力，以及用于实时设备控制的边缘计算的低延迟响应能力。这种边缘-云连续体方法为管理分布式智能电网中的能源提供了一种高效的解决方案。使用 Gridlab-D 和 Omnet++ 仿真获得的实验结果表明，与传统的基于云的系统相比，JEMADAR-AI 将决策速度提高了 32.25%，并将家庭能源费用降低了 22.11%。