Adaptive reinforcement learning framework for sustainable microgrid optimization in arid urban environments.

Managing energy in urban microgrids is a major issue because of the high degree of variability of renewable energy sources and the dynamic nature of the urban demand especially in regions with arid climates that pose extreme temperatures that provide volatile cooling loads. Existing energy management systems (EMS) often relied on static or rule-based control systems, which lack the responsiveness needed to manage the variability inherent in renewable generation and fluctuating demands. This study presents a simulation-based and adaptive reinforcement learning (RL)-based energy management framework that addresses persistent inefficiencies in coordinating diverse energy sources within urban microgrids, particularly in arid regions, to bridge this gap. In this study, it was tried to develop a simulation-driven platform combining EnergyPlus with Python/TensorFlow RL agents to dynamically optimize the dispatch of solar, wind, diesel, and battery resources. Unlike prior approaches, the proposed system also combined inter-microgrid communication via MQTT protocols, enabling real-time energy sharing. The framework was validated in a case study reflecting Riyadh's climatic conditions, where it significantly improved operational and environmental performance. Simulation outcomes demonstrated high predictive accuracy for hourly and annual consumption (R² = 0.94 and 0.90, respectively). Compared to baseline methods, the RL-based approach reduced CO