A Novel Deep Reinforcement Learning Framework for Autonomous Energy Management in Off-Grid Solar Microgrids Serving Nigerian Rural Communities

Rural electrification through solar microgrids is central to Nigeria's energy access strategy, but the energy management systems currently deployed in rural microgrids are rule-based and fail to optimise generation, storage, and demand management efficiently under the stochastic demand patterns of Nigerian rural communities. This study develops a novel deep reinforcement learning framework for autonomous energy management in off-grid solar microgrids serving Nigerian rural communities. The framework employs a twin-delayed deep deterministic policy gradient (TD3) algorithm extended with three original contributions: a stochastic demand prediction module trained on load data from twelve rural Nigerian microgrids in Plateau and Cross River States; a microgrid environment simulator calibrated with Nigerian solar irradiance profiles, battery degradation models, and community demand behaviour; and a multi-objective reward function balancing energy reliability, battery health preservation, and diesel backup minimisation. The framework was trained in simulation over 10,000 episodes and evaluated against rule-based baseline controllers in both simulation and in a hardware-in-the-loop testbed. The TD3 agent achieved a 23 percent reduction in unmet demand hours and a 18 percent improvement in battery cycle life extension compared to the best rule-based baseline. Diesel dependency decreased by 31 percent. The study constitutes an original contribution to deep reinforcement learning application in Sub-Saharan African energy systems and provides an open-source simulation environment for future research.

Keywords: deep reinforcement learning, energy management, solar microgrid, rural electrification, Nigeria