Smart energy supply scheduling for green remote telecom with data-driven deep reinforcement learning

The backbone of modern mobile communication networks is comprised of wireless telecom base stations, which serve vital functions. A significant challenge arises in remote or underdeveloped regions where power supply to these base stations is often unreliable or entirely absent. Integrated energy systems, combining solar, wind, diesel generators, and the electrical grid, present a promising solution. Effective and intelligent scheduling of these systems is paramount for ensuring uninterrupted base station operation, maximizing the integration of renewable energy, and lowering overall energy expenses. The inherent dual uncertainty of energy demand and supply poses a primary obstacle to intelligent scheduling. This research proposes a robust energy scheduling modeling framework for integrated energy systems, grounded in empirical risk minimization (ERM), forecasting methodologies, and deep reinforcement learning (DRL). This framework strategically coordinates the activation of various energy sources – grid, solar, diesel generators, and battery storage – to meet fluctuating base station power demands. By seamlessly blending predictive control with DRL, utilizing rolling forecasts as system state indicators, and employing the proximal policy optimization (PPO) algorithm for training, the proposed model demonstrably surpasses traditional predictive control and rule-based methods in both renewable energy utilization efficiency and total energy cost, as evidenced by real-world data from remote telecom base stations.