Energy scheduling strategy for energy hubs using reinforcement learning approach

The rapid integration of Renewable Energy Sources (RES) into energy systems is critical for achieving a fossil fuel-free future. However, it also introduces significant challenges, such as increased system complexity and uncertainties in energy output. To overcome these challenges, this paper presents a model-free Deep Reinforcement Learning (DRL) method to energy hub scheduling using state-of-the-art algorithms, specifically Soft Actor-Critic (SAC) and Proximal Policy Optimization (PPO). Breaking away from traditional energy management systems that typically use discrete action spaces, this approach employs a multi-dimensional, continuous action space, providing a more accurate framework for decision-making. Furthermore, the current study adopts a comprehensive perspective by considering not only energy supply and operational costs, but also the often-overlooked impact of emitted emissions in the objective function. The results yield a Pareto space encompassing various criteria, offering a diverse scheduling framework that balances economic feasibility with environmental sustainability. By presenting a range of optimal solutions, this study contributes to the development of more resilient and sustainable energy systems. • Deep Reinforcement Learning is applied to determine optimal energy system scheduling. • Continuous state and action spaces represent the nature of EMS more accurately. • The objective function encompasses operational costs, energy supply, and emissions. • Weight factors are introduced to examine trade-offs between costs and emissions. • The Pareto space for operational strategies is explored and identified.