SFNAS-DDPG: A Biomass-Based Energy Hub Dynamic Scheduling Approach via Connecting Supervised Federated Neural Architecture Search and Deep Deterministic Policy Gradient

Abstract

The transition to a near-zero-emission power and energy industry for facing up to global warming issues is dominated by the use of renewable energy resources such as bioenergy and solar energy. When these resources are coordinated within an energy hub framework, the system’s flexibility is increased and dispatchable energy is provided by enhancing the share of renewable-dominated power. This paper proposes a dynamic scheduling framework for an energy hub with a biomass-solar hybrid renewable system. A hybrid forecasting model based on convolutional neural networks (CNNs) and Gated recurrent units (GRUs) is developed first to capture solar-related uncertainty sensibly, which will provide a great opportunity for the learning-based controller to determine an effective operation strategy in an optimal manner, especially on a cloudy-weather day. Then, a supervised federated neural architecture search (SFNAS) technique has been presented to eliminate the need for manual engineering of deep neural network models and the unnecessary computational burden associated with them. Finally, the deep deterministic policy gradient (DDPG), as an actor-critic deep reinforcement learning (DRL) methodology, enables the biomass-based energy hub to achieve cost-effective dynamic control strategies by addressing the decision-making problem as a highly dynamic continuous state-action model. The major conclusions of the numerical results show the effectiveness of the proposed SFNAS-DDPG method from average operating cost reduction up to 7.31% compared to the conventional DDPG model.

Keywords

• Actor-critic deep reinforcement learning
• biomass energy
• energy hub
• federated learning (FL)
• neural architecture search (NAS)