Published: Jan. 1, 2023

The optimal dispatch of energy storage systems (ESSs) presents formidable challenges due to the uncertainty introduced by fluctuations in dynamic prices, demand consumption, and renewable-based generation. By exploiting generalization capabilities deep neural networks (DNNs), reinforcement learning (DRL) algorithms can learn good-quality control models that adaptively respond distribution networks' stochastic nature. However, current standard DRL are limited constraint satisfaction unable provide feasible actions. To address this issue, we propose a framework effectively handles continuous action spaces while strictly enforcing environments space operational constraints during online operation. Firstly, proposed trains an action-value function modeled using DNNs. Subsequently, is formulated as mixed-integer programming (MIP) formulation, enabling consideration environment's constraints. Comprehensive numerical simulations show superior performance MIP-DRL framework, all delivering high-quality decisions when compared with state-of-the-art solution obtained perfect forecast variables.

Language: Английский