Optimization Scheduling of Environmentally Friendly Microgrid Incorporating Energy Storage and Carbon Sequestration Using Deep Reinforcement Learning Approach

This paper addresses challenges in power system operations such as high carbon emissions and low efficiency, proposing an environmentally friendly microgrid (ECO-MG) energy management strategy based on deep re-inforcement learning (DRL). The strategy comprehensively considers economic benefit, environmental friendliness, and energy comfort index. Firstly, a novel microgrid model is constructed, incorporating a wind-solar integrated generation system, controllable diesel generators, and an advanced battery system with energy storage and carbon sequestration capabilities. The model also includes thermo-statically controlled loads and residential loads influenced by electricity prices. Secondly, this paper establishes a ladder-type carbon trading mechanism, participating in carbon markets based on daily actual emissions. Then, leveraging a proposed two-stage multi-objective reward function based on carbon sequestration, this paper em-ploys Soft Actor-Critic with Long Short-Term Memory (LSTM-SAC) algorithm to optimize internal energy dis-patch within the microgrid. Simulation results demonstrate that the proposed strategy achieves economically viable and environmentally sustainable operations while ensuring comfort, effectively addressing uncertainties and achieving desirable convergence outcomes.