基于深度强化学习的微电网源-荷低碳调度优化研究

Research on Source-load Low-carbon Optimal Dispatching for Microgrid Based on Deep Reinforcement Learning

摘要:

提升可再生能源在能源供给中的比例成为实现低碳经济的重要举措之一。为减少碳排放量并降低用电成本，提出了一种基于深度强化学习的微电网低碳经济优化调度模型。首先，介绍了碳排放流理论并基于此构建了碳计量模型以及阶梯碳价模型；其次，将低碳经济优化问题转换为一个马尔科夫决策；最后，利用深度强化学习对该多目标优化问题求解。实验结果表明，所提方法通过控制发电机组的出力以及负荷的转移，有效地提升了系统经济性并降低了碳排放量。

Abstract:

Enhancing the proportion of energy supply from renewable energy sources in the system becomes a significant initiative to realize a low-carbon economy. A model based on deep reinforcement learning (DRL) for optimal allocation of low carbon economies in microgrid is proposed to mitigate system carbon emission and decrease system electricity cost. Firstly, carbon emission flow theory is introduced on which a carbon measurement model and a stepped carbon price model are constructed. Secondly, the low-carbon economy optimization problem is converted into a Markov decision. Finally, the multi-objective optimization issue can be addressed utilizing DRL. The experimental results demonstrate that the proposed approach is effective in boosting system economy and mitigating carbon emissions by regulating the capacity of generating units and shifting the load.