Aggregating Learning Agents for Microgrid Energy Scheduling During Extreme Weather Events
Aggregating Learning Agents for Microgrid Energy Scheduling During Extreme Weather Events
复制标题
DOI：
10.1109/pesgm46819.2021.9637949
复制
发表时间：
2021-07
期刊：
影响因子：
--
通讯作者：
Avijit Das;Z. Ni;Xiangnan Zhong
中科院分区：
文献类型：
--
作者：
Avijit Das;Z. Ni;Xiangnan Zhong
有效利用微电网发电资源（MGR）是非常重要的，尤其是在天气相关事件期间。现有的单智能体和合作多智能体强化学习（RL）方法可能是不可行的，计算昂贵的调度MGRs有效时，有一些概率的天气相关的紧急事件。在本文中，我们提出了一个Q-学习方法与多个本地代理的并网微电网应用程序，并表明所提出的集成能够有效地调度MGRs正常和天气相关的紧急事件。具体来说，我们利用不同的本地Q学习代理来学习不同的微电网事件，并将学习到的值函数聚合到以概率方式处理整体微电网能量调度的全局代理。数值仿真验证了该方法的有效性。讨论了发电机组的有效利用率和停电时间对系统性能的影响。两个案例研究不同的停电概率，我们提出的方法来评估性能。
Efficient utilization of the microgrid generation resources (MGRs) is important especially during weather-related events. Existing single-agent and cooperative multi-agent based reinforcement learning (RL) approaches may be infeasible and computationally expensive for scheduling MGRs effectively when there are some probabilistic weather-related emergency events. In this paper, we propose a Q-learning approach with multiple local agents for a grid-connected microgrid application and show that the proposed integration is capable of scheduling MGRs efficiently for both normal and weather-related emergency events. Specifically, we utilize different local Q-learning agents to learn different microgrid events, and aggregate the learned value functions to the global agent who handles the overall microgrid energy scheduling in a probabilistic way. Numerical simulations are performed to validate the effectiveness of the proposed method. The influences of the effective utilization of the MGRs and power outage duration are discussed. Two case studies with different power outage probabilities are presented to evaluate the performance of our proposed method.