基于MADDPG算法的建筑群柔性负荷优化调控方法  

Optimized control method for flexible load of a building complex based on MADDPG reinforcement learning

作　　者：包义辛 徐椤赟 杨强[1] BAO Yixin;XU Luoyun;YANG Qiang(College of Electrical Engineering,Zhejiang University,Hangzhou 310027 China;Zhejiang Baima Lake Laboratory Company Limited,Hangzhou 310056,China)

机构地区：[1]浙江大学电气工程学院,杭州310027 [2]浙江省白马湖实验室有限公司,杭州310056

出　　处：《综合智慧能源》2023年第7期61-69,共9页Integrated Intelligent Energy

基　　金：中国电机工程学会青年人才托举工程项目(CSEEYESS-2021020)。

摘　　要：随着电网调度环境和信息整理环境日趋复杂,电网调控的难度也随之增加。针对深度强化学习技术具有有效感知复杂系统运行状态、适应性强、可扩展性好等特点,提出了基于深度强化学习的配网优化调度方法。构建了考虑源-网-荷-储的模拟建筑体配网模型,从原理出发对多智能体深度确定性策略梯度(MADDPG)算法进行静态优化,将模型与真实数据导入适用于电网级目标的多智能体强化学习框架中,尝试用优化后的算法对配网系统进行电压调控。结果表明,所用算法基本消除了配网系统的违规峰值电压,降低了总体电压偏差;优化后的多目标导向算法在保持电压稳定的同时减小了负载-发电功率差,使负载功率损耗维持较低水平,表明基于深度强化学习的建筑群柔性负荷优化调控方法具有一定有效性。The power grid dispatch environment and information organization environment have become more complex,and the difficulty of power grid regulation has gradually increased.Since deep reinforcement learning technology is of effective perception on complex system operation statuses,strong adaptability and good scalability,a distribution network optimization scheduling method based on deep reinforcement learning is proposed.Based on the simulated source-networkload-storage integrated distribution network model of a building complex,Multi-Agent Deep Deterministic Policy Gradient(MADDPG)algorithm was statically optimized from its principle.The model and real data were input into a multi-agent reinforcement learning framework suitable for grid-level objectives,and the optimized algorithm was tried to regulate the voltage of the distribution network system.The results show that the algorithm basically eliminates the abnormal peak voltages and reduces the overall voltage deviation.The optimized multi-objective oriented algorithm reduces the loadgenerated power difference while levelling the voltage off at a low level.The optimized control method for building complex flexible load based on reinforcement learning is proven to be effective.

关 键 词：微电网调控 能量管理 深度强化学习 确定性策略梯度 多目标优化 源网荷储 建筑群柔性负荷 

分 类 号：TK018[动力工程及工程热物理] TU18[建筑科学—建筑理论]