基于强化学习的多园区综合能源系统经济调度  

Economic dispatch of multi-area integrated energy system based on reinforcement learning

作　　者：王丙文 付明 黄堃 WANG Bingwen;FU Ming;HUANG Kun(NARI Group Corporation,(State Grid Electric Power Research Institute),Nanjing 210000,China)

机构地区：[1]南瑞集团有限公司(国网电力科学研究院),南京210000

出　　处：《电测与仪表》2024年第9期32-39,共8页Electrical Measurement & Instrumentation

基　　金：国家重点研发计划项目(2018YFB0905000)。

摘　　要：多园区综合能源系统中新能源出力与负荷的波动性以及多种能量之间的耦合关系,给多园区综合能源系统的优化调度带来诸多挑战。为此提出一种基于数据驱动的多智能体近端策略优化(multi-agent proximal policy optimization,MAPPO)多园区综合能源系统经济调度方法。文章综合考虑园区间的能量交易与碳市场交易,以园区日运行成本最小为目标,建立多园区综合能源系统实时优化调度模型;将该优化问题建模为马尔科夫决策过程,并设计状态空间、动作空间以及奖励函数,通过大量历史数据的训练得到多园区综合能源系统优化调度神经网络模型,实现多园区分散式实时优化调度。仿真结果表明:在新能源出力与负荷随机性波动影响下,所提方法能够在降低各园区运行成本的同时减少园区间的信息交互,有助于提高各园区隐私信息的安全性。Due to the fluctuation of renewable energy output and load in multi-area integrated energy system,as well as the coupling relationship among multi-energy,it brings many challenges to the real-time optimal scheduling of multi-zone integrated energy system.To this end,this paper proposes a data-driven based multi-agent proximal policy optimization(MAPPO) algorithm for economic dispatch method of multi-area integrated energy system.Considering the energy trading and carbon market trading between areas,a real-time optimal scheduling model of multi-area integrated energy system is established to minimize the daily operating cost of the area.The optimization problem is modeled as a Markov decision process,and the state space,action space and reward function are designed.Through a large number of historical data training,the optimization scheduling neural network model of multi-area integrated energy system is obtained to realize multi-area decentralized real-time optimal scheduling.The results show that,under the influence of random fluctuations of new energy output and load,the proposed method can reduce the operating cost of each area,as well as the information interaction,which helps to improve the security of private information in each area.

关 键 词：多园区综合能源系统 实时经济调度 强化学习 多智能体近端策略优化 

分 类 号：TM73[电气工程—电力系统及自动化] TK018[动力工程及工程热物理]