基于多智能体强化学习的多园区综合能源系统协同优化运行研究  

Research on cooperative optimal operation of multi-park integrated energy system based on multi agent reinforcement learning

作　　者：杨照 黄少伟[1] 陈颖[1,2] YANG Zhao;HUANG Shao-wei;CHEN Ying(Department of Electrical Engineering,Tsinghua University,Beijing 100084,China;New Energy(Photovoltaic)Industry Research Center,Qinghai University,Xining 810016,China)

机构地区：[1]清华大学电机工程与应用电子技术系,北京100084 [2]青海大学新能源光伏产业研究中心,青海西宁810016

出　　处：《电工电能新技术》2021年第8期1-10,共10页Advanced Technology of Electrical Engineering and Energy

基　　金：青海省科技计划项目(2018-ZJ-748)。

摘　　要：多园区综合能源系统协同优化运行能充分发挥多能耦合的灵活性,降低系统运行成本,但多主体利益分配问题、隐私保护需求及多重不确定量的存在给多园区协同优化运行带来了巨大挑战。为此,本文建立了一个多园区协同优化运行架构,并采用多智能体深度确定性策略梯度算法进行求解。仿真结果表明,无论是面对确定性场景还是不确定性场景,本文所提方法均能在保护各园区隐私的前提下,降低各园区的运行成本,且该方法不依赖于对不确定量的准确预测,可应用于实时调度中。The collaborative optimization operation of the multi-park integrated energy system can give full play to the flexibility of multi-energy coupling and reduce the operation cost of the system.However,the distribution of benefits of multiple entities,the need for privacy protection,and the existence of multiple uncertainties have brought huge challenges to the collaborative optimization operation of the multi-park integrated energy system.To this end,this paper establishes a multi-park collaborative optimization operation architecture,and uses the Multi-Agent Deep Deterministic Strategy Gradient Algorithm(MADDPG)to solve it.The simulation results show that the proposed method can reduce the operation cost of each park under the premise of protecting the privacy of each park regardless of whether it is facing deterministic scenarios or uncertain scenarios.And this method does not rely on the accurate prediction of the uncertainty variables,so it can be applied to real-time operation scheduling.

关 键 词：多园区综合能源系统 协同优化运行 MADDPG 

分 类 号：TM732[电气工程—电力系统及自动化]