引用本文:陈艺璇,张孝顺,余涛.基于纳什均衡迁移学习的碳-能复合流自律优化[J].控制理论与应用,2018,35(5):668~681.[点击复制]
CHEN Yi-xuan,ZHANG Xiao-shun,YU Tao.Nash equilibrium inspired transfer learning for self-organizing optimal carbon-energy combined-flow[J].Control Theory and Technology,2018,35(5):668~681.[点击复制]
基于纳什均衡迁移学习的碳-能复合流自律优化
Nash equilibrium inspired transfer learning for self-organizing optimal carbon-energy combined-flow
摘要点击 1998  全文点击 746  投稿时间：2017-08-29  修订日期：2018-01-31
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2017.70612
  2018,35(5):668-681
中文关键词  纳什均衡解  碳排放责任分摊  分散自律  最优碳–能复合流  迁移学习  强化学习  电力系统
英文关键词  Nash equilibrium solution  shared responsibility of carbon emission  decentralized self-organization  optimal carbon-energy combined-flow  transfer learning  reinforcement learning  power system
基金项目  国家重点基础研究发展计划项目(“973”计划)(2013CB228205), 国家自然科学基金项目(51777078)资助.
作者单位E-mail
陈艺璇 华南理工大学电力学院 yxchen_diana@foxmail.com 
张孝顺* 华南理工大学电力学院 xszhang1990@sina.cn 
余涛 华南理工大学电力学院  
中文摘要
      提出了一种全新的纳什均衡迁移学习算法, 并应用于求解大规模电力系统分散式碳–能复合流自律优化. 首次引入碳排放责任分摊机制, 避免了碳排放责任的重复计算. 将大规模电网分解成若干小型区域电网, 每个小型 区域电网被定义为一个智能体, 通过纳什博弈实现分散式自律优化. 智能体利用记忆矩阵对寻优知识进行存储, 并 通过多个个体与环境的反复交互实现记忆更新; 采用状态–动作链对记忆矩阵进行降维, 有效避免了“维数灾难”. 此外, 基于相似度的迁移学习可以对历史任务知识进行高效提炼, 提高了新任务寻优效率. IEEE 57和300节点系统 仿真表明: 所提算法非常适合求解大规模电网的碳–能复合流自律优化, 在保证纳什均衡解质量的同时, 明显加快 寻优速度.
英文摘要
      This paper proposes a novel Nash equilibrium inspired transfer learning (NETL) for decentralized selforganizing optimal carbon-energy combined-flow of large-scale power systems. A shared responsibility of carbon emission is firstly considered, such that a double counting of carbon emission can be eliminated. Moreover, the whole power system is partitioned to the multiple subsystems, in which each subsystem is treated as an agent. The Nash game is introduced to satisfy the self-organizing optimal operation of each agent. Every agent stores its knowledge by the memory matrix, and a group of individuals is employed by agents to update their memories by interactions with the environment. The associated state-action chains are adopted to handle the curse of dimension. Transfer learning mechanism can refine the knowledge of the prior tasks efficiently thus dramatically accelerating the new tasks. The simulation on IEEE 57-bus system and IEEE 300-bus system verify that NETL is particularly geared to handle the self-organizing optimal carbon-energy combinedflow of large-scale power systems, which can ensure the quality of the Nash equilibrium solution as well as significantly accelerate the searching speed.