强化Q学习——模糊多目标AGC调节容量动态优化分配策略的研究
Dynamic Optimized Allocation Policy for AGC Regulation Capacity Based on Reinforcement Q-Learning and Fuzzy Multiple Objectives
钱虹 1姚一鸣 2陈丹 3费敏锐4
作者信息
- 1. 上海电力学院,上海 200090;上海大学,上海 200072
- 2. 上海电力学院,上海 200090
- 3. 北京工商大学,北京 100037
- 4. 上海大学,上海 200072
- 折叠
摘要
针对AGC调节容量多目标优化分配存在着无法完全描述机组特性的问题,通过研究提出采用基于强化Q学习的优化分配方法.将AGC系统视为“不确定的随机系统”,结合ACE的调节死区以及CPS评价标准,建立了AGC调节容量优化分配问题的马尔可夫决策过程模型,并引入Q学习方法对MDP的最优值函数进行学习,仿真结果表明,该强化Q学习-模糊多目标AGC调节容量动态优化分配策略能够适应电网环境变化的要求.
Abstract
The unit characteristics can't be fully described in the multi-objective optimized allocation of AGC regulation capacity; therefore,this paper proposes the optimal allocation approach based on the reinforcement Q-learning.With AGC system,the "uncertain stochastic system",combined with the ACE adjusting dead zone and CPS evaluation standard,the Markov decision process (MDP) model is established for optimized allocation of AGC regulation power; the Q-learning approach is introduced to study MDP's optimal value function.The simulation results show that the proposed allocation strategy can adapt to the changing requirement of grid environment.
关键词
自动发电控制/强化Q学习算法/调节容量动态优化分配Key words
AGC/reinforcement Q-learning algorithm/dynamic optimized allocation of regulation capacity引用本文复制引用
出版年
2014
NETL
NSTL
万方数据