| 引用本文: | 余涛,王宇名,甄卫国,叶文加,刘前进.基于多步回溯Q学习的自动发电控制指令动态优化分配算法[J].控制理论与应用,2011,28(1):58~64.[点击复制] |
| YU Tao,WANG Yu-ming,ZHEN Wei-guo,YE Wen-jia,LIU Qian-jin.Multi-step backtrack Q-learning based dynamic optimal algorithm for auto generation control order dispatch[J].Control Theory and Technology,2011,28(1):58~64.[点击复制] |
|
| 基于多步回溯Q学习的自动发电控制指令动态优化分配算法 |
| Multi-step backtrack Q-learning based dynamic optimal algorithm for auto generation control order dispatch |
| 摘要点击 2993 全文点击 2232 投稿时间：2009-11-09 修订日期：2010-03-23 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 |
| 2011,28(1):58-64 |
| 中文关键词 Q(λ)学习 自动发电控制 控制性能标准 随机最优 调节费用 |
| 英文关键词 Q(λ) learning automatic generation control CPS stochastic optimization AGC production cost |
| 基金项目 国家自然科学基金资助项目(50807016); 广东省自然科学基金资助项目(9151064101000049). |
|
| 中文摘要 |
| 单步Q学习在火电占优、机组时延较大的自动发电控制(AGC) 功率指令动态优化分配中的应用表现出收敛速度慢等不足而影响最优策略的获取. 具有多步预见能力的多步回溯Q学习(Q(λ))显式利用资格迹进行高效回溯操作, 能够有效解决火电机组大时滞环节带来的延时回报问题, 算法平均收敛时间较Q学习缩短50%以上. 算法奖励函数引入调节费用一项, 形成多目标动态最优控制. 两区域模型及南方电网模型仿真研究分析显示, Q(λ)算法在随机、大负荷扰动的复杂系统环境中有效提高系统控制性能标准(CPS)控制品质和适应性, 并且在保证CPS合格率的前提下, 使AGC调节费用下降超过5%. |
| 英文摘要 |
| This paper presents the application of multi-step backtrack Q(λ) learning-based methodology on CPS order dynamic dispatch problem. The proposed Q(λ) learning can effectively solve the long time-delay assessment for the action strategy of one step Q-learning in the thermal dominated power system. AGC production cost is formulated as Markov decision process(MDP) reward function by means of linear weighted aggregative approach in the CPS order multiobjective dynamic optimal dispatch. Simulation of institute of electrical and electronics engineers(IEEE) two-area LFC model shows that the convergence time of the Q(λ) algorithm is reduced by more than 50% comparing with Q-learning. The statistical experiments of Q(λ) in the China Southern Power grid show that the proposed method can effectively enhance the robustness and dynamic performance of AGC systems in CPS assessment and save more than 5% of AGC production cost while the CPS compliances are ensured. |