Accelerating bio-inspired optimizer with transfer reinforcement learning for reactive power optimization

详细信息    查看全文

• 作者：Xiaoshun Zhang^a ; ^{xszhang1990@sina.cn} ; Tao Yu^a ; ^{taoyu1@scut.edu.cn} ; ^{yutao99@tsinghua.org.cn} ; Bo Yang^b ; ^{yangbo_ac@outlook.com} ; Lefeng Cheng^a ; ^{chenglf_scut@163.com}
• 关键词：Accelerating bio-inspired optimizer ; Transfer reinforcement learning ; Memory matrix ; Cooperating multi-bion ; WoLF-PHC ; Reactive power optimization
• 刊名：Knowledge-Based Systems
• 出版年：2017
• 出版时间：15 January 2017
• 年：2017
• 卷：116
• 期：Complete
• 页码：26-38
• 全文大小：1794 K
• 卷排序：116

文摘

This paper proposes a novel accelerating bio-inspired optimizer (ABO) associated with transfer reinforcement learning (TRL) to solve the reactive power optimization (RPO) in large-scale power systems. A memory matrix is employed to represent the memory of different state-action pairs, which is used for knowledge learning, storage, and transfer among different optimization tasks. Then an associative memory is introduced to significantly reduce the dimension of memory matrix, in which more than one element can be simultaneously updated by the cooperating multi-bion. The win or learn fast policy hill-climbing (WoLF-PHC) is also used to accelerate the convergence. Thus, ABO can rapidly seek the closest solution to the exact global optimum by exploiting the prior knowledge of the source tasks according to their similarities. The performance of ABO has been evaluated for RPO on IEEE 118-bus system and IEEE 300-bus system, respectively. Simulation results verify that ABO outperforms the existing artificial intelligence algorithms in terms of global convergence ability and stability, which can raise one order of magnitude of the convergence rate than that of others.