文章摘要
温可瑞,李卫东,张明泽,等.基于Markov决策过程的电池储能一次调频能量管理策略[J].电力系统自动化,2019,43(19):77-86. DOI: 10.7500/AEPS20190107001.
WEN Kerui,LI Weidong,ZHANG Mingze, et al.Energy Management Strategy of Battery Energy Storage System Participating in Primary Frequency Control Based on Markov Decision Process[J].Automation of Electric Power Systems,2019,43(19):77-86. DOI: 10.7500/AEPS20190107001.
基于Markov决策过程的电池储能一次调频能量管理策略
Energy Management Strategy of Battery Energy Storage System Participating in Primary Frequency Control Based on Markov Decision Process
DOI:10.7500/AEPS20190107001
关键词: 一次调频  电池储能系统  能量管理策略  Markov决策过程  辅助服务
KeyWords: primary frequency control  battery energy storage system  energy management strategy  Markov decision process  ancillary service
上网日期:2019-08-20
基金项目:国家自然科学基金资助项目(51677018);国家电网公司科技项目(2018ZX-14)
作者单位E-mail
温可瑞 大连理工大学电气工程学院, 辽宁省大连市 116024  
李卫东 大连理工大学电气工程学院, 辽宁省大连市 116024 wdli@dlut.edu.cn 
张明泽 大连理工大学电气工程学院, 辽宁省大连市 116024  
王振南 国网大连供电公司, 辽宁省大连市 116001  
吴港 国网大连供电公司, 辽宁省大连市 116001  
摘要:
      一次调频市场机制下的电池储能系统能量管理,需要在维持应对频率波动双向调节能力的基础上权衡运行成本和调频收益,以追求电池生命周期内的经济效益最大化。揭示了能量管理序贯决策本质上属于受控Markov过程,据此,通过频率响应需求动态转移的连续时间Markov链描述,以及基于生命周期吞吐量角度的储能电池容量动态衰退刻画,建立了以电池生命周期内经济效益期望值最大化为目标的Markov决策模型。针对运用标准迭代算法求解上述模型所面临的“维数灾”问题,提出了具有状态空间分解及后继状态辨识特征的降维并行值迭代(DRPVI)算法。算例结果表明:所得动态阈值结构能量管理策略可以显著提升储能经济效益,DRPVI算法能够有效缩减冗余计算,改善求解效率。
Abstract:
      For the energy management of battery storage system(BESS)in the primary frequency control(PFC)market, in order to maximize the economic benefits within battery life span, it is necessary to weigh operation costs and frequency control revenues on the basis of maintaining bidirectional frequency regulation capability. This paper reveals that the sequential decision of energy management is essentially a controlled Markov process. Therefore, the dynamic transfer of frequency response demand is described by continuous-time Markov chain, and the dynamic degradation of battery capacity based on lifecycle throughput is characterized. Then, a Markov decision model is built to maximize expected economic benefits within battery life span. Against the curses of dimensionality in solving the above model by using standard iterative algorithm, a dimensionality reduction parallel value iteration(DRPVI)algorithm with the characteristics of state space decomposition and subsequent state identification is proposed. Results show that the dynamic threshold strategy can significantly improve economic benefits, and DRPVI algorithm can effectively reduce redundant calculation and accelerate the efficiency of solution.
查看全文(Free!)   查看附录   查看/发表评论  下载PDF阅读器