文章摘要
方睿,董树锋,唐坤杰,等.基于最大测点正常率与GPU并行加速的不良数据辨识方法[J].电力系统自动化. DOI: 10.7500/AEPS20181029003.
fang rui,dong shufeng,tang kunjie, et al.Bad Data Identification Method Based on Maximum Normal Measurement Rate and GPU Parallel Acceleration[J].Automation of Electric Power Systems. DOI: 10.7500/AEPS20181029003.
基于最大测点正常率与GPU并行加速的不良数据辨识方法
Bad Data Identification Method Based on Maximum Normal Measurement Rate and GPU Parallel Acceleration
DOI:10.7500/AEPS20181029003
关键词: 数据辨识  状态估计  测点正常率  GPU并行计算
KeyWords: data identification  state estimation  normal measurement rate  GPU parallel computing
上网日期:2019-06-11
基金项目:国家电网公司科技项目
作者单位E-mail
方睿 浙江大学电气工程学院 garrybest@foxmail.com 
董树锋 浙江大学电气工程学院 dongshufeng@zju.edu.cn 
唐坤杰 浙江大学电气工程学院 tangkunjie1994@163.com 
朱承治 国网浙江省电力公司 chengzhi_zhu@163.com 
裴湉 国网浙江省电力公司 welcome_pt@163.com 
宋永华 澳门大学电机及电脑工程系 yhsongcn@zju.edu.cn 
摘要:
      基于测量不确定度的概念,以测点正常率最大(maximum normal measurement rate,MNMR)为目标的电力系统抗差状态估计方法具有较好的不良数据辨识能力。然而,该模型求解困难,已有研究对该模型进行了近似等效,并采用现代内点法进行求解,但存在因近似而辨识效果降低的问题。为此,本文基于MNMR状态估计模型,采用杂交变异粒子群(particle swarm optimization,PSO)算法,提出一种基于GPU并行加速的不良数据辨识算法。该算法不对MNMR模型进行近似等效,并根据GPU并行计算架构特点,设计了粗粒度和细粒度结合的并行加速策略。算例结果表明,本文所提的算法对不良数据的误检率和漏检率较低,具有较好的不良数据辨识能力,且计算时间短,加速效率高,能够满足实际运行需求。
Abstract:
      Based on the concept of uncertainty of measurement, the power system robust state estimation method with maximum normal measurement rate (MNMR) has good bad data identification ability. However, the model is difficult to solve. Research has been proposed that the model can be approximated and solved by the modern interior point method. However, there are problems such as lower identification effect due to approximation. Therefore, based on the MNMR state estimation model, a hybrid particle swarm optimization (PSO) algorithm with gaussian mutation is used to propose a bad data identification algorithm based on GPU parallel acceleration. The algorithm does not approximate the MNMR model, and according to the characteristics of the GPU parallel computing architecture, a parallel acceleration strategy combining coarse and fine granularity is designed. The results of the case analysis shows that the proposed algorithm has low false detection rate and missed detection rate for bad data, and has good bad data identification ability, short calculation time and high acceleration efficiency, which can meet the actual operation requirements.
查看全文   查看附录  查看附录   查看/发表评论  下载PDF阅读器