基于云计算的电力设备监测数据的集中并行处理与诊断

发布时间：2018-02-15 07:37

本文关键词： 状态监测云计算并行计算局部放电故障诊断　出处：《华北电力大学(北京)》2017年博士论文　论文类型：学位论文

【摘要】：数据集成与信息共享是建设坚强智能电网的必然趋势。随着电网调度中心向调控一体化的发展,越来越多的电力设备状态数据将被送往调控中心。海量的历史状态数据和实时在线监测数据的快速处理正面临挑战,本文基于云计算和大数据处理技术对电力设备状态监测数据的集中并行处理与诊断问题展开了研究。设计了一种面向电力设备状态监测数据处理中多应用场景的综合云计算平台架构。在该平台下,多种并行计算框架共享同一套IT基础设施,既节约了计算平台投资与维护成本,又利于数据集成与信息共享。根据监测数据的处理模式和用户对实时性的需求,计算任务会以指定资源分配给最合适的计算框架。提出了一种局部放电信号基础放电参数的自适应提取算法,并在此基础上提出了基于Hadoop Map Reduce的大量局部放电信号的集中并行批处理方法。采用自适应的放电幅值阈值和放电间隔阈值对局部放电信号的局部极值点进行双重过滤,实现了放电次数、放电量与相位的自动化提取。借助Map Reduce计算框架,实现了局部放电信号从基础放电参数提取、谱图绘制与特征计算到类型识别整个分析过程的并行处理,提高了大数据量下的处理效率。提出了基于Spark内存计算技术的集合经验模态分解(EEMD)的并行化算法,弥补了Hadoop Map Reduce应用于复杂数据处理场景时的不足。通过对EEMD算法处理时序波形信号时的并行性分析,设计并实现了波形分段并行与EMD过程并行这两种不同结构的并行EEMD算法。将所提并行算法应用于局部放电波形信号的特征提取,并在计算性能上与串行EEMD算法和基于Hadoop Map Reduce的并行EEMD算法进行了比较。提出了一种基于Storm实时处理技术和变量预测模型分类(VPMCD)的电力设备在线并行故障诊断方法。为了满足在线诊断需求,将增量学习机制引入VPMCD中,采用递推最小二乘法实现了变量预测模型(VPM)的增量更新。考虑到大规模电力设备的在线故障诊断问题,设计了基于Storm的流式数据实时处理框架,通过构建Storm拓扑组件的监听机制实现了VPM在Storm上的初始化、增量学习与应用分类。以变压器油中气体分析为应用实例,测试了所提方法的分类性能和计算性能。
[Abstract]:Data integration and information sharing are the inevitable trend of building a strong smart grid. More and more power equipment status data will be sent to the regulatory center. Rapid processing of massive historical state data and real-time online monitoring data is facing challenges. Based on cloud computing and big data processing technology, the centralized parallel processing and diagnosis of power equipment state monitoring data are studied in this paper. A multi-application scenario for power equipment condition monitoring data processing is designed. Integrated cloud computing platform architecture. Under this platform, Many parallel computing frameworks share the same set of IT infrastructure, which not only saves the cost of investment and maintenance of computing platform, but also facilitates data integration and information sharing. The computing tasks are allocated to the most appropriate computing framework with the specified resources. An adaptive extraction algorithm for the fundamental discharge parameters of PD signals is proposed. On this basis, a centralized and parallel batch processing method for a large number of partial discharge signals based on Hadoop Map Reduce is proposed. The local extremum of partial discharge signal is filtered by adaptive discharge amplitude threshold and discharge interval threshold. The automatic extraction of discharge frequency, discharge quantity and phase is realized. With the help of Map Reduce calculation framework, the parallel processing of partial discharge signal from basic discharge parameters, spectrum drawing and feature calculation to type recognition is realized. In this paper, the processing efficiency of large amount of data is improved. A parallel algorithm of EMD based on Spark memory computing technology is proposed. It makes up for the deficiency of Hadoop Map Reduce in complex data processing scene. By analyzing the parallelism of EEMD algorithm when processing time series waveform signal, This paper designs and implements two parallel EEMD algorithms with two different structures: waveform piecewise parallelism and EMD process parallelism. The proposed parallel algorithm is applied to the feature extraction of PD waveform signals. Compared with serial EEMD algorithm and parallel EEMD algorithm based on Hadoop Map Reduce, a parallel fault diagnosis method for power equipment based on Storm real-time processing technology and variable prediction model classification is proposed. To meet the online diagnostic needs, The incremental learning mechanism is introduced into VPMCD, and the incremental updating of variable prediction model is realized by using recursive least square method. Considering the problem of on-line fault diagnosis of large-scale power equipment, a real-time processing framework of flow data based on Storm is designed. The initialization, incremental learning and application classification of VPM on Storm are realized by constructing the monitoring mechanism of Storm topology components. Taking gas analysis in transformer oil as an application example, the classification performance and computational performance of the proposed method are tested.
【学位授予单位】：华北电力大学(北京)
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TM507

【相似文献】