基于云平台的电网设备监测井喷大数据的接收与并行分析

The condition monitoring of electrical primary apparatuses is being largely advanced in power utilities. With the dispatching center of a power grid being evolved towards to the integration of a power dispatch center and a remote control center, much more condition monitoring data of the apparatuses need to be sent to the center. But the present SCADA system is incapable to receive and process the huge data in time.To timely acquire the monitoring data of large-scale remote apparatuses in a power grid, the project is focus on the processing of the huge mass of blowout data faced by the monitoring system in extreme weather conditions. Therefore, based on Storm and Spark platforms, the architecture of the real-time cloud platform for large-scale electrical power equipment condition monitoring should be designed, the reliable receiving and rapid distribution on the monitoring data based on distributed technology is to be studied, the methods for the parallel processing and feature extraction on pieces of divided data of a monitoring signal waveform relating to a primary equipment are tried to be researched, the learning method under small sample set of a given parameter monitoring of a power equipment is to be found out, and the dynamic parallel learning, self-triggering and self-fusion of multiple classifiers for diagnosis need to be proposed.

目前，智能化电力一次设备和常规电力设备的在线监测都得到了较大发展。随着电网调度中心向调控一体化的发展，更多的电力设备状态数据将被送往调控中心，而现有的电网SCADA系统在这种巨量的监测数据面前显得难以应对，难以接入并及时分析这些数据。为实现电网中众多远方电力设备状态监测大数据的及时收集和诊断，本项目重点研究在极端天气下监测系统可能面临的井喷式报警数据处理问题，为此拟提出基于Storm和Spark的大规模电力设备状态监测实时云平台架构，研究基于分布式技术的监测大数据的可靠接收及快速分发，提出在Storm和Spark混合云平台下的输变电设备波形信号的分段并行分析和特征提取方法，找出电力设备在某种监测参数故障案例少的条件下的小样本诊断学习方法，并提出输变电设备诊断分类器的动态学习和自触发自融合方法。

随着智能电网的不断推进，监测数据变得日益庞大，电力设备在线监测系统面临海量数据（尤其极端天气下巨量井喷报警数据）及时处理的瓶颈问题。Storm云平台具有超低时延特性（毫秒级），对大量高并发实时或准实时数据的处理速度能控制在1秒以内（含数据分发时间），可作为大量监测和井喷报警数据的准实时处理平台；而Spark云平台的功能比较全面，可用于电力设备故障诊断模型的训练。本项目围绕构建基于实时大数据平台的新一代电力设备监测系统的关键技术开展了多方面研究，包括平台架构的设计、监测大数据可靠接收及其快速分发、电力设备波形信号的分段并行分析和特征提取、电力设备故障小样本识别方法、电网设备诊断分类器并行学习和多分类器融合，提出了一系列的解决方法，并从多角度进行了大量的仿真实验。理论研究和实验结果表明，本项目提出的基于Storm和Spark混合云平台构建新一代电力设备监测系统的设想是可行的，混合云平台对于上述海量监测数据有可接受的响应速度和处理能力；所提方法和理论研究成果能初步解决监控系统可靠收集、快速处理和诊断问题。