论文部分内容阅读
针对智能调度监测中大量实时数据的快速处理问题，在实时流计算框架和并行拓扑编程基础上，研究一种配电网监测信息流计算的并行滑动窗口实时处理新方法。结合连续遥测对象的拓扑并行处理过程，利用并行滑动窗口的窗格聚集、窗格元组缓存与窗口聚集等设计技巧，实现监测信息流连续计算的快速滑动窗口拓扑实例。以一个配电网调度监控的工程系统为算例，对若干组拓扑实例进行快速滑动窗口的多节点集群测试，结果表明：集群环境下各元组可得到 ms级的窗口计算平均处理延时，合理设置拓扑工作进程的并行度和螺栓组件执行线程的并发数等并行参数有利于降低滑动窗口处理延时，验证了流计算拓扑的集群和并行度对提高智能配电网大规模连续监测信息的实时响应能力和处理效率具有重要意义。“,”In order to realize fast processing of massive real-time data in intelligent dispatching monitoring, a parallel sliding window stream computing real-time processing method for distribution network monitoring information is studied based on real-time stream computing framework and parallel topological programming. Combined with topology parallel processing procedure of continuous telemetry objects, fast sliding window topology instance for continuous computation of monitoring information flow is realized with sliding window design techniques, such as pane aggregation, pane tuple buffer and window aggregation. Taking engineering system of distribution power network dispatching monitoring as example, fast sliding window cluster tests for multiple sets of topology instances are performed. Results show that under cluster environment, each tuple gets millisecond-level average processing latency time of window computation, and window processing latency is effectively reduced by reasonable parallel parameter setting, including parallelism of work process and concurrent number of bolt component executor threads. It verifies that stream computing topology’s cluster and parallelism extension are important in improving real-time response ability and processing efficiency for large-scale continuous monitoring information flow in intelligent distribution network.