一种基于卷积神经网络的配电网主站告警窗口的图像识别方法

An Image Identification Method for the Alarm Window of Distribution Network Master Station Based on CNN

随着配电网拓扑结构与设备规模的不断扩大,配电网主站中集聚了海量告警信息,亟需高效计算能力进行识别与分析。以配电网信息物理主站系统的强大硬件系统为依托,基于卷积神经网络的先进事件处理能力,开发了告警图像智能分析技术,实现告警信息在线综合处理、显示与推理,支持汇集和处理。以实际主站告警业务为例,对大量告警信息进行精准识别,实现对图像数值分析、阈值对比、趋势分析,对异常状态进行自动预警,实时对电网运维人员进行故障告警推送。基于实际配电网实景场景验证了设备识别与缺陷识别的准确性,实现了配网运行监控、故障处理的一体化运行。

As the topology and equipment scale of the distribution network continue to expand, a huge amount of alarm information is gathered in the distribution network master station, which requires efficient computing power for identification and analysis. Based on the powerful hardware of the distribution network information physical master station system and the advanced event processing capability of convolutional neural network, an alarm image intelligent analysis technology is developed to realize online integrated processing, display and reasoning of alarm information, supporting aggregation and processing. Taking the actual main station alarm service as an example, it accurately identifies a large amount of alarm information, realizes numerical analysis of images, threshold comparison, trend analysis, automatic warning of abnormal states, and real-time fault alarm pushing to grid operation and maintenance personnel. The accuracy of equipment identification and defect identification is verified based on actual distribution network real-world scenarios, showing that the integrated operation of distribution network operation monitoring and fault handling is realized.