基于机器学习的电力调度主站SCADA系统告警信号自动识别  

Automatic Identification of Alarm Signals in SCADA System of Power Dispatch Master Station Based on Machine Learning

作　　者：章杜锡 胡铁军 管金胜 张霁明 ZHANG Du-xi;HU Tie-jun;GUAN Jin-sheng;ZHANG Ji-ming(State Grid Ningbo Electric Power Supply Company,Ningbo 315000 China)

机构地区：[1]国网宁波供电公司,浙江宁波315000

出　　处：《自动化技术与应用》2024年第3期31-34,共4页Techniques of Automation and Applications

摘　　要：为解决现有在线评估方法识别结果不准确的问题,提出基于机器学习的电力调度主站SCADA系统告警信号自动识别。先构建神经网络结构,采用反向传播方式训练神经网络,利用计算差值更新权值和偏置,由此完成攻击行为判断。将灰度图像二值化处理后,使用基于长方形的轮廓链代码滤除噪声。增强文本和背景的对比度,并设计告警界面文字分割及处理流程,使文本区域更加明显。采用告警窗口自动点对点接收方法,识别每一行警报窗口中的文本信息,完成告警信号识别。由实验结果可知,该方法最高查全率为96%,最高查准率为99.2%,可实现精准识别。In order to solve the problem of inaccurate identification results of existing online evaluation methods,an automatic identification of alarm signals in SCADA system of power dispatching master station based on machine learning is proposed.Firstly,the neural network structure is constructed,the neural network is trained by back propagation,and the weight and bias are updated by calculating the difference,so as to judge the attack behavior.After the gray image is binarized,the contour chain code based on rectangle is used to filter the noise.It enhances the contrast between text and background,and designs the text segmentation and processing flow of alarm interface to make the text area more obvious.The automatic point-to-point receiving method of alarm window is adopted to identify the text information in each line of alarm window and complete the identification of alarm signal.The experimental results show that the maximum recall rate of this method is 96% and the maximum precision rate is 99.2%,which can realize accurate recognition.

关 键 词：机器学习 主站SCADA系统 告警信号 自动识别 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TM76[自动化与计算机技术—控制科学与工程]