基于深度卷积神经网络的电力系统故障预测  

Power System Fault Prediction Based on Deep Convolutional Neural Network

作　　者：朱燕芳[1] 闫磊[1] 常康 赵文娜[1] 李远[1] 徐利美[1] ZHU Yanfang;YAN Lei;CHANG Kang;ZHAO Wenna;LI Yuan;XU Limei(Dispatching and Control Center,State Grid Shanxi Electric Power Company,Taiyuan 030002,China;School of Electrical Engineering and Automation,Harbin Institute of Technology,Harbin 150001,China;NARI Group Corporation(State Grid Electric Power Research Institute),Nanjing 211106,China)

机构地区：[1]国网山西省电力公司电力调度控制中心,太原030002 [2]哈尔滨工业大学电气工程及自动化学院,哈尔滨150001 [3]南瑞集团(国网电力科学研究院)有限公司,南京211106

出　　处：《电源学报》2024年第S01期179-185,共7页Journal of Power Supply

基　　金：国家重点研发计划资助项目(2017YFB0902200);国网山西省电力公司科技资助项目(WD160274)。

摘　　要：通过对深度卷积神经网络的深入研究,提出基于深度卷积神经网络的电力系统故障预测方法,保障系统安全运行。采用广域测量系统测量每个支路与节点,将获得的功率与关键特征值分别作为深度卷积神经网络模型输入、输出,训练这2个数据,并使用深度卷积神经网络AlexNet分析输入数据与输出数据的映射关系,建立基于深度卷积神经网络的电力系统故障预测模型,通过特征值分组、振荡模式筛选、数据预处理、模型训练和模型评估,实现电力系统运行状态评估,完成电力系统故障预测。实验结果说明:该方法的关键特征值计算结果与实际结果基本一致,可靠性高;使用正则化可提升模型泛化效果,防止模型过拟合;与其余方法的准确率和误报率指标相比,所提方法的准确率高达99.52%,误报率为1.16%,综合评价指标较高,评估效果优势显著。Based on the research of deep convolutional neural network,a fault prediction method for power system is proposed to ensure the safe operation of the system.Each branch and node are measured by a wide-area measurement system,and the obtained power and key eigenvalues are taken as input and output for the deep convolutional neural network model,respectively.The two data are trained,the deep convolutional neural network AlexNet is used to analyze the mapping relationship between the input and output data,and a power system fault prediction model based on deep convolutional neural network is established.Through eigenvalue grouping,oscillation mode screening,data preprocessing,model training and model evaluation,the power system operation state evaluation is realized,and the power system fault prediction is completed.Experimental results show that the results of key eigenvalues calculated by this method are basically consistent with the actual results,indicating a high reliability.Regularization can improve the model generalization effect and prevent the model from over-fitting.The accuracy of the proposed method is as high as 99.52%,and its false positive rate is 1.16%,indicating that the comprehensive evaluation index is high,and the evaluation effect has a significant advantage.

关 键 词：深度卷积 神经网络 电力系统 故障预测 AlexNet 

分 类 号：TM715[电气工程—电力系统及自动化]