摘要
电子信息与通信技术推动了电力系统的更新换代,信息技术给电力系统带来发展的同时,潜在的信息安全问题也逐渐严峻。研究提出了基于深度自动编码器的量测篡改攻击检测方法和基于同态密码学的隐私保护估计框架。仿真结果显示,随阈值增大,算法精确率逐渐稳定在90%左右,而准确率、召回率和F1值则呈下降趋势,其中F1值最大为91.07,对应最优阈值为8。算法准确率为0.927,精确率为0.934,召回率为0.847,F1值为0.912,四项性能均表现较好,提升幅度明显。IEEE9节点系统的正常量测误差均值为2.367,异常量测误差为22.781,显著的差异可明显区分出异常量测的重构误差。当密钥为1536时,运行效率仍然较高,且均方根误差大小与实际工程需要大小差值处于合理范围之内,有效保证了电力系统信息的安全性。
Electronic information and communication technologies have driven the renewal of power systems.Information technology has brought development to power systems while the potential information security problems have become progressively more severe.The study proposes a method for detecting quantitative tampering attacks based on a deep auto-encoder and a framework for estimating privacy protection based on homomorphic cryptography.The simulation results show that the accuracy rate of the algorithm gradually stabilises at around 90%as the threshold increases,while the accuracy rate,recall rate and F1 value show a decreasing trend,with the maximum F1 value being 91.07,corresponding to the optimal threshold of 8.The accuracy rate of the algorithm is 0.927,the accuracy rate is 0.934,the recall rate is 0.847 and the F1 value is 0.912,with all four performance items performing well and improving significantly.The mean normal measurement error for the IEEE9 node system was 2.367 and the abnormal measurement error was 22.781,with significant differences clearly distinguishing the reconfiguration error of the abnormal measurements.When the key is 1536,the operating efficiency is still high and the root mean square error size meets the actual engineering needs,effectively guaranteeing the safe operation of the power system.
作者
刘柯余
范永学
李雨
南颖
马方远
LIU Keyu;FAN Yongxue;LI Yu;NAN Ying;MA Fangyuan(State Grid Sichuan Procurement Company,Chengdu 610000,China;Beijing Guodiantong Network Technology Co.,Ltd.,Beijing 100071,China)
出处
《自动化与仪器仪表》
2024年第2期91-95,共5页
Automation & Instrumentation
基金
《多稳定约束下电网临界切除时间批量计算方案研究》(SGNXDK00DWJS2000171)。
关键词
智能电网
深度学习
攻击
电力隐私
密码学
power systems
deep learning
attacks
power privacy
cryptography
