基于生存分析的智能电网安全告警事件持续时间预测模型
PREDICTION MODEL OF SMART NETWORK SECURITY ALARM EVENT DURATION BASED ON SURVIVAL ANALYSIS
刘萧 1李静 1许珂1
作者信息
- 1. 国网四川省电力公司信息通信公司 四川成都 610041
- 折叠
摘要
针对目前智能电网安全风险预测系统解释性不强的问题,提出一种基于生存分析中DeepSurv模型的改进模型来预测网络安全告警事件持续时间.为了加快运算速度,模型对原DeepSurv的神经网络部分进行改进.模型基于K-means,提出一种降维算法对输入数据进行降维,通过改进的DeepSurv,获得智能电网安全告警事件持续时间的生存函数,并以此为依据计算C-index与MAPE.通过总耗时、C-index与MAPE这三个指标将模型和原DeepSurv进行比较,发现模型在预测准确率相差不大的情况下,大幅提高了运算速度.此外,由于模型是基于生存分析的,解释性较强,且能够提供网络安全告警事件关于持续时间的生存函数,即关于持续时间的概率预测,对智能电网安全风险预测研究有很大的参考意义.
Abstract
To solve the problem that the current smart electric network security prediction system is not sufficiently interpreted,an improved model based on the DeepSurv model in survival analysis is proposed to predict the duration of network security alarm events.In order to speed up the operation speed,the neural network part of the original DeepSurv is improved.Based on K-means,a dimensionality reduction algorithm was proposed to reduce the input data.Through the improved DeepSurv,the survival function of the duration of the smart electric network security alarm event was obtained,and the C-index and MAPE were calculated on this basis.By comparing the model with the original DeepSurv in terms of total time consuming,C-index and MAPE,it is found that the model significantly improved the operation speed while the prediction accuracy was not much different.In addition,as the model is based on survival analysis,it has strong explanatory power and can provide the survival function of network security alarm event on duration,namely,probability prediction on duration,which is of great reference significance for the research of smart electric network security risk prediction.
关键词
智能电网/网络安全告警事件/生存分析/神经网络/持续时间预测Key words
Smart electric network/Network security alarm events/Survival analysis/Neural network/Prediction of duration引用本文复制引用
基金项目
国网四川省电力公司科技项目(SGSCXT00XGJS2000191)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据