基于概率模型和主成分分析的多变量时序数据网络监测框架

打开文本图片集

中图分类号：TP309 文献标志码：A 文章编码：1672-7274（2026）02-0008-04

A Multi-Variable Time Series Data Network Monitoring Framework Based on Probabilistic Models and Principal Component Analysis

DINGJu （ShanghaiMunicipalPublic SecurityBureau，Shanghai 2oooo1，China）

Abstract： Inactual scenarios，time series data is oftenaffected by noise such assensor errors and environmental disturbances，resulting inarelativelyhigh false alarmrate innetwork monitoring.To thisend，amulti-variable time series data network monitoring framework based on probabilistic models and principal component analysis is designed.Theslidingwindowalgorithm isadoptedtodividetheoriginaldataset，and principalcomponentanalysis is utilized totransformthehigh-dimensionaldatasetintothelow-dimensionalspace.One-dimensionaldilatedconvolutionis usedtoextracttmporalfeaturesatdiferentlevels，andtheweightsareadaptivelyadjustedforfusion.Constructthenetwork timeseries matrix basedonthe probabilistic model.A monitoringframework is constructed usingLSTMtoachieve effective monitoring of multivariable time series data.The experimental results show that the average lossvalue of the proposed method is O.296，and the error reduction rate is 4.40% ，which reduces the monitoring error to a certain extent.

Keywords： probability model; principal component analysis; multivariate time series data; network monitoring

0 引言

在公安系统网络安全工作中，对多变量时间序列数据进行实时监控和分析，是识别网络攻击、防止数据泄露、维护社会稳定的核心任务[1]。（剩余3598字）