基于长短周期特征的用户异常行为检测

打开文本图片集

中图分类号：TP309 文献标志码：A 文章编号：1671-6841（2025）06-0065-09

DOI：10.13705/j. issn.1671-6841.2024077

Abstract：With the increasing number and types of users，the energy big data platform is now facing prominent internal security threats.User abnormal behavior detection is an effective technique to resist such security threats. However，current mainstream detection approaches did not take behavior pattrn of diferent types of users in the same platform and their long-term and short-term behavior characteristics into consideration，therefore leading to low user abnormal behavior detection performance.To solve these challenges，a method was proposed to extract the long-term and short-term behavior characteristics of different users in the energy big data platform.Specifically，the long short periods isolated forest model and the multiple time windows gate recurrent neural network were proposed to construct the long-term and short-term user behavior paterns respectively，and then the results of two models were effctively integrated for better detection ability.Moreover，an abnormal behavior detection framework was constructed with the consideration of diferent platform user types. Finally，the proposed framework was verified in a provincial energy big data platform，and the experimental results showed that our framework effctively characterized diffrent user behavior patterns in this platformand achieved a high accuracy of abnormal user behavior detection as well as high processing efficiency.

Key words： user behavior; abnormal behavior detection; long-term characteristics; short-term characteristics

0 引言

随着大数据技术在能源领域的不断深入，电力、燃气、石油等能源数据对国民经济发展的重要性日益凸显，因此能源大数据平台的建设受到了各级政府的高度关注。（剩余14868字）