统计与决策2025，Vol.41Issue(24)：58-63,6.DOI:10.13546/j.cnki.tjyjc.2025.24.010
基于有监督神经网络的CUSUM在线变点检测方法及应用
Supervised Neural Network-based CUSUM Online Change-point Detection Method and Its Application
摘要
Abstract
This paper mainly investigates the online change-point detection problem,and proposes a detection framework that integrates supervised neural networks with the traditional CUSUM method.The approach employs a neural network classifica-tion model to capture complex characteristics of data distribution shifts and introduces a DBSCAN clustering strategy based on the silhouette coefficient and Davies-Bouldin index to mitigate the impact of data imbalance on detection performance.On this basis,the cumulative duration of streaming data is dynamically adjusted to further optimize the real-time performance and accuracy of the detection process.An empirical study based on NOAA meteorological data(2012-2023)demonstrates that the proposed mod-el significantly outperforms the conventional CUSUM method in terms of detection accuracy and time error control,achieving an 83%reduction in maximum time error and a 52%decrease in false detections.For the detection of high-temperature and precipi-tation events,optimal performance is achieved under data windows of 72 hours and 36 hours,respectively.关键词
CUSUM变点检测/有监督神经网络/在线检测/气象监测Key words
CUSUM change-point detection/supervised neural network/online detection/meteorological monitoring分类
数理科学引用本文复制引用
Li Yufeng..基于有监督神经网络的CUSUM在线变点检测方法及应用[J].统计与决策,2025,41(24):58-63,6.基金项目
国家社会科学基金重大项目(21&ZD153) （21&ZD153）
成都市哲学社会科学研究中心咨政服务能力建设专项重点项目(2024-35) （2024-35）
重庆市高校网络舆情与思想动态研究咨政中心科研创新项目(23yqzxxs005) （23yqzxxs005）
