基于Stacking集成学习的OLT流量预测系统研究

打开文本图片集

中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2026）05-0010-05

（中国电信股份有限公司新乡分公司，河南新乡453000）

Research on OLT Traffic Prediction System Based on Stacking Ensemble Learning

LIU Qing （XinxiangBranch，China Telecom CorporationLimited，Xinxiang 453ooo， China）

Abstract： Machine Learning algorithms process multidimensional data and present promising application prospects intrafic predictionat theOptical Line Trminal （OLT）.However，traditional MachineLearningalgorithms fail toachieve satisfactory prediction performance in practical applications.Toadressthis isse，thispaper proposesa fusion modelbased onthe Stackingensemble learning method.The modeladopts LightweightGradientBoosting trees （LGB），Extreme Gradient Boosting trees （XGB）andRandom Forests （RF）as base learmers，anduses Kemel Ridge Regresionas the meta-learner.It optimizes feature engineering througha historical shift plus sliding window approach，and thencomparesand analyzes the predictionresults ofsingle modelsand the Stacking leamer model.Italsodesigns anautomatedclosed-loopfeedback system. Compared with traditional Machine Learning methods，the proposed model reduces the Root Mean Square Error （RMSE）and Mean Absolute Error （MAE） in OLT traffic prediction by 26.59% and 25.62% respectively. This method is applicable to OLT traffic prediction and provides decision support for subsequent traffc expansion.

Keywords：Optical Line Terminal; traffic prediction; sliding window; Stacking ensemble learning

0 引言

随着数字化进程的深入发展，宽带用户的网络需求显著增长，运营商需对网络设备无源光网络（PassiveOpticalNetwork，PON）及OLT设备进行持续维护和优化[]。（剩余6140字）