基于贝叶斯优化LightGBM算法的主动式搜索时间调整方法  

An Active Search Time Tuning Model Based on the Bayesian Optimization LightGBM Algorithm

作　　者：刘青[1] 鲁成 马天祥 段昕 LIU Qing;LU Cheng;MA Tianxiang;DUAN Xin(State Key Laboratory of Alternate Electrical Power System with Renewable Energy Sources(North China Electric Power University),Baoding 071003,China;Power Research Institute of State Grid Hebei Electric Power Co.,Ltd,Shijiazhuang 050021,China)

机构地区：[1]新能源电力系统国家重点实验室(华北电力大学),河北保定071003 [2]国网河北省电力有限公司电力科学研究院,河北石家庄050021

出　　处：《华北电力大学学报（自然科学版）》2024年第1期30-38,48,共10页Journal of North China Electric Power University：Natural Science Edition

基　　金：河北省省级科技计划项目(20314301D);国家电网公司科技项目(kj2020-064)。

摘　　要：针对5G配电终端延迟波动较大导致保护闭锁的问题,提出一种基于贝叶斯优化LightGBM算法的主动式搜索时间调整方法。该方法以无线通信终端延迟波动历史数据以及温度、日期时间等特征变量为输入,对延迟进行预测并动态调整设备参数。首先,对原始特征变量进行特征工程预处理,然后同历史延迟数据一并通过LightGBM算法进行数据的拟合,在训练过程中引入贝叶斯优化算法进行参数寻优,并利用最终加权组合,结合终端实时监测延迟进行预测值的调整,最终实现5G终端延迟的高精度预测。以河北南网某5G配网试点的数据进行训练与验证,结果表明所提方法能有效实现延迟预测,较随机森林回归,XGBoost等算法有更高的预测精度。A proactive search time adjustment model based on the Bayesian optimization LightGBM algorithm was proposed for the problem of large delay fluctuations in 5G distribution terminals leading to protection blocking.The model used historical data on delay fluctuations of wireless communication terminals and feature variables such as temperature and date and time as inputs to predict delays and dynamically adjust equipment parameters.Firstly,the original feature variables were pre-processed via feature engineering,and then the data were fitted together with the historical delay data by the LightGBM algorithm.Secondly,a Bayesian optimization algorithm was introduced for parameter search during the training process,and the final weighted combination was used to adjust the predicted values in combination with the real-time monitoring delay of the terminal.Finally,the high-precision prediction of 5G terminal delay is achieved.We conducted training and validation with data from a 5G distribution network pilot in the southern network of Hebei province.The results show that the proposed method can effectively achieve delay prediction and has higher prediction accuracy than random forest regression and XGBoost algorithms.

关 键 词：贝叶斯优化 机器学习 5G通信 馈线自动化 延迟波动 短期预测 

分 类 号：TM73[电气工程—电力系统及自动化]