基于RGPSO-LightGBM的套管磨损深度预测

打开文本图片集

Qin Yanbin，Wang Jian，Wan Zhiguo，et al.Prediction of casing wear depth based on RGPSO-LightGBM[J].Chi-naPetroleumMachinery，2025，53（5）：139-146.

关键词：套管磨损深度；井筒完整性；LightGBM；粒子群优化；机器学习中图分类号：TE931 文献标识码：ADOI：10.12473/CPM.202404075

Prediction of Casing Wear Depth Based on RGPSO-LightGBM

Qin Yanbin 1，2 Wang Jian'Wan Zhiguo1，²Li Linlin³Dou Yihua 1，2 （1.CollegeofMechancalEngnering，Xi'anShiyou University；2.Xi'anKeyLaboratoryofWelboreIntegrityEvaluaion；3.Well Testing Branch of CNPC Bohai Drilling Engineering Company Limited）

Abstract： Traditional casing wear prediction models fail to achieve satisfactory accuracy under ideal assumptions，and the derivation method relying on test data is also time-consuming and costly.This paper presents a casing weardepth prediction model based onreactive global particle swarm optimizationand lightweight gradient boosting machine （RGPSO-LightGBM）.First，the Pearson corelation coefficient method and feature importance were used to analyze the report dataof the multi-arm caliper imaging logging tool and the dilling logs and extract key feature values.Then，the LightGBM was used to predict the wear depth，and RGPSO was combined for global optimization on multiple hyperparameters of LightGBM.Finally，the RGPSO-LightGBM model was compared with the BP neural network （BPNN）and extreme gradient boosting（XGBoost）models.The results show that the RGPSOLightGBM model yields the highest goodness of fit （ R2 ）up to O. 997 6， indicating better prediction accuracy，robustness and generalization.The research results provide effective basis for inteligent control of subsequent oil and gas well production，and areof great practical significancefor maintaining welbore integrity and ensuring safe production operations of oil and gas wells.

Keywords： casing wear depth； wellbore integrity ； LightGBM； PSO； machine learning

0引言

随着传统的浅层油气储备逐渐枯竭，我国推动了对深层油气资源的关注和探索。（剩余11754字）