基于Blending模型融合的抽油機井檢泵周期預(yù)測方法
打開文本圖片集
中圖分類號：TE933文獻標識碼：ADOI：10.12473/CPM.202405015
Jiang Minzheng，Zhang Qi，Wang Xinmin，et al.Pump detection period prediction of pumping well using blending ensemble model [J]．China Petroleum Machinery，2025，53（5）：10-17.
Pump Detection Period Prediction of Pumping Well Using Blending Ensemble Model
Jiang Minzheng'Zhang Qi1Wang Xinmin2Meng Bo'Zhou Yufeng1Dong Kangxing' （1.SchlofMechnicalScienceandEnginering，NorthstPetroemUniersity；aqngOilfeldProductionTecholoIsiute）
Abstract：A single model for predicting the pump detection period of pumping well is low in stability and accuracy.For improved prediction accuracy，a Blending ensemble model was proposed by efectively combining RF， GBDT，XGBoost and LightGBM algorithms.The LOF isolation detection method and normalization were used to preprocess the historical pump detection data from a block of Daqing Oilfield.A tree model based feature fusion screening method was used to screen out the main influencing parameters，and each of the above-mentioned four algorithms was compared with the Blending ensemble model forthe prediction accuracy.Finally，15O sets of new pump detection data were used to verify the prediction accuracy and generalization performance of the Blending ensemble model.The results show that the Blending ensemble model yields a greatly improved performance，with a goodness offit determination coefficient of 0.954.The model verificationusing 15O sets of new pump detection data demonstrates the goodnessof fit determination coeffcient of 0.947.Thus，the Blending ensemble method is verified effective and feasible.The research results provide reference for the production and management of oilfields.
Keywords：pump detection period prediction；Blending ensemble model； normalization；parameter optimization；field verification
0引言
有桿泵開采是目前世界上使用最普遍的一種采油方式，在中國使用有桿泵開采的井約占總井數(shù)的90%[1] 。（剩余12156字）