Research On Selection And Production Parameters Prediction Of Bailing Well Based On Data Drive

Bailing production method,also known as bailing production,is an economic and feasible production method aimed at remote and scattered wells,wells with ultra-low permeability,insufficient fluid supply and long shutdown.Under the current low oil price environment,some ultra-high water cut oil fields in China have also explored ways of bailing production,but there is no discriminant standard and production performance prediction method for converting conventional mechanical production wells into bailing wells.To solve these problems,this article adopts the method of data mining and machine learning,based on the dynamic and static data of reservoir,build selection model for converting mechanical production wells into bailing wells and bailing well production index prediction model,guide bailing well selection and dynamic prediction,to realize the data mining and machine learning technology application in the field of oil and gas field development.The research contents are mainly divided into four aspects:(1)Analyzing the production characteristics of bailing wells and determining the characteristic indexes of production effect of bailing wells;Based on reservoir engineering theory,multi-source data were fused to analyze the influencing factors of bailing effect from six aspects: physical property parameters,pressure and flow field,structural type,residual potential,seepage environment,and injectionproduction connectivity characteristics.Pearson correlation analysis was used to screen out the sensitivity factors,and the sample library of bailing well screening model was established.(2)Entropy value method and comprehensive scoring method were used to calculate and determine the limit comprehensive evaluation value of bailing production effect;Based on the sample library of bailing well screening model,a bailing well screening model was established by using multiple linear regression method and combined with the comprehensive evaluation value of bailing well production limit,and the potential replacement wells were selected from the target block by using the model.Based on the historical dynamic and static data of bailing well,the favorable conditions for bailing production are summarized through reservoir engineering analysis.(3)The sample database was established in the initial stage and the later stage of bailing production,and the monthly liquid production of bailing was determined as the forecast production index;through data cleaning and characteristic parameter processing,the characteristic parameters that affect the monthly fluid production in the early and late stages of bailing well were obtained.Pearson and Spearman correlation analysis were used to screen out the sensitive characteristic parameters of monthly fluid production in the early and late stages of bailing well and form the sample library of production indicators in the early and late stages of bailing well.(4)Based on early and late learning samples established for bailing well,using gradient boost decision tree and gated recurrent unit neural network respectively,the model is trained and parameter optimization,set up for bailing well in the early and later production index prediction model,using cyclic update methods established for multi-month fluid production prediction model at the later stage of bailing production,realize the historical and potential bailing wells dynamic prediction.The research results show that the bailing well selection model and production index prediction model established in this paper have the characteristics of comprehensive consideration of dynamic and static production parameters and strong feasibility.Through field application,11 potential bailing wells were screened out from low production and low efficiency and closed wells in the target block.The average relative error of production index prediction model test at the initial stage of bailing production is 9.70%.The average relative error of the model for predicting the production index in the later stage of bailing production is11.85%.When it is applied to the prediction of multi-month bailing fluid production in the later stage of historical bailing well,the average relative error is all less than 10%.The prediction effect is good and meets the application requirements of the mine.