基于CatBoost-Adam结合的煤层气产量通用预测与优化方法研究

Research on the General Prediction and Optimization Method of Coalbed Methane Production Based on Catboost-adam Combination

在石油与天然气开采领域,压裂改造是必不可少的增加产量的方法,也是成本极高的工作。被压裂后的井的产量预测是衡量这口井压裂价值的重要指标,该指标受地质参数,测井参数和操作参数的综合影响,全部影响因素多达数十个,且数据方差大,处理难度高。针对这种情况,该研究首次用CatBoost对鄂尔多斯盆地临汾区块的5号和8号煤层进行特征筛选与产气量预测,并用适应性矩估计算法(Adam)对操作变量进行寻优。结果表明:1,CatBoost对低产井的预测精度为94%,对中产井的预测精度为86%,对高产井的预测精度为79%。经CatBoost初次训练得到煤层气显著影响因子为总液量、施工排量、电阻率、射孔深度、含气饱和度、孔隙度、全烃峰值、声波时差、射孔井段长、煤层厚度。运用Adam算法对该变量优化发现,最优排量为3.5 m3/min。保持排量不变的前提下提升总液量,有利于模型输出的提升。

In the field of oil and gas extraction,fracking is an essential way to increase production and a costly endeavor.The production prediction of the well after being fractured is an important index to measure the fracturing value of the well.This index is influenced by geological parameters,logging parameters and operational parameters comprehensively,with dozens of influencing factors,large data variance and high processing difficulty.In response to this situation,CatBoost was used for the first time to perform feature screening and gas production prediction for No.5 and No.8 coal seams in Linfen block,Ordos Basin,and adaptive moment estimation algorithm(Adam)was used to optimize the operating variables.The results showed that CatBoost predicted low-yield Wells with 94%accuracy,middle-yield Wells with 86%accuracy,and high-yield Wells with 79%accuracy.After CatBoost initial training,the significant influence factors of CBM were obtained as total fluid volume,construction displacement,resistivity,perforation depth,gas saturation,porosity,total hydrocarbon peak,acoustic time difference,perforation length,coal seam thickness.Using Adam algorithm to optimize the variable,it was found that the optimal displacement was 3.5 m^3/min.It is beneficial to improve the output of the model by increasing the total liquid quantity under the premise of keeping the displacement constant.