在石油钻进工程中,钻进速率(rate of penetration,简称ROP)是影响钻进效率和成本的重要指标.基于经验的ROP预测方法具有成本较高和效率较低的问题,为了解决这一局限性,本研究以澳大利亚昆士兰州的苏拉特盆地煤层气矿区的五口钻井为研究对象,在考虑数据是否完整和是否过滤的情况下,通过基于人工神经网络和极致梯度提升的监督式机器学习方法,利用其中三口钻井的钻进参数进行模型训练,并将训练后的模型应用于其他两口钻井的ROP预测来对比不同模型的预测优劣情况,从而获得较优的ROP预测方法.结果显示:人工神经网络和极致梯度提升在ROP预测中不同方面都显示出良好性能,其中人工神经网络在训练集的准确性上表现略优(R2=0.57),而极致梯度提升在测试集数据中有更好的性能.这为未来在石油钻进领域进一步利用机器学习技术进行ROP预测具有一定的指导意义.
Machine learning-based prediction of coal seam gas mining areas drilling rate:A case study of Surat basin in Australia
In petroleum drilling engineering,the rate of penetration(ROP)is a crucial indicator affecting drilling efficiency and cost.Empirical methods for predicting ROP are often costly and inefficient.To overcome these limitations,this study focuses on five wells in the coal seam gas mining area of the Surat basin in Queensland,Australia.Considering data completeness and filtering,the study utilizes supervised machine learning methods based on Artificial Neural Networks(ANN)and eXtreme Gradient Boosting(XGBoost).The drilling parameters from three of these wells were used for model training,and the trained models were then applied to predict the ROP for the remaining two wells to compare the predictive capabilities of different models,thereby identifying a superior method for ROP prediction.The results indicate that both ANN and XGBoost demonstrate good performance in various aspects of ROP prediction,with ANN slightly outperforming in training set accuracy(R2=0.57),while XGBoost shows better performance in the test dataset.This offers valuable guidance for the future application of machine learning technology in ROP prediction within the petroleum drilling field.
rate of penetrationcoal seam gasmachine learningANNXGBoost
国家科技期刊平台
NSTL
万方数据
