基于机械比能的钻速预测模型优选

Optimization of ROP-Increase Prediction Model Based on Mechanical Specific Energy Theory

提高钻井效率是石油钻探过程中的重要环节之一,主要通过调整钻井参数、减小机械比能、提高机械钻速等方法实现。机械比能(MSE)表示钻头破碎单位体积岩石所需的机械能量,是评价钻井效率的主要指标之一,然而复杂的钻井作业导致大多数评价模型在应用过程中受到限制,地层不确定性会带来钻速预测模型泛化能力下降、非生产时间增加等问题。文章针对目标井的不同地层建立多种预测模型,经评估优选出表现最佳的模型。首先,基于机械比能理论和互信息法分析了影响钻速的可控参数;其次,以历史钻速均值为提速阈值将回归预测转换为分类预测,评估K最近邻(KNN)、多层感知机(MLP)、朴素贝叶斯(NB)、逻辑回归(LR)等分类算法模型的预测性能;最后优选出适用于目标井各地层的KNN模型,并将优选模型应用于同区域邻井中。实际验证结果表明:同区域邻井的四个地层预测准确率分别为0.94、0.94、0.92、0.96,AUC值分别为0.98、0.97、0.96、0.98,模型表现良好,能够助力钻井施工科学决策。

Improving drilling efficiency is an important part in the process of oil drilling.The optimization of drilling operation is mainly achieved by adjusting drilling parameters,reducing mechanical specific energy and increasing ROP.The specific mechanical energy(MSE)is one of the main indexes to evaluate drilling performance.However,complex drilling operations cause the limitations of most evaluation models in the application process,and formation uncertainty will lead to the problems such as reducing generalization ability of ROP prediction model and non-production time.In this study,multiple prediction models are established for different formations of the target well,the best performance model is selected by evaluation.Firstly,the controllable parameters affecting ROP are analyzed based on the theory of MSE and mutual information method.Secondly,the historical mean value is used as the ROP-increase threshold to convert the regression prediction into the classification prediction,and the prediction performance of classification algorithm models such as K nearest neighbor(KNN),multilayer perceptron(MLP),na ve Bayes(NB),and logistic regression(LR)is evaluated.Finally,the KNN model suitable for each formation of the target well is optimized,and applied to adjacent well in the same area.The actual verification results show that the prediction accuracy in the four formations of the adjacent well is 0.94,0.94,0.92,0.96,respectively,and the AUC value is 0.98,0.97,0.96,0.98,respectively.The model performs well and can help scientific decision-making of drilling construction.