| 基于融合特征选择算法的钻速预测模型研究 |
| 投稿时间：2022-04-25 修订日期：2022-06-17 点此下载全文 |
| 引用本文：周长春,姜杰,李谦,等.基于融合特征选择算法的钻速预测模型研究[J].钻探工程,2022,49(4):31-40.ZHOU Changchun,JIANG Jie,LI Qian,et al. Research on drilling rate prediction model based on fusion feature selection algorithm[J]. Drilling Engineering, 2022,49(4):31-40. |
| 摘要点击次数: 831 |
| 全文下载次数: 804 |
|
| 基金项目:中海石油（中国）有限公司项目“南海西部油田上产2000万方钻完井关键技术研究”子课题“乐东10区超高温高压井综合提速技术研究”（编号：CNOOC-KJ135ZDXM38ZJ05ZJ）；四川省科技支撑计划应用基础研究项目“四川深层页岩气产能大数据挖掘和智能评估方法研究”（编号：2021YJ0360） |
|
| 中文摘要:钻速预测是钻井优化的重要组成部分，机器学习算法是当前实现准确钻速预测的重要手段，准确的特征选择是保证机器学习精度的关键途径。基于南海某井眼的实际钻井数据，本文采用一种融合特征选择法从钻井特征参数中选出井径、钻井液出口温度、钻井液入口密度、钻井液出口密度、K值、塑性粘度、滤失量、上覆压力、孔隙压力、和喷嘴等效直径共10种参数。将优选出的参数作为模型输入，引入集成的梯度提升树（Gradient Boosting Decision Tree，GBDT）算法建立机械钻速预测模型。将建立的模型与常规机器学习算法模型进行对比试验。试验结果显示，所提出的融合特征选择算法模型精度较全特征模型高2%，较常用机器学习模型平均高14.5%，该研究为钻井参数的准确、快速寻优提供了有效解决方案，对提高钻进速率具有一定的指导意义和实际应用价值。 |
| 中文关键词:钻速预测 机器学习 融合特征选择 梯度提升树算法（GBDT） |
| |
| Research on drilling rate prediction model based on fusion feature selection algorithm |
|
|
| Abstract:ROP prediction is an important part of drilling optimization， machine learning algorithms are currently an important means to achieve accurate ROP prediction， and correct feature selection is the key way to ensure machine learning accuracy. Based on the actual drilling data of a well in the South China Sea， this research uses a fusion feature selection method to select 10 drilling characteristic parameters， including well diameter， outlet temperature， inlet density，outlet density， K value， plastic viscosity， filtration loss， overburden pressure， pore pressure， and nozzle equivalent diameter. The optimized parameters are taken as model inputs， and the integrated Gradient Boosting Decision Tree （GBDT） algorithm is introduced to establish a ROP prediction model. The established model is compared with the conventional machine learning algorithm model， and the test results show that the accuracy of the proposed fusion feature selection algorithm model is 2% higher than that of the full feature model， and the average accuracy is 14.5% higher than that of the commonly used machine learning model. The research provides an effective solution for the accurate and rapid optimization of drilling parameters， and have guiding significance and practical application value for improving the drilling rate. |
| keywords:ROP prediction machine learning fusion feature selection Gradient Boosting Decision Tree（GBDT） |
| 查看全文 查看/发表评论 下载PDF阅读器 |
