论文部分内容阅读
Influx and loss are the two kinds of bottom hole complex accidents.They not only cause the reservoir damage,increase the exploration cost,reduce the drilling ef-ficiency;but also induce major malignancy,such as pipe sticking,wellbore collapse and blowout.Therefore accurate and early detection of influx and loss during drilling has signif-icant meaning.Traditional influx and loss detection methods have the shortcoming of moni-toring time lagging and high costs.As the rapid development of artificial intelligence tech-niques,researchers start to detect influx and loss using artificial intelligence method.One of the most important data driven machine learning method,random forests,has the char-acteristics of high prediction accuracy,less adjustment parameters and strong robustness.It has been applied widely in different fields.This work adopted random forests to detect influx and loss in real-time.The work includes five steps: (1) Generating raw influx/loss data set by combing real-time drilling data and drilling history data;(2)Pre-processing raw data set;(3) Generating influx/loss training data set using bootstrap tech-nique;(4)Using CART(Classification and Regression Tree)algorithm to generate classi-fication trees for each training data set;(5)Predicting influx/loss by voting result of the classification trees.The field example shows that influx and loss can be detected accurately in real-time by proper pre-processing real-time drilling data and adopting random forests algorithm.The importance of input variables are also analyzed in this work using random forests method,and the result indicated that the difference between inflow and outflow is the most important variable for influx/loss detection,which is consistent with the real-ity.The results also demonstrated that input variables,such as rotation speed,out flow temperature,etc,has little impact on the detection result,therefore these variables can be removed in the future study to improve the calculation speed.