Intelligent Prediction of Pore Pressure Using Real-time Drilling Parameters
The existing pore pressure prediction methods have limitations like restricted application,insuffi-cient accuracy,complex computation,and inability to predict in real-time manner.This paper presents a new method for predicting pore pressure based on real-time drilling data.Firstly,the logging data is used to calculate the theoretic actual value of pore pressure,which serves as the learning objective for prediction.Secondly,by the correlation coefficient and model selection methods,eight key parameters were determined,including hook load,pump pressure,rate of penetration,weight on bit,RPM,flow rate,density and viscosity.Then,three ma-chine learning algorithms were adopted respectively to build real-time pore pressure prediction models.The predic-tion results of train set show that the XGBoost and LightGBM models yield good results of key performance indica-tors(KPIs),while the random forest(RF)model has the problem of over fitting.The prediction results of the test set show that the XGBoost and LightGBM models are more superior in prediction accuracy and stability.All models produce translation deviations when parameter change and bit wear occur after the bit is replaced.The rela-tionship between the bit features and the prediction deviation can be investigated,or the models be modified to properly correct the prediction results,so as to achieve higher prediction accuracy.The proposed method is more accurate,and also provides real-time support for field decision making,thereby facilitating drilling optimization and risk reduction.
NETL
NSTL
万方数据
