ROP Prediction Method Based on Stacking Ensemble Learning

Rate of penetration(ROP)is an important indicator to evaluate the petroleum drilling performance.To accurately predict the ROP at an oilfield in the Xinjiang work area,the historical drilling data from the area were processed using the local outlier factor(LOF)algorithm,and an ROP prediction model based on Stacking ensemble learning was established.The model integrated by Stacking strategy with the K-nearest neighbor(KNN),support vector machine(SVM)or random forest(RF)algorithm showed inaccurate classification in the verification.The genetic algorithm was then adopted to optimize the parameters of the basic models.The optimized models integrating KNN,SVM,RF and Stacking algorithms yielded the prediction results with accuracy of 73.7%,78.9%,81.6%,and 97.4%,respectively.Clearly,the Stacking-based model gets the highest accuracy.Thus,a software was developed using the Stacking-based model.It was applied to predict the ROP under two sets of parameters.The results show that the predicted ROP matches well with the actual ROP,and the software works stable.This proves the generalization and accuracy of the Stacking-based model.This intelligent algorithm has provided a new means to predict ROP and optimize drilling parameters at the oilfield of Xinjing work area.