An Empirical Analysis Of Oil Well Production Prediction Based On Ensemble Learning

Oil well production prediction is one of the important methods for oilfield enterprises to refine their production goals and production decisions.The parameters in the process of oil wells contain the production law and relations of each factor.If these laws can be effectively tapped and used,it will provide suggestions for oilfield enterprise production managers to predict and reasonably control production.Aiming at the problem of oil well daily production prediction,this paper mainly uses the prediction technology in machine learning to solve the problem.First,a variety of methods are used for feature selection processing of oil well data.Based on this,individual learner models such as multivariate linear regression,decision tree,and Knearest neighbor,and ensemble learning models such as gradient boosting trees,random forest,and extremely randomized trees are constructed.Then,the cross-validation technique is used in the training model process to obtain the evaluation index value of the above-mentioned models.Based on the model score and the mean square error value,the prediction and performance advantages of the ensemble learning over the individual learner are compared and analyzed.Next,according to the prediction accuracy and stability of the model,a prediction model suitable for the daily production data set of the oil well is selected.The optimal model calculates the importance of the characteristics,and obtains the features that have a key impact on the four output dependent variables.Finally,the prediction results are analyzed and related improvement measures are proposed.The experimental results show that by sorting the importance of the variable features of each model,the main working parameters that affect production are: oil pressure,lower oil pressure limit,oil nozzle,pump diameter,water content,and stroke time.The individual learner has certain feasibility for oil well production prediction,but various metrics of the ensemble learning model are better than models such as decision trees and K-nearest neighbors.Combining the characteristics of the algorithm and the final evaluation index,the extremely randomized trees regression model in ensemble learning performs optimally on different oil well product predictions,indicating that extremely randomized trees regression can be effectively applied to the analysis and prediction of oil well daily production.Combined with the actual working parameters of oil well production,the application of the extremely randomized trees regression model to the empirical research will help relevant decision makers in oil field companies to reasonably control the dynamic production behavior of oil wells.