Study On Prediction Method Of Formation Pressure In Oilfield Water Drive Based On Machine Learning

Formation pressure refers to the pressure of fluids in the pores of a formation.Accurate and effective prediction of formation pressure is a prerequisite for achieving high-quality,efficient,and safe drilling.In response to the problems of high labor cost and low timeliness in traditional methods for measuring multiple wells or large areas,this paper proposes an oilfield water drive formation pressure prediction model based on Blending ensemble model,designs and implements an oilfield water drive formation pressure system.The specific content is as follows:Firstly,in response to the problem of excessive number of features in on-site data,this paper analyzes the main factors that affect formation pressure in water injection wells and production wells through a combination of data-driven and theoretical analysis.In the first step,Pearson Correlation Coefficient and Random Forest Importance Coefficient are used to determine the main factors that affect formation pressure.The second step is to compare and analyze the influencing factors selected through data-driven screening with the theoretical formulas related to formation pressure.Therefore,a total of 14 main factors that affect formation pressure are ultimately determined for water injection wells,and 15 main factors that affect formation pressure for production wells.Secondly,four machine learning models,namely random forest,Adaboost,XGBoost and LightGBM,are constructed to predict the formation pressure of oilfield water drive,and the parameters most suitable for the model are determined by grid search method.Through comparative analysis of the results,the accuracy rates of the four models on the water injection wells test set are 85.76%,87.14%,88.04%,and 88.51%,respectively.The accuracy rates of the four models on the production wells test set are 88.53%,85.21%,89.03%,and 89.93%,respectively.It shows that the four models we choose are very suitable for the research questions in this paper.Thirdly,aiming at the problem of formation pressure prediction of oilfield water drive,this paper proposes a two-layer Blending integrated model,and selects four models,namely random forest,Adaboost,XGBoost,and LightGBM,as the base model,and selects a three-layer BP neural network as the meta model.Through comparative analysis of the results,the accuracy of the two-layer Blending integrated model proposed in this paper is 93.53% on the test set of water injection wells and 94.21% on the test set of oil production wells,both of which are higher than that of the single model.In order to verify the generalization ability of the model,this paper uses verification set data to verify and compare four single models and Blending integrated model.The results show that the Blending integrated model proposed in this paper has the best effect on the prediction of oilfield water drive formation pressure,and effectively improves the accuracy of oilfield water drive formation pressure prediction.Finally,in order to verify the effectiveness of the method proposed in this article,the PyQT5 framework in Python language was used to design and implement an oilfield water drive formation pressure system.The system includes a model module,data query module,statistical module,and contour map module.The system can achieve formation pressure prediction,historical data query of formation pressure,statistical analysis and visualization of formation pressure data,and has practical application value.