Optimization Of Hydraulic Fracturing Treatment Parameters For Tight Gas Reservoirs Using Machine Learning

The gap between the supply and demand of natural gas resources in China has become increasingly prominent,and the rapid development of tight gas reservoirs has a strategic significance for ensuring national energy security and sustainable natural gas development.Hydraulic fracturing technology is one of the effective means for commercialized production of tight gas.Fracturing treatment parameters can significantly impact the post-fracturing productivity of gas wells.Traditional analytical methods and reservoir numerical simulation methods have introduced many assumptions in characterizing of fracture geometries and gas flow mechanism in porous media,and the realization process is complicated and the calculation efficiency is low.With the rise of artificial intelligence,machine learning methods have been widely applied in oil and gas field development,and many progresses have been made.This thesis proposes a method for optimizing fracturing treatment parameters based on machine learning,which aims to improve the effectiveness of fracturing stimulation.This thesis takes the vertical fractured wells in the eastern region of the Sulige gas field as case study.First,the raw data with 20 factors including geology,fracturing treatment,and production of 712 wells in the region are analyzed and preprocessed,and the multiple imputation method is selected to treat the missing values in the data set.The Pearson correlation coefficients are used to analyze the factors that have a strong correlation with the absolute open flow(AOF)of the gas well.The Min-max method and the principal component analysis method are used to standardize and reduce the dimension of the data set.Subsequently,K-means clustering and KNN classification algorithms were used to establish a diagnosis model for the main production control factors of vertical fractured wells,and the differences between specific factors between high-productivity wells and poor-productivity wells were analyzed,and the main fracturing treatment factors on the fractured wells include the total fracturing fluid volume,the pad fluid volume ratio,proppant volume,pump rate,and the average sand ratio.The built diagnostic model shows high prediction accuracy on the test set.The classification accuracy of high-productivity wells and poor-productivity wells is perfect,and the accuracy of medium wells has reached 97.9%.The application shows that the KNN classification diagnosis model has high prediction accuracy on the type of gas well in the region.Through the evaluation of the AOF prediction model established by machine learning algorithms such as multiple stepwise regression,support vector machine,BP neural network and random forest,it is concluded that the productivity prediction model established by the random forest algorithm on the training data set and the test data set has the best performance,followed by the support vector machine algorithm.The diagnosis model and prediction model are used to analyze 5 vertical fractured wells in this region,and taking a poor-productivity well as an example,through the optimization method based on support vector machine and referring to the fracturing treatment parameter values of similar wells including the pad fluid ratio,pump rate,proppant volume and average sand ratio were optimized.The optimized treatment design increased the AOF of the gas well by 50.5%,which can effectively improve the fracturing effectiveness and well production.