Forecasting the Results of Hydraulic Fracturing Based on Methods of Clustering

The real data collected from the field for the prediction of the results of hydraulic fracturing of the reservoir are investigated. The technical and geological factors of the process at the well are aligned with the interval of the starting volumes of oil production. To solve the problem, methods of machine learning are used, the clustering problem is solved by the proximity distribution method, the subsequent classification problem by the methods of the nearest neighbors and a random forest, the visualization of a multidimensional feature space. For clarity, the naive Bayes method is compared. The significant parameters are selected and their influence on the hydraulic fracturing process is determined. The obtained prediction accuracy of 95%, the existence of patterns was revealed and 5 parameters of the greatest impact on the fracturing were found with a total fraction of about 44%. The Python programming language is used for analysis and solution. The task is applied and extremely relevant for the industry, whose solution will help in optimizing the process of hydraulic fracturing and maximizing investment.