Operational Selection of Wells for Hydraulic Fracturing Treatment Through Machine Learning

The purpose of the work is the development of methodology for the operational selection of wells for hydraulic fracturing treatment based on machine learning for Field A. In order to prepare information for machine learning the data from the simulation model, monthly field reports, well test, HF reports are collected and analyzed. Principal component analysis was used in order to find out correlating between themselves parameters and exclude them from the database. Target for binary classification parameter is efficiency of HF treatment.

After the final database has been formed, machine learning process was realized. Six different models of machine learning were trained and evaluated: k-neighbors, support vector method, decision tree, random forest, gradient boosting, neural network. Comparison of models was carried out and best classification model was selected on base of several independent metrics. The parameters most influencing on the result of the success of the HF were identified. Machine learning model based on neural networks has been trained to estimate the average annual oil production rate after HF treatment. Based on the selected model, predictions are made for the remaining wells not participating in the training. Well candidates for HF treatment were proposed. Economic efficiency was estimated.