Fracture Flow Rate Estimation Using Machine Learning on Temperature Data

Near-wellbore fracture characterization methodologies help in identifying fluid entry points as well as flow rate in order to assess the effectiveness of hydraulic fracturing treatments, optimize the completion plan or identify the need for refracturing. Temperature transient analysis is one of such methods and previous work has shown that it allows for the estimation of flow rate coming out of fractures.
In this work, a machine learning approach to fracture flow rate estimation using temperature data is presented. The problem was formulated as a time series regression problem where the temperature data is used as the input of a reverse model that estimates flow rate. The Lasso Regression, Random Forest and Kernel Ridge Regression algorithms were tested in the study and three case studies are presented with varying levels of complexity.
The Kernel Ridge Regression approach was found to outperform the other two algorithms in the most complex case due to the specific formulation of the features as well as the mathematical similarity of the learning algorithm with the analytical solution of the physical problem.