Application of Machine Learning for Target Selection and Acid Treatment Design

Abstract

Acid treatment is commonly used to enhance the production capacity of wells drilled in carbonate deposits. However, field outcomes of this procedure may vary significantly. Current approaches to acid treatment design rely on advanced software tools that evaluate major acidizing factors. Machine learning is a valuable complement to the existing techniques: it facilitates the selection of target wells and aids in defining initial parameters for design engineering on reliable and effective software platforms. This study examines potential applications of machine learning in target selection based on the history of treatment outcomes influenced by the initial well conditions, operational conditions, treatment frequency, acid volumes, acid system types, pretreatment strategies, acid system diverters, and acid residence time. Acid treatment design requires complex laboratory work to investigate the kinetics of acid-rock interactions determined by the mineral composition of the rock formation and the chemical properties of the acid system, including the concentrations of its components. The problem of predicting the reaction kinetics of acid systems by processing an array of laboratory data using machine learning methods, specifically linear regression and random forest methods, was discussed. It was demonstrated that the incorporation of machine learning enables the development of robust decision-making algorithms that optimize acid treatment by considering its multifactorial effects. These algorithms significantly simplify the tasks of acid treatment design.

Keywords

• vibrational spectra
• molecular structure
• electronography
• quantum chemical calculations