Artificial Intelligence and Machine Learning Approach for Modelling Injectivity Decline due to Particle Invasion

ABSTRACT Particle invasion is a major contributor to well impairment in waterflooding operations, resulting in permeability (K) reduction and formation plugging. It significantly affects the injectivity decline. In this study, supervised Machine Learning (ML) utilising an Artificial Neural Network (ANN) has been employed to model and predict K decline due to particle invasion in core flooding experiments. Mathematical modelling has traditionally been used for K decline, but a single model struggles to generalise across different experiments. This is where Artificial Intelligence (AI) and ML play a crucial role by training a model on a large dataset, incorporating data from numerous experiments. This paper has successfully demonstrated the feasibility of the proposed approach using data from five distinct core flooding experiments obtained from published works for model training. The model exhibited excellent performance in predicting K, with an R2 value approaching 1. Additionally, the logistic regression model accurately predicted the type of filter cake formed, achieving a classification accuracy score of 1. These findings have significant implications for designing waterflooding and injection operations, which are becoming increasingly important in meeting the growing oil demand.