Enhancing Water Influx Prediction: A New Machine Learning-Based Model for Accurate Water Cut Forecasting

Abstract Accurate prediction of water influx is a critical aspect of reservoir management, particularly in fields with water injection and multiple production wells. However, traditional methods for forecasting water influx often struggle to handle the complex dynamics of reservoir behavior and fluid interactions, especially in fields with limited geological variability and a limited number of wells. This study introduces a novel machine learning-based model to predict water influx in a specific oil reservoir containing three production wells and two water injection wells. With over 1,000 production data points available, including real-time measurements of water cut, pressure, and temperature, the proposed machine learning framework aims to improve water influx forecasting accuracy by using data-driven correlations instead of conventional empirical models. The model integrates various machine learning algorithms, such as Random Forests and Neural Networks, to predict water cut and influx based on historical production data and well parameters. The dataset used for training and validation spans a significant period of field operation, allowing the model to learn and adapt to the temporal variations in water influx behavior. Preliminary results indicate a significant improvement over traditional methods, with an R2 value of 0.92 and a reduction in the root mean square error (RMSE) from 180 bbl/day to 120 bbl/day. This study addresses the limitations of conventional methods by demonstrating the potential of machine learning to provide a more accurate, adaptive, and scalable solution for water influx prediction, specifically for fields with a limited number of wells. By improving water influx forecasting, the model helps in optimizing water injection rates, enhancing oil recovery, and reducing operational costs.