Accurate Prediction of Post-Fracturing Productivity Index Using Advanced Machine Learning Models: A Field-Validated Framework

Abstract

Forecasting the post-fracturing productivity index (PI) accurately remains essential as it enables both hydraulic fracturing optimization and better well performance as well as financial sustainability. The prediction of post-fracturing productivity index through traditional methods such as empirical correlations along with analytical models and numerical simulations and well testing results in both time inefficiencies and high costs and restricted capabilities in complex reservoirs. The objective of this research focuses on creating advanced machine learning models that deliver accurate predictions for post-fracturing PI. This research used standard hydraulic fracturing and reservoir parameters to develop nine advanced machine learning model systems. The trained developed models received their training data from actual post-fracturing PI measurements taken from pressure gauges deployed in wells and traditional production test results. Machine learning algorithms were developed by analysing a large dataset from 258 wells. The most efficient algorithms for post-fracturing PI prediction yielded excellent results, with RMSE values of 0.019 for AdaBoost, 0.021 for Gradient Boosting, 0.021 for Neural Network, and 0.047 for Random Forest. In a field application of AdaBoost, the model demonstrated a 3.2% prediction error, which closely matched the actual PI results. The study established that ML methods deliver precise PI predictions while offering both budget-friendly features and high operational speed, making them viable substitutes for traditional prediction tools in hydraulic fracturing design applications and reservoir management needs.