Advanced Production Diagnostics and Optimized Drilling Strategy: Unlocking the True Potential of Production Data Using Physics and Machine Learning

Reliable well survival statistics forecast for shale wells is crucial for investment decisions, optimized drilling rates, and energy security policies. At present, the industry lacks a standard analytical or data-driven solution for survival probability estimation in both space and time. We present a state-of-the-art physics-guided and data-driven model for well survival statistics forecast with an error of less than 5%. The implications include optimized long-term drilling strategy, robust well production diagnostics, and profitable economic decisions. Post pressure interference time, τ, the production rate of an unconventional well declines exponentially. Early and accurate estimation of τ is of utmost importance. Using 5000 wells (from Barnett, Marcellus, Haynesville, and Eagle Ford) as the training sample, we developed a universal neural network model to estimate the pressure interference time, τ, of a well. The estimated τ is coupled with the physics-based physical scaling curve to estimate the well survival statistics with time. Finally, the well survival statistics is used to calculate the residual production rate and its consequences on future well fracking strategy. The robustness of the neural network model (ANN) is tested on a test set comprising 6000 wells from Barnett, Haynesville, Eagle Ford, and Marcellus. Using initial well production data, and a single ANN model the predicted pressure interference time τ, has an R2 of 0.98. The well survival statistics estimates for the next 10 years in the future achieved an accuracy of 95%, 93%, 96%, and 95% for Barnett, Marcellus, Haynesville, and Eagle Ford respectively. We found that with time τ has decreased for all the plays (Barnett, Haynesville, Eagle Ford and Marcellus). The estimated well survival statistics is shown to have a strong effect on the optimized and profitable drilling strategy. We have used production data and a physics-informed neural network to unlock future production diagnostics with high certainty. The new and robust state-of-the- art hybrid model estimates the future pressure interference time of wells with an accuracy of 98%. The model calculates the future well survival statistics with time with an average accuracy of 95%. Early and accurate estimation of future production diagnostics governed future investment decisions, re-fracking strategy, and improved energy security strategy.