Applications of Inverse Theory and Machine Learning in Rate/Pressure Transient Analysis


The applicability of decline relations (empirical or analytical) in rate transient analysis (RTA) to forecast the production of an unconventional reservoir depends on the validity of assumptions made during the development of these decline relations. None of the present rate decline relations used to estimate the ultimate recovery (EUR), are conceptually applicable in a reservoir where hydrocarbon production exhibits variable rate and pressure. We propose a new methodology based on an inverse deconvolution problem formulation to accurately forecast performance in such reservoirs. To handle the instability in deconvolution, we propose the use of elastic net regularization and a new weighting scheme in our deconvolution algorithm.

In this work, we propose a non-parametric density-based outlier detection method, which identifies outliers by classifying the data into clusters and assigning local outlier factors to the individual data points. We validate our method using synthetic examples generated using numerical models of multi-stage hydraulically fractured wells in unconventional reservoirs. Upon validation we demonstrate our method using field examples. Our work demonstrates that this new methodology integrating, pressures into decline curve analysis is theoretically and practically more robust than the analysis of pressure normalized decline curves currently used to solve the problem.



Deconvolution, Inverse theory, Decline curve analysis, Local Outlier Factor, Outlier Detection, Reservoir Engineering