Production data analysis and forecasting: Bayesian probabilistic methods to quantify uncertainty, analyze production performance trends in stack

Abstract

In this work, a Bayesian approach of probabilistic estimation for decline curve analysis in unconventional reservoirs is presented. The primary objectives of this study are the quantification of the uncertainty for production forecasting and do a parent-child analysis for wells from the same play. MCMC-based Metropolis algorithm is used for sampling from the proposal distributions to generate posterior distributions for the decline curve parameters. This sampling technique is applied for three models: Arps, Duong, and power law exponential models. Prior and likelihood distributions are established for the three models based on our understating of the data and the models. Forecast estimates are generated using multiple intervals of initial production data to understand how the sampling algorithm generates better estimates with increasing amount of training data. 282 oil and gas wells Meramec STACK unconventional play are used in this work to quantify the production forecasting uncertainty. Results show that the MCMC-based approach was able to establish uncertainty bounds, matching MAP estimates for cumulative production. Based on the amount of production data available and the nature of the flow, the model that fits best can vary. Using the estimated decline curve parameters, parent-child well comparison analysis is done to understand the changing production dynamics in the Meramec STACK play.