Machine Learning as a Reliable Technology for Evaluating Time/Rate Performance of Unconventional Wells

Abstract

Summary Production-data analysis is a practice fraught with inconsistencies. In the application of any single model, the quantity of answers arrived at by experienced evaluators is often equal to the number of evaluators analyzing the data. The cause of such inconsistency is bias on the part of evaluators. Although the colloquial use of bias typically implies systematic error, in this paper, we define bias as an expression of belief by the evaluator. With the lack of recognition of bias, no means exists with which to gauge its accuracy. A method that requires explicit expression of one's bias in time/rate decline behavior can provide an objective means with which to evaluate it. In this work, we present a machine-learning method to forecast production in unconventional, liquid-rich shale and gas-shale wells. Methods were developed for probabilistic decline-curve analysis with Markov-chain Monte Carlo simulation (MCMC) as a means to quantify reserves uncertainty, to incorporate prior information (i.e., bias), and to do so quickly. We extend the existing approaches by (a) a modified likelihood-distribution function to improve “learning” of production data, (b) integration of the transient hyperbolic model (THM) to explicitly define the various flow regimes present in unconventional wells, (c) a method for construction of discretized “percentile neighborhood” forecasts, and (d) construction of type wells from an analyzed well population. The accuracy and calibration of the method are demonstrated by an analysis of 136 wells in the Elm Coulee Field of the Bakken. Quantification of change in time/rate behavior caused by completion design, and the inference of physical behavior and properties, is demonstrated with a tight oil play in the Cleveland sand formation of the Anadarko Basin, as well as a shale play in the Wolfcamp formation of the Permian Basin. We show that this implementation of supervised machine learning, in combination with well-calibrated bias, improves the estimation of uncertainty of the posterior distribution of forecasts. In addition, hindcasts performed at various time intervals result in accurate estimation of mean five-year cumulative production. We observe that the “percentile neighborhood” forecasts are reasonable fits of production data comparable to those that may be created by a human evaluator, and that the type well computed is representative of the decline behavior of the well population upon which it is based. We conclude that, given the speed and accuracy of the process, machine learning is a reliable technology as defined by the US Securities and Exchange Commission (SEC), and can significantly improve the process of production forecasting by human evaluators for most unconventional wells with consistent trends of production history.

Methods of Decline Curve Analysis for Shale Gas Reservoirs

“…Bayes's Theorem [58] has been widely used for uncertainty quantification [59][60][61][62]. Gong et al [59] introduced a Bayesian probabilistic methodology using Markov chain Monte Carlo (MCMC) coupled with the standard Metropolis-Hasting (MH) algorithm [63,64].…”

“…Applying the new method to 197 Barnett shale gas wells, the authors concluded that their methodology can quantify the uncertainty in hindcasted production with a narrower P90-P10 interval and higher computational efficiency than the modified bootstrap method [65], which was originally proposed by Jochen and Spivey [66]. A Bayesian machine learning method with the MCMC coupled with the MH algorithm was proposed in Fulford et al [60], and was used to forecast production in shale wells by utilizing the transient hyperbolic model introduced by Fulford and Blasingame [67]. However, the behavior of b and D parameters as a function of time is only valid for homogeneous reservoirs with equal fracture half-length and spacing, not suitable for heterogeneous reservoirs with varying fracture half-length and spacing [68].…”

Combining Physics, Statistics, and Heuristics in the Decline-Curve Analysis of Large Data Sets In Unconventional Reservoirs

“…The Arps (1945) decline-curve family has been widely applied in the industry to estimate reserves. This practice also has been extended to unconventional reservoirs (Gong et al 2014), where the hyperbolic model is the most suitable to capture the production decline during the transient state:…”

“…When investing in a field-development plan, it is essential to be aware of the risks taken and determine a probable range of reserves volumes. For this reason, uncertainty-quantification algorithms have been widely applied to decline-curve analysis (Cronquist 1991;Chang and Lin 1999;Cheng et al 2008;Gong et al 2014;Fulford et al 2016;Yu et al 2016). Purvis and Kuzma (2016) provides an overview of methods commonly used.…”

Generalized Extreme Value Statistics, Physical Scaling and Forecasts of Oil Production in the Bakken Shale

“…To make things worse, the empirical DCA fits of particular wells are ill-suited to forecasting production from a wide area of a given shale play in which reservoir properties vary and uncertainties abound. Therefore, some authors have developed probabilistic models to introduce a range of possible outcomes into their production forecasts [15][16][17][18][19]. The most common assumption is that well productivities in shales are log-normally distributed.…”