A Bayesian approach to deconvolution in well test analysis

TL;DR

This paper introduces a Bayesian deconvolution method for well test analysis that estimates reservoir parameters and true pressure and flow rate values, allowing for uncertainty quantification and incorporation of prior physical knowledge.

Contribution

It presents a novel Bayesian framework using a parametric physical model and MCMC sampling for joint inference of reservoir parameters and true signals in well testing.

Findings

Method provides comparable results to existing solutions.

Allows for quantification of parameter uncertainty.

Enables incorporation of prior physical constraints.

Abstract

In petroleum well test analysis, deconvolution is used to obtain information about the reservoir system. This information is contained in the response function, which can be estimated by solving an inverse problem in the pressure and flow rate measurements. Our Bayesian approach to this problem is based upon a parametric physical model of reservoir behaviour, derived from the solution for fluid flow in a general class of reservoirs. This permits joint parametric Bayesian inference for both the reservoir parameters and the true pressure and rate values, which is essential due to the typical levels of observation error. Using a set of flexible priors for the reservoir parameters to restrict the solution space to physical behaviours, samples from the posterior are generated using MCMC. Summaries and visualisations of the reservoir parameters' posterior, response, and true pressure and rate values can be produced, interpreted, and model selection can be performed. The method is validated through a synthetic application, and applied to a field data set. The results are comparable to the state of the art solution, but through our method we gain access to system parameters, we can incorporate prior knowledge that excludes non-physical results, and we can quantify parameter uncertainty.