Bridging The Gap Between Material Balance And Reservoir Simulation For History Matching And Probabilistic Forecasting Using Machine Learning

Recorded On: 10/20/2021

In this talk we will start by discussing probabilistic forecasting methods using reservoir simulation for complex reservoir studies. We will emphasise the use of ensembles of simulation models, and discuss some of the difficulties in creating a valid probabilistic ensemble. We will describe Markov Chain Monte Carlo methods, and in particular random walk and Hamiltonian methods. We will show how these methods may be used with proxy models to generate probabilistic simulation ensembles.

In some cases, the bottleneck for rapid reservoir decision support is building and maintaining a reservoir simulation model. We will show how the reservoir simulation methods can be modified to support a data driven approach which includes known physics such as a material balance.

To generate a data driven model, we take historical measurements of rates and pressures at each well, and apply multi-variate time series analysis, with automatic feature selection, to generate a set of differential-algebraic equations (DAE) which can then be integrated over time using a fully implicit solver. We combine the time series models with material balance equations, including a simple PVT and Z factor model. The parameters are adjusted in a fully Bayesian manner to generate an ensemble of forecasts. The use of a DAE distinguishes the approach from normal statistical time-series analysis, where an ARIMA model or state space model is used, and is only suitable for short term forecasting.

We illustrate these approaches with examples of case studies of complex field history matching and probabilistic forecasting.

These approaches have many possible applications within the oil and gas industry, from subsurface to downstream.

This webinar is categorized under the Data Science and Engineering Analytics technical discipline.

All content contained within this webinar is copyrighted by Nigel Goodwin and its use and/or reproduction outside the portal requires express permission from Nigel Goodwin.

Nigel Goodwin

Director, EssencePS

Nigel Goodwin has an MMath and PhD in theoretical physics from Cambridge and Manchester universities, and has worked in the oil industry in a range of disciplines, modelling gas transmission systems, petrophyics and reservoir engineering. Since 2000 he was a co-founder and software director for Energy Scitech, which produced the leading commercial tool for history matching and probabilistic forecasting. Since then, he has continued to develop robust and validated workflows for reservoir simulation based probabilistic forecasting. More recently he has been working at Shell, applying time series machine learning techniques to a range of industry problems.

Yasin Hajizadeh (Moderator)

CTO, Maillance

Yasin Hajizadeh is the CTO of Maillance,(pronounced mai'ans) a startup focusing on democratizing ML for petroleum engineers. Prior to Maillance, Yasin was a program manager of Azure ML and IoT at Microsoft. He also worked for Schlumberger as a data scientist and reservoir engineer. Yasin holds a PhD in Petroleum Engineering from Heriot Watt University and a Master of Technology Management from Memorial University of Newfoundland.

SPE Webinars are FREE to members courtesy of the



Bridging The Gap Between Material Balance And Reservoir Simulation For History Matching And Probabilistic Forecasting Using Machine Learning
10/20/2021 at 11:00 AM (EDT)   |  90 minutes
10/20/2021 at 11:00 AM (EDT)   |  90 minutes
20 Questions
0.15 CEU/1.5 PDH credits  |  Certificate available
0.15 CEU/1.5 PDH credits  |  Certificate available