Prediction Focused Analysis: Forcasting Reservoir Performance Using Production Data Without History Matching
This is the companion code for "Addy Satiya; Celine Scheidt; Lewis Li; Jef Caers, Direct forecasting of reservoir performance using production data without history matching: a Libyan reservoir case study, submitted to Computational Geosciences"
The conventional paradigm for predicting future reservoir performance from existing production data involves the construction of reservoir models that match the historical data through iterative history matching (left triangle). We propose an alternative re-formulation of the problem, in which the role of the reservoir model is reconsidered. Instead of using the model to match the historical production, and then forecasting, the model is used in combination with Monte Carlo sampling to establish a statistical relationship between the historical and forecast variables. The estimated relationship is then used in conjunction with the actual production data to produce a statistical forecast. This allows us to quantify posterior uncertainty on the forecast variable without explicit inversion or history matching. We call this approach for direct forecasting Prediction Focused Analysis. In this repository, we show the application of this methodology to a real field case in Libya.
The case study is modeled after the WinstersHall Concession C97-I in Libya. 5 Producers and 3 Injectors have been in place for 3500 days, and production data is available for all 5 days. A decision needs to be made regarding drilling a new infill well (PNEW). The workflow consists of two parts a) Constructing the prior models and forward simulating to obtain responses b) Applying canonical functional correlation analysis to estimate the posterior forecast distribution.
The prior models are generated using the script here. A walkthrough is provided in the Jupyter notebook here. The models are then simulated using 3DSL.
A Jupyter Notebook providing a step-by-step walkthrough of PFA using CFCA can be found here. Alternatively, a demo script is provided here.
Jupyter notebooks may have issues rendering equations in certain browsers, refer to this pdf if equations appear to be mangled. Chrome is recommended.
The Functonal Data Analysis is courtesy of Jim Ramsay.
Contact lewisli@stanford.edu