(242c) Forecasting Gas-Oil Ratios and Solution Gas Production from Liquid Rich Shale Reservoirs

The basic issue: To enhance our knowledge about the production mechanisms of shale oil reservoirs and better exploit their economic potential, it is important to be able to forecast production of the secondary phase, namely gas.

Why it is important: Stakeholders often focus on oil forecasts when it comes to shale oil reservoirs, ignoring the equally important solution gas produced from these plays. Lack of production data, complicated flow mechanisms of liquid rich shale reservoirs, production pattern of producing gas-oil ratios among other factors, make the task of forecasting solution gas production difficult.

What we are proposing: The objective of this paper is to present a new method for estimating producing gas-oil ratios (GOR) and solution gas production from shale volatile oil reservoirs. The method is data driven while observing the fundamental equations that govern hydrocarbon production over time. It entails the following steps:

Step 1 â?? Generate representative collection of data: To develop the functional form that describes the evolution of GOR over time for a producing well, run a large number of representative simulations over a fairly long production time  (e.g. 30 years) so that a variety of wells in various reservoirs under diverse producing conditions can be well represented in the sample.

Step 2 â?? Identify natural GOR shapes through PCA: Stack GOR profiles on top of each other, to construct a matrix  of dimension perform principal component analysis (PCA) on , and then approximate  as , where . The vectors  can be thought of as natural shapes over time, whose linear combination can reproduce â?? with reasonable approximation â?? every conceivable GOR profile over time.

Note:Steps 1 and 2 will have to be completed perhaps only once or at worst a limited number of times, since the natural GOR shapes they produce will be universal.

Step 3 â?? Fit a linear combination of natural GOR shapes to data over limited time: Given GOR data for a limited production history over time , identify estimates  by finding .

Step 4 â?? Use the model of Step 3 for future predictions: The value of GOR at a future time  along with a corresponding confidence interval can be estimated using standard formulas as  where  is the information matrix.

Case study: A commercial compositional simulator was used to simulate 30 years of production from a shale volatile oil reservoir with 4 different reservoir fluids. The reservoir permeability and porosity were 0.001 md and 6% respectively. Several values for the following parameters were considered in a total of the  simulations:

1. 1. Critical gas saturations between 5% and 20%;
2. 2. Constant bottomhole pressures (BHP) between 500 and 1500 psi; and
3. 3. Initial reservoir pressures between 3000 and 5000 psi.

PCA on the GOR data produced by these simulations revealed (through cross-validation) that a small number of principal components adequately capture the GOR profiles observed over the 30-year period. To test how well the resulting model can forecast future production from production data over limited time, we considered a horizontal well of length 5000 ft, with 20 hydraulic fractures of 150 ft half-length, spaced 250 ft apart. We simulated producing GOR over 30 years, fit the model to the corresponding simulated data over periods from 6 months to 3 years, and then compared GOR forecasts by the model to GOR values produced by the simulator into the future. The future profiles produced by the PCA model matched those produced by the simulator with reasonable level of accuracy. The PCA model was also found to match the simulator on solution gas production, calculated by integrating producing GOR over cumulative oil production.

Conclusions: This work presents a novel approach of forecasting production from liquid rich shale plays, based on a well established statistical approach. It bypasses the need to invent an ad hoc functional form for GOR over time that would be both simple for computation and realistic enough to capture real data. The proposed approach can help improve reserves estimation, reservoir management, and project economics.