The classical Material Balance equation (F = N*Et + We) is a zero-dimensional reservoir modeling methodology that is used to estimate original oil-in-place volume (N), gas cap size, and aquifer influx (We). The material balance equation is a single equation with many unknowns (e.g. N, m, We); thus the solution to the equation is inherently non-unique. In other words a range of original oil in place (STOOIP) gives good match of the material balance equation. The range of STOOIP solutions of the material balance equation is often too high. In this paper a methodology is suggested for reducing the uncertainty in the material balance derived STOOIP values. In the proposed methodology, the material balance equation is further constrained by history matching the average fluid contacts in the reservoir. Three case studies were used to illustrate the application of material balance and fluid contact match in STOOIP estimation. The first case study is a synthetic reservoir with a STOOIP of 100 MMBO, initial gas cap and aquifer influx. The result shows that solving only the material balance equation gives a very wide range of STOOIP of 80 - 300 MMBO, while including the fluid contact match reduced the STOOIP range to 80 – 125 MMBO. The second case study is a real reservoir with initial gas cap and aquifer influx. Using material balance alone the STOOIP range was 80 – 800 MMBO whereas including the fluid contact match gave a lower STOOIP range of 100 – 120 MMBO. The last case study also shows a reduction in the range of STOOIP estimation from 50 - 500 MMBO for solving the material balance equation alone to 90 - 120 MMBO when the fluid contacts are history matched. These case studies show that uncertainty in the material balance STOOIP estimates are greatly reduced by matching fluid contacts.
The authors have used this paper to demonstrate how material balance was applied in field development planning for a green gas field. In this work, we have used one of the reservoirs as case study. Deterministic tank model was initially built for the reservoir using MBAL™. Petrophysical properties, aquifer parameters and relative permeability data were all added into the model. Well flow models were generated using PROSPER™ and then imported into MBAL™. Facility constraints were imposed, and deterministic prediction run was performed. Key impacting parameters on the recovery factor were assessed, and corresponding ranges were estimated for each. A probabilistic prediction workflow was developed and applied to the deterministic model. This uses experimental design to generate multiple runs with the aid of OpenServer™. Response/proxy function for gas recovery was then generated and tested for consistency with "observed" data. Multiple Monte Carlo runs were then done using Crystal ball, and the 10th, 50th and 90th percentiles were extracted. The corresponding parameters for these respective percentiles were then tested in MBAL™ to check for reliability. Finally, all reservoirs were rolled-up using GAP™, and the recovery factors were checked for consistency with MBAL™. The recovery factors (P10, P50 and P90) from the probabilistic material balance work were then compared with results from grid-based simulation work done on the reservoir. The figures were further compared with estimates from local and global analogues, as well as analysis done by a third-party. Results from the MBAL™ work compared reasonably with recovery factors from the other methods. Probabilistic material balance approach helps to remove bias/anchoring while estimating a range of outcomes for recovery factor. It also gives reasonable estimates, as demonstrated by the closeness of results with other methods. However, it is not a replacement especially for the grid-based simulation, but should rather be a complement. The methodology has been successfully applied to other gas fields and reliable results were also obtained. The work was equally adapted to more complex systems as multi-tank models.
Experimental design method is very useful in green field development. It helps to understand the impact of uncertainties on ultimate recovery, and hence, gives guidance on business decision making. Outputs from the process (Pareto/Tornado charts, selected models) are good exhibits for use in the uncertainty management plan, which also drives data acquisition and work plans for future stages/phases of the project. This work shares some lessons from using experimental design for development planning of two Non-Associated Gas (NAG) condensate reservoirs. It demonstrates the importance of selecting appropriate design for proxy generation/Monte Carlo simulation runs – and eventual model selection. This paper has two case studies: (i) a gas condensate Reservoir "A," with total of 20 parameters (9 discrete and 11 continuous variables), and (ii) another gas condensate Reservoir "B", which has 22 parameters. Folded Plackett-Burman (FPB) design was first used in both cases. However, owing to limited number of runs, only linear proxies could be created. This did not meet the objective of the process, because it does not allow for generating interaction terms among parameters. The FPB runs were therefore used only as screening studies, while 3-level D-Optimal designs were subsequently used for response surface model (proxy) generation. Four sensitivities were done on Reservoir "A": (i) 20 parameters with 300 D-Optimal runs; (ii) 14 parameters with 300 D-Optimal runs (after screening out less impacting parameters on objective functions); (iii) 20 parameters with 500 D-Optimal runs; and (iv) sensitivity (i), but with additional fresh 200 D-Optimal runs. The two sensitivities done on Reservoir "B" are: (i) 22 parameters with 200 runs; and (ii) 15 parameters with 300 runs. It was observed that only sensitivities (ii) and (iii) for Reservoir "A", and sensitivity (ii) for Reservoir "B" yielded meaningful proxies. In conclusion, using Folded Plackett-Burman (FPB) designs alone in cases with many variables, as shown in this work, may not lead to meaningful proxies (especially, when there are interactions among parameters) because it is restricted to only linear proxies. Also, it is important to have adequate number of 3-level design (D-Optimal) runs for both process efficiency and proxy generation. Too few runs result in unreliable proxies, whereas too many runs take more time/computing resources. In addition, carrying large number of variables into the 3-level design stage requires more runs and also leads to more cumbersome proxies.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.