ISI Logo

Building & Validating a Data Driven Reservoir Model In D.J. Basin, Using The Top- Down Modeling (TDM) Technology

Problem Statement

An innovative top-down modeling (TDM) methodology is developed to more accurately simulate unconventional resources. While being completely data-driven, TDM is unlike approaches such as CRM (capacitance resistance modeling) that oversimplifies the complexities of fluid flow in the porous media and ignores the geology of the asset. TDM is conditioned to any and all field measurements (drilling data, well logs, cores, well tests, seismic, production history, etc.) to build comprehensive full-field reservoir models. The results of comprehensive analysis by TDM can provide insightful guidelines for improved field development planning and decision making.

The TDM approach was used to analyze an area in the Wattenberg Field in Weld County, Co., producing from the Niobrara formation using publicly available data. The TDM was built using data from more than 145 wells. Well logs, production history, well design parameters and dynamic production constrains were the main data attributes used to perform the data-driven analysis.

Methodology

As opposed to bottom-up modeling processes, TDM takes a completely different tactic for full-field reservoir simulation. In this state-of-the-art technique, reservoir engineering, statistical analysis, advanced data-driven analytics and machine learning are integrated to build a reservoir model as an alternative (or complement) to traditional reservoir modeling approaches.

This study analyzed 145 wells in the Wattenberg Field using various data-driven techniques to model the Niobrara reservoir. All data were publicly available and obtained through the Colorado Oil and Gas Conservation Commission's website. Intelligent Solutions Inc.'s IMagine software was used for the data-driven analysis. Fuzzy pattern recognition was implemented to provide the best producing locations as well as identifying underperforming wells. Also, an artificial intelligence-based, history-matched model was developed and validated for predictive practices, sensitivity analysis, infill drilling locations and other reservoir management purposes

Location and geology of the Asset

The workflow being presented here is part of the Intelligent Solutions Inc.'s IMagine software application that provided all the required tools and techniques for the development of the TDM. The area of review in this study consisted of six sections (21, 22, 27, 28, 33 and 34) located in Range 66W and Township 3N.

There were 222 wells that had production from Codell, Niobrara and J Sand formations in this area. Excluding the wells that produced from only the Codell or J Sand resulted in a total of 145 wells in six sections with production from the Niobrara formation only, or comingled with production from the Codell. In order to allocate the approximate production to each formation, a production ratio was used based on the porosities, formation thicknesses and water saturations obtained from the logs for each well. Again, all data analyzed in the study was acquired from COGCC's website.

Location and geology of the Asset

Results

One of the practical implementations of data driven analysis is Field Wide Pattern Recognition which depicts the areas representing high to low production history and remaining reserves. Fuzzy Pattern Recognition was used to determine different sections the reservoir with varying levels of contribution to the production during a specified time interval. This method can be used in order to assign different indices to each part of the reservoir based on their Relative Reservoir Quality (RRQ).

Fuzzy Pattern Recognition

The darker colors represent areas with higher cumulative production (Excellent RRQ) during the specified time interval. It is worth mentioning that RRQ is a good indicator of anticipated cumulative production (during a time interval) for a new drilled well in a specific area of the reservoir. Additionally, it is possible to determine Production Indicator (PI) for each well and compare it with the RRQ of the area that the well is located in. If the PI is less than RRQ, the well can be considered as under-performing well which could be a good candidate for re-stimulation.

Fuzzy Pattern Recognition applied to multiple producion indicators.

Fuzzy Pattern Recognition applied to Recovery Factor and Remaining Reservers

The most important aspect of data driven analysis for the oil production of this asset was developing a time-based reservoir model. This model used pattern recognition techniques (artificial intelligence) and honored reservoir engineering practices to regenerate production histories and predict the future performance of existing and new wells. In order to make an artificial intelligence-based reservoir model, it was necessary to train the TDM to able to recognize patterns between the production histories of the wells and the corresponding static (reservoir properties, well completion information) and dynamic (operational information such as days of production per month) data

Production data from all 145 wells beginning in 1986 and continuing through December 2012 were used in TDM training. Reasonably good history match results were achieved for both the entire field and individual wells. The green line represents TDM's production estimates and future predictions, while the dots represent actual data. Cumulative production for actual data and the TDM's predictions are shown by gray and green shaded areas, respectively. Future well production also was predicted for three years from 2013 to 2015 (yellow shadow).

In TDM production is conditioned to a static, dynamic and operational parameters.

History matching production in the entire asset

History matching production from individual wells

Another key capability of the TDM process is performing sensitivity analysis. Reservoir and well design properties are uncertain, and any change in any property can have an effect on production predictions. Using sensitivity analysis, the effect of uncertainty in these parameters on production forecasting was investigated using Monte Carlo simulation for uncertainty/sensitivity analysis. The advantage of TDM was fast computation time, which allowed performing up to thousands of production scenarios for Monte Carlo simulation.

History matching and forecasting production from individual wells

References: