A Cost Effective Approach to Modeling and Managing Large Gas Fields

Yee, David (PHH Petroleum Consultants Ltd.) | Poitevien, Ricot (Apache Canada Ltd.)

OnePetro 

Management of gas reservoirs can be a difficult task if there is a varying degree of interference between wells. This difficulty increases with well count and the number of compressors and inter-connected processing plants. Numerical simulation of the integrated network and reservoir can help substantially. However, stability issues in complex networks and extensive data requirements have made simulation costly for large and complex projects.

With an extremely stable and efficient numerical model combined with data preparation techniques that rely on the use of existing databases and automated techniques, as well as streamlined history matching approaches, it has now become economically viable to use integrated models for large shallow gas properties. Without such a model to predict future performance, there is significant risk of over-building and over-drilling for future development. This paper discusses the setup, calibration and day to day use of an integrated model for a shallow gas field in Southern Saskatchewan, Canada.

Background

The Hatton reservoir has three main productive geological zones, all of which are of low permeability. As of October, 2001 the Hatton gas field in this project contains over 2900 wells with 38 years of production history. Production from the field began in 1964 and was taken over by Fletcher Challenge Energy in the late 1960's. Rapid development started in 1986 and has continued to the present. The field has since been acquired by Apache Canada Limited.

This field has three shallow Cretaceous gas zones, Milk River (MR), Medicine Hat (MH) and Second White Speckled Sandstone (SWS). The MR and MH zones contain shale and silt layers with sandy lenses. The MH pay is all at the top of the formation immediately below the MR. The Milk River and Medicine Hat zones have very low permeability and produce at low rates while the SWS zone has higher permeability and produces at higher initial rates. Productive Milk River exists over the entire field while the Medicine Hat pay disappears to the east and the productive SWS exists only in the extreme south. While the three zones are still segregated in some wells it is now common practice to commingle the upper two zones and segregate the SWS. The Milk River-Medicine Hat well spacing varies from 64.7 to 32.4 ha (160 to 80 acres) while the SWS spacing varies from 259 to 64.7 ha (640 to 160 acres). The main part of the field has been developed under closer spacing and most of the remaining reserves are in the Milk River zone as the other productive zones have less net pay thickness and area and had higher initial producing rates.

In the year 2001 program 200 wells were added and in 2002, a 600 well infill program was initiated. At the end of October, 2001 there were 1846 wells producing out of a total of 2136 wells in the Apache operated area. The surface network in Hatton is very complex with multiple delivery points, compressor stations and significant flow splitting. There were four sales points and six compressor stations, two of which were boosters. The maximum rate for the Apache wells in this field reached 4000 E3m3/d during the late 1980's and has since decreased to approximately 1500 E3m3/d.

The original model for this field was built in 1993 using a "pseudo well" model where wells associated with a given battery were grouped and averaged into tank reservoirs. Gathering lines connected these pseudo wells to the compressors and plants. This model was limited because of the use of tanks to represent reservoirs, where new wells may start at lower pressures than actual. In addition, pressure drops along group lines are not properly accounted for and uncertainty existed in the calculation of effective diameters for the pseudo wells.

With improved computer performance and simulation programs it is now practical to model such a system on an individual well basis with reasonable computing times. A decision was made to build a full-scale model of the system to help plan facilities optimizations for a 600 well infill program for 2002.

A discrete reservoir surface network model was chosen to improve the field analyses. All 2136 historic wells in the Apache operation and 818 outside wells were included in the reservoir history match.