Better understanding of production decline in shale gas wells

Abstract

Production data from the Eagle Ford shale (an analog to the Alaska Shublik shale) was collected from two neighboring counties and analyzed to correlate well performance with completion parameters including length of horizontal wellbore and number of hydraulic fracturing stages. Thirty-eight dry gas wells with production history range of 18-43 months were analyzed using 6 different decline curve analysis (DCA) models including Arps' exponential, harmonic and hyperbolic, power law exponential (PLE), logistic growth analysis (LGA) and Duong's models. In the matching process, 2/3 of history was used to tune the DCA models and their forecasts were compared to the remaining 1/3 of real history. The matching results were analyzed based on production history length and flow regime to have better understanding of limitations and capabilities of each DCA model. Reservoir simulation models, constructed using range of realistic data and actual completion practices of 4 select wells, were employed to assess reasonable values of remaining reserve and remaining well life that were used as benchmarks for comparison with DCA results. The results showed that there was no strong correlation between well performance (average first year production rate) and the horizontal leg or the number of fracturing stages. This was an indication of extremely heterogeneous medium. In most cases, the accuracy of the DCA models increased when longer production history was used to tune the model parameters. LGA seems to be the most accurate DCA model since it gave the highest matching accuracy 71% of the total wells when using longest history length of 31 months. As the flow regime is concerned, LGA model also performed very well matched in 57% of the wells exhibiting only transient flow and 63% for the wells showing transient flow during early production time followed by boundary-dominated flow during late production. Moreover, the remaining reserve and well life of the select wells predicted by LGA fell into reasonably close range of the estimates from the reservoir simulations.