In this paper, we present a methodology to predict the performance of horizontal gas wells in unconventional reservoirs using publicly available completion data. Our process combines public domain data with statistical analysis and probabilistic simulation methods to forecast well performance without a detailed reservoir characterization. We have tested our methodology using a 425-well dataset from the unconventional Montney resource play in British Columbia, Canada. We believe this workflow can be applied to other resource plays with similar data. In our SPE Paper 167154 [1], we determined the sensitivity of production performance to completion parameters using multivariate regression analysis on the same 425-well dataset from the Montney formation. We found that the number of fracture stages and the number of perforation clusters per stage were the most influential predictors of well performance. In this paper, we discuss how we combined the regression analysis results with probabilistic methods to predict well performance. The model converts the deterministic regression coefficients into probabilistic distributions to account for parameters not considered in the original regression analysis, including reservoir properties. The results of our study show that by using this model, we can match the range of actual well performance outcomes with a 95% confidence. Considering the importance of shale gas resources to the North American energy supply and the difficulty of characterizing shale gas reservoirs, this methodology offers a distinct advantage by providing a predictive model for well performance without the need for a detailed reservoir characterization. This also could be a beneficial tool to use in scoping studies where high-level, rapid evaluation is required. Introduction In SPE paper 167154, we reviewed completion reports of 425 wells located in British Colombia to study the influence of several completion parameters (lateral length, number of fracture stages, perforation cluster per fracture stage and fluid volume) on production performance. Since the slickwater fracture fluids have become the most common industry practice in the area, we limited the final results presented in that paper to a dataset comprised of 127 horizontal wells using slickwater fracture fluids. Also in that study, our group developed a model using multivariate regression analysis to predict the well performance of Montney horizontal wells based on completion parameters alone. In that paper, we applied a multivariate regression analysis to our database, utilizing four completion parameters to predict the average gas rate over the first year of production (B12). The regressed data set matched the mean value of the observed data very well and generated useful trends to determine how much individual completion parameters influence B12. However, one deficiency of that work was that the range of predicted B12’s was not nearly as large as the range of observed values. In this paper, we describe how we added probabilistic techniques to the regression-based predictive equation to create a predictive model that matches both the mean and scatter of the data. The objective is to generate a distribution of predicted B12’s similar to the observed data. This is accomplished by increasing the variance of the input distributions (both the variables and coefficients in our equation). We used Monte Carlo simulation to impose the increased variances on each of four different models and chose the best model by comparing the distribution of predicted B12 values from each model to the distribution of B12 values of the observed field data. Multivariate Regression Model and Probabilistic Model Eq. 1 is one of our models for the slickwater fracture fluid well dataset. This equation is formed using the tabular output (Table 1) from multivariate regression analysis in Excel to determine the coefficient of each variable. This equation relates B12 (Mcf/day) to the completion parameters. The coefficients of all parameters in the equation (Lateral length per stage, FS, PC/stage and Fluid/stage) are positive, which indicate an increase in each of them will increase the B12. This equation tells us that adding one fracture stage will increase the B12 average rate by 169 Mcf/D and increasing the perforation clusters by one will increase the average B12 by 284 Mcf/D. As shown by Table 1, PC and FS have more impact on the production rate compared to LL /Stage and Fluid/stage. This is determined by evaluating their coefficient, t-stat and p-value. A description of each variable in Table 1 was included in SPE paper 167154.

Publications with shared reference:

0 publications found