New forecasting method for liquid rich shale gas condensate reservoirs with data driven approach using principal component analysis
Academic Article
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
2017 Elsevier B.V. Accurate production performance evaluation and forecasting in shales during the early stages of development can play an important role in minimizing uncertainties associated with unconventional reservoirs. Given the limited reliability in forecasts from traditional decline models when applied to unconventional reservoirs, new tools to supplement the ones in use today are required to improve the accuracy of production forecasts. In this study, we present a method involving principal component analysis (PCA), which is a simple, non-parametric method of extracting relevant information from large data sets to perform production forecasting of liquid rich shale gas condensate reservoirs. We used a comprehensive compositional reservoir model to create several iterations of synthetic production histories from liquid rich shales (LRS) wells based on Monte Carlo simulation with predefined probability distributions. Cumulative gas, gas rate, and condensate-to-gas ratio (CGR) for the simulated cases were decomposed into principal component (PC) scores and coefficients were used to recreate the original data. The dataset was cross-validated to check its ability to predict the missing production data based on PC scores and coefficients of the limited production data. Principal component analysis was further applied to the field data from several wells from Eagle Ford shale. We re-created and cross-validated the field data by using limited PC which led to good matches of the original production data. Two to three PC's were required to recreate the initial data with reasonable accuracy depending on the quality of the input data. During the validation step, we observed that some of the wells exhibited significant error which could be attributed to significantly different production profiles of those wells compared to the other wells. For simulated data, four PC was enough to yield the prediction with average error of 0.16%, 0% and 0.77% respectively for gas rate, cumulative gas and CGR respectively. For field data, three PC yielded the best prediction with average error of 1.63% and 2.98% for gas rate and oil rate respectively. This work shows that multivariate statistics and data driven methods can be used as an important approach to complement existing tools like reservoir simulation and decline curve analysis to perform production data analysis. PCA can also be used and can generate accurate results relatively quickly. We recognize that even more rapid approximate methods will be required for routine analysis. Understanding the limitations of different approximate methods and application of methods to overcome these limitations in given circumstances should lead to optimal use of these methods.
