New approach to estimating ultimate recovery for hydraulically fractured horizontal wells, A

Abstract

Reserve estimation is an essential part of developing any reservoir. Predicting the long-term production performance and estimated ultimate recovery (EUR) in unconventional wells has always been a challenge. Developing a reliable and accurate production forecast in the oil and gas industry is mandatory as it plays a crucial part in decision making. Several methods are used to estimate EUR in the oil and gas industry and each method has its advantages and limitations. Decline curve analysis is a traditional reserve estimation technique that is widely used to estimate EUR in conventional reservoirs. However, when it comes to unconventional reservoirs, traditional methods are frequently unreliable to predict production trends for low permeability plays. In recent years, many approaches have been developed to accommodate the high complexity of unconventional plays and establish reliable estimates of reserves. The objective of this study is to develop a methodology to predict EUR for hydraulically fractured horizontal wells that outperforms current methods and overcomes some of the limitations of using decline curve analysis or other traditional methods to forecast production. A new approach is introduced in this study to estimate EUR for hydraulically fractured horizontal wells that consists of a workflow for production history matching using decline curve analysis (DCA), artificial neural networks (ANN) predictive models, and probabilistic and statistical analysis. The developed workflow for production history matching combines different DCA models with least-squares fitting to match actual production data and reliably forecast production. The production history matching workflow resulted in very accurate matches (correlation coefficient of 0.99) of actual production data for the entire production history and produced accurate production forecasts (correlation coefficients of 0.73-0.99) using only limited periods of early production history (three months to two years). The predictive models use ANN to predict EUR given short early production history data along with well completion data. The ANN models showed good accuracy (correlation coefficients of 0.85-0.99) in predicting EUR for around 1,000 hydraulically fractured horizontal wells used in this study given only three months to two years of production data. The ANN models utilize the production forecasts from the production history matching workflow along with well completion data to improve EUR predictions. Probabilistic analysis performed on all the models using Monte Carlo simulation techniques established a range of probabilities of the ANN models predictions (P90, P50 and P10). Also in this study, several statistical data analysis were conducted to study the production behavior of more than 1,000 hydraulically fractured horizontal wells. The end product of this work is a comprehensive workflow to predict EUR that can be implemented in different formations by utilizing well completion data along with short production history data.The credibility of the work in this study was tested using actual production and completion data from more than 1,000 hydraulically fractured horizontal wells from five different formations and compared with other models from literature. One of the main added values of this study is the ability to predict EUR given a short early production history (one month to two years), overcoming a major limitation in DCA as it tends to require longer production history to produce reliable forecasts. Furthermore, this study utilizes well completion data in predicting productivity using ANN models, leading to much better predictions of EUR than solely relying on production data alone. Finally, a comprehensive computer-based interface (software) was assembled to conduct and utilize all the analyses and models developed in this study.