Accurate estimation of the estimated ultimate recovery (EUR) is critical in decision making processes related to the development of conventional oil reservoirs. Existing methods have limitations when it comes to predicting such long-term production behaviors. This study analyzes the performance of deep learning methods such as long short-term memory (LSTM) neural networks on time-series data, and their effective application to accurately estimate the EUR in conventional reservoirs. Synthetic data that are realistic and representative of many major conventional oil reservoirs were generated for this study. The generated dataset was used by the LSTM model for the purpose of forecasting the EUR. The results of the LSTM model were compared with that of a reservoir simulation model from a full-physics reservoir simulator. EUR forecasts from the physics-based reservoir simulator is used as a benchmark and the LSTM model shows a good predictive accuracy while forecasting the long-term production behavior from a well in a conventional oil reservoir. The LSTM model-based deep learning method can be effectively used with real-field data obtained from wells in conventional reservoirs to accurately predict the EUR, and the study provides a comparative analysis of the results and factors affecting the EUR forecasts from the LSTM model and reservoir simulation model. Deep learning methods such as LSTMs have an inherent advantage in identifying trends in time-series data and making forecasts using the data. The existing literature has a limited number of studies that outline the use of deep learning methods for EUR forecasts and this study covers this gap by providing details analyses, best practices and workflows on the use of such methods for conventional oil reservoirs.
Machine learning models have worked as a robust tool in forecasting and optimization processes for wells in conventional, data-rich reservoirs. In unconventional reservoirs however, given the large ranges of uncertainty, purely data-driven, machine learning models have not yet proven to be repeatable and scalable. In such cases, integrating physics-based reservoir simulation methods along with machine learning techniques can be used as a solution to alleviate these limitations. The objective of this study is to provide an overview along with examples of implementing this integrated approach for the purpose of forecasting Estimated Ultimate Recovery (EUR) in shale reservoirs. This study is solely based on synthetic data. To generate data for one section of a reservoir, a full-physics reservoir simulator has been used. Simulated data from this section is used to train a machine learning model, which provides EUR as the output. Production from another section of the field with a different range of reservoir properties is then forecasted using a physics-based model. Using the earlier trained model, production forecasting for this section of the reservoir is then carried out to illustrate the integrated approach to EUR forecasting for a section of the reservoir that is not data rich. The integrated approach, or hybrid modeling, production forecasting for different sections of the reservoir that were data-starved, are illustrated. Using the physics-based model, the uncertainty in EUR predictions made by the machine learning model has been reduced and a more accurate forecasting has been attained. This method is primarily applicable in reservoirs, such as unconventionals, where one section of the field that has been developed has a substantial amount of data, whereas, the other section of the field will be data starved. The hybrid model was consistently able to forecast EUR at an acceptable level of accuracy, thereby, highlighting the benefits of this type of an integrated approach. This study advances the application of repeatable and scalable hybrid models in unconventional reservoirs and highlights its benefits as compared to using either physics-based or machine-learning based models separately.
One of the key parameters in identifying the success of fracture placement is to determine the number of perforations contributing to creating fractures from each hydraulic fracturing stage. One of the popular methods of estimating perforation contribution and near-wellbore pressure frictional losses is by performing step-down tests (SDT). The only drawback of this methodology is that the rate has to be dropped in a step-wise fashion, which introduces operational constraints. At the area of the implementation, frac treatment designed rate is 92 barrel-per-minute (bpm) and is reduced to zero to shut-in the well in step-wise fashion. The objective of this paper is to exploit the opportunity of utilizing the common practice of soft shutdown (SSD), where rate is dropped to 60 (bpm) then to zero at the end of the stage frac job to act like a mini SDT as a practical alternative solution. The proposed methodology entails conducting a tailored rate SDT on one well, and utilizing the shutdown period as a substitute to SDT on another well, specifically selected to be of the same conditions in terms of formation landing zone, stimulation treatment design and perforation count. A typical SDT is conducted by dropping the pump rate gradually in step-wise decrements. In this particular approach, two SDTs were performed within a single frac job, one at the beginning of the job before the introduction of proppant, and the other at the end after the flush period. Whereas in the SSD test approach, rate is dropped in only two steps at the end of the job and no time interval is specified. Pressure and rate are then selected as data points from each step, with instantaneous shut-in pressure (ISIP) considered as the final data point. It is important to keep as many variables fixed as possible in order to have the pressure response contributed by wellbore and perforation frictional components only. The selected data points are then plotted as pressure versus rate and matched with frictional losses and number of open perforations. The methodology capitalizes on the availability of SSD data, and evaluate its feasibility as a substitute to SDT. By performing this type of analysis, an estimate of perforation efficiency from both methodologies was achieved. Although two different results were retrieved from the SDT obtained from the beginning and the end of the frac job, the one performed at the flush stage was the focus of this study as it mimics the most realistic setting of perforation efficiency post treatment. Although lower number of data points are obtained from the SSD approach, it did not obscure matching the calculated pressure to the selected pressure-rate data points. In fact, the results from the SSD indicated a variance of as low as 2% when compared to SDT results from a mirror stage. This small variation demonstrated the technical and practical feasibility of utilizing SSD as a strong substitute to SDT, promoting the effectiveness of this robust methodology. Novelty of this approach lies within the utilization of readily available data retrieved from the original practice to substitute SDTs that could be operationally time consuming. The results from SSD tests validated the results from SDT, which allowed for the extrapolation of this approach to future wells within the same field without the necessity of performing any additional data acquisition.
Evaluating EUR and production forecasting in multi-frac horizontal wells completed in unconventional shale reservoirs during early, exploration and appraisal stages is very challenging. With the absence of suitable production facilities to handle produced fluids, production data are limited to short flow backs and extended production for some key wells to early production facility (EPF). This method uses the flow capacity as an intrinsic parameter that captures and reflects the major drainage mechanism and recovery characteristics of the well within the unconventional reservoir. Thus, the flow capacity may serve as a reference parameter that can be estimated from early-time data. This parameter may have the ability to reflect the future production behavior of the well in terms of cumulative production through proportional comparison of flow capacity to EUR. To show applicability of the workflow, it was applied to an existing unconventional project and showed consistent results. The workflow incorporates RTA analysis for several wells to create a single correlation that describes the relationship between Ac√k and EUR. Plotting gas flow capacity (Ac√k), estimated during the early 3-4 weeks of flowback, versus the estimated EUR from modeling techniques showed a strong correlation. The proposed workflow is designed to estimate EUR for wells completed in unconventional reservoirs during the early phases of development, where there is no production facility to handle produced hydrocarbons and flow back period is limited to cleanup only. The advantages of the proposed workflow over the currently available methods are: (1) Incorporating several wells analyses to create a single correlation that describes the relationship between Ac√k and EUR (2) The ability to evaluate EUR from as-early-as 3-4 weeks of flowback data. The early evaluation of these wells will expedite critical completion and development decisions which will impact project economics.
There are documented cases of machine learning being applied to different segments of the oil and gas industry with different levels of success. These successes have not been readily transferred to production forecasting for unconventional oil and gas reservoirs because of sparsity of production data at the early stage of production. Sparsity of unconventional production data is a challenge, but transfer learning can mitigate this challenge. Application of machine learning for production forecasting is challenging in areas with insufficient data. Transfer learning makes it possible to carry over the information gathered from well-established areas with rich data to areas with relatively limited data. This study outlines the background theory along with the application of transfer learning in unconventionals to aid in production forecasting. Similarity metrics are utilized in finding candidates for transfer learning by using key drivers for reservoir performance. Key drivers include similar reservoir mechanisms and subsurface structures. After training the model on a related field with rich data, most of the primary parameters learned and stored in a representative machine or deep learning model can be re-used in a transfer learning manner. By employing the already learned basic features, models with sparse data have been enriched by using transfer learning. The approach has been outlined in a stepwise manner with details. With the help of the insights transferred from related sites with rich data, the uncertainty in production forecasting has decreased, and the accuracy of the predictions increased. As a result, the details of selecting a related site to be used for transfer learning along with the challenges and steps in achieving the forecasts have been outlined in detail. There are limited studies in oil and gas literature on transfer learning for oil and gas reservoirs. If applied with care, it is a powerful method for increasing the success of models with sparse data. This study uses transfer learning to encapsulate the basics of the substructure of a well-known area and uses this information to empower the model. This study investigates the application to unconventional shale reservoirs, which have limited studies on transfer learning.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.