Search citation statements
Paper Sections
Citation Types
Year Published
Publication Types
Relationship
Authors
Journals
Prediction and analysis of well performance is based on accurate estimations of reservoir pressure, flowing bottom-hole pressure (Pwf), and flow rate. Among these three parameters, Pwf is the most challenging one due to its dynamic nature, inaccessibility, and varying conditions. This paper applies field data of high-pressure high-temperature sour gas well equipped with downhole sensors to study the accuracy of the industry standard fluid flow in pipes correlations, and to develop an AI model to compare with the correlations and the actual measurement. The objective is to determine what is the most accurate method we can reliably use as an alternative to downhole sensors. AI models were developed to estimate the flowing bottom-hole pressure and compared with actual measurements of flowing bottom-hole pressure measured by the Permanent Down-hole Measurement System, also known as PDHMS. The benefit of this application is illustrated in how fast the AI achieves results versus the conventional process of flowing bottom-hole pressure lengthy process, which involves careful study and selection of the fluid flow in pipelines correlations, such as the ten (10) models described in this paper. All of which require careful entry of and considerations of key fluid flow impacting well attributes. Which include completion design, tubing accessories, geothermal gradient, and deviation survey. These attributes are considered constant and none-changing based on completion design reports, however, they could vary during the lifetime of the well due to multiple factors, such as scale build-up with could create choke points and restrict flow in some tubing locations. On the other hand, an ANN model only needs the real-time measured data for a sufficient period of time, and would instantly be able to predict the output without the need of tedious entry and considerations of ever-changing well attributes. AI models were developed to estimate the flowing bottom-hole pressure and compared with actual measurements of flowing bottom-hole pressure measured by PDHMS. The results show that when an ANN model was trained on Well-A data, it was able to achieve 0.034% MAPE as compared to an average MAPE of 2.13% from the industry correlations. This is a significant difference which demonstrates the accuracy of AI models. This is especially convenient considering the short time required to train an AI model to the more involved and tedious process of developing a well structure and model through inputting PVT/well data, and yet not matching the accuracy of the ANN. Another notable result is how close the Well-B results between the correlations and ANN model, it’s worth mentioning that the ANN model used for Well-B was developed on Well-A dynamic data, and yet managed to achieve extremely satisfactory results (less than 1% MAPE). The application in this paper demonstrated the robustness of the ANN model with 30 simulation runs ranging from 5 neurons to 35. The 1st model (15 nodes) achieved an acceptable match in the test well (Well-B) with a MAPE of 2%. However, to demonstrate the flexibility of the ANN, 30 more runs was swiftly developed which resulted in an excellent 0.9% MAPE by the 13 neurons model. The possibility with ANN is enormous, much more accurate models could be developed considering the different aspects that go into the ANN such as number of hidden layers, number of neurons, training function, transfer function, and the performance function. In addition, this paper only investigated the ANN technique which is one of many AI algorithms, such as Fuzzy Logic and Support Vector Machine.
Prediction and analysis of well performance is based on accurate estimations of reservoir pressure, flowing bottom-hole pressure (Pwf), and flow rate. Among these three parameters, Pwf is the most challenging one due to its dynamic nature, inaccessibility, and varying conditions. This paper applies field data of high-pressure high-temperature sour gas well equipped with downhole sensors to study the accuracy of the industry standard fluid flow in pipes correlations, and to develop an AI model to compare with the correlations and the actual measurement. The objective is to determine what is the most accurate method we can reliably use as an alternative to downhole sensors. AI models were developed to estimate the flowing bottom-hole pressure and compared with actual measurements of flowing bottom-hole pressure measured by the Permanent Down-hole Measurement System, also known as PDHMS. The benefit of this application is illustrated in how fast the AI achieves results versus the conventional process of flowing bottom-hole pressure lengthy process, which involves careful study and selection of the fluid flow in pipelines correlations, such as the ten (10) models described in this paper. All of which require careful entry of and considerations of key fluid flow impacting well attributes. Which include completion design, tubing accessories, geothermal gradient, and deviation survey. These attributes are considered constant and none-changing based on completion design reports, however, they could vary during the lifetime of the well due to multiple factors, such as scale build-up with could create choke points and restrict flow in some tubing locations. On the other hand, an ANN model only needs the real-time measured data for a sufficient period of time, and would instantly be able to predict the output without the need of tedious entry and considerations of ever-changing well attributes. AI models were developed to estimate the flowing bottom-hole pressure and compared with actual measurements of flowing bottom-hole pressure measured by PDHMS. The results show that when an ANN model was trained on Well-A data, it was able to achieve 0.034% MAPE as compared to an average MAPE of 2.13% from the industry correlations. This is a significant difference which demonstrates the accuracy of AI models. This is especially convenient considering the short time required to train an AI model to the more involved and tedious process of developing a well structure and model through inputting PVT/well data, and yet not matching the accuracy of the ANN. Another notable result is how close the Well-B results between the correlations and ANN model, it’s worth mentioning that the ANN model used for Well-B was developed on Well-A dynamic data, and yet managed to achieve extremely satisfactory results (less than 1% MAPE). The application in this paper demonstrated the robustness of the ANN model with 30 simulation runs ranging from 5 neurons to 35. The 1st model (15 nodes) achieved an acceptable match in the test well (Well-B) with a MAPE of 2%. However, to demonstrate the flexibility of the ANN, 30 more runs was swiftly developed which resulted in an excellent 0.9% MAPE by the 13 neurons model. The possibility with ANN is enormous, much more accurate models could be developed considering the different aspects that go into the ANN such as number of hidden layers, number of neurons, training function, transfer function, and the performance function. In addition, this paper only investigated the ANN technique which is one of many AI algorithms, such as Fuzzy Logic and Support Vector Machine.
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.
customersupport@researchsolutions.com
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.