Prediction of oil production rate in multiple wells of a producing field applying combined deep–learning and optimization techniques; Fuel; Vol. 406, pt. A

Bibliographic Details
Parent link:Fuel.— .— Amsterdam: Elsevier Science Publishing Company Inc.
Vol. 406, pt. A.— 2026.— Article number 136847, 14 p.
Other Authors: Makarov N. S. Nikita Sergeevich, Al-Shargabi M. A. T. S. Mokhammed Abdulsalam Takha Sallam, Wood D. A. David, Burnaev E. Evgeny, Davoodi Sh. Shadfar
Summary:Title screen
Accurate prediction of oil production rate (OPR) is crucial for optimizing reservoir management and maximizing production efficiency. This study develops high-accuracy predictive models for OPR using well performance data from three wells in Norway’s Volve field. The key objective was to create a robust, generalizable model that leverages deep learning and optimization techniques to overcome the limitations of traditional methods. The methodology involved preprocessing the data with wavelet denoising and normalization. A hybrid non-dominated sorting genetic algorithm and long short-term memory (LSTM) model was used for feature selection, identifying seven key input parameters. Two deep learning models, LSTM and a convolutional neural network, were then hybridized with particle swarm optimization and the cuckoo optimization algorithm (COA) to fine-tune their hyperparameters. The results showed that the hybrid LSTM-COA model generated superior performance, achieving the lowest prediction error on the blind test well data (root-mean-square error = 2.1534 m3/day). Shapley additive explanation (SHAP) analysis identified bottomhole temperature as the most influential feature. This study’s novelty lies in integrating multi-well data, hybrid optimization, and explainable artificial intelligence into a single workflow, offering a practical tool for fast and reliable OPR forecasting to guide field development and optimize production in mature reservoirs
Текстовый файл
AM_Agreement
Language:English
Published: 2026
Subjects:
Online Access:https://doi.org/10.1016/j.fuel.2025.136847
Format: Electronic Book Chapter
KOHA link:https://koha.lib.tpu.ru/cgi-bin/koha/opac-detail.pl?biblionumber=686549
Description
Summary:Title screen
Accurate prediction of oil production rate (OPR) is crucial for optimizing reservoir management and maximizing production efficiency. This study develops high-accuracy predictive models for OPR using well performance data from three wells in Norway’s Volve field. The key objective was to create a robust, generalizable model that leverages deep learning and optimization techniques to overcome the limitations of traditional methods. The methodology involved preprocessing the data with wavelet denoising and normalization. A hybrid non-dominated sorting genetic algorithm and long short-term memory (LSTM) model was used for feature selection, identifying seven key input parameters. Two deep learning models, LSTM and a convolutional neural network, were then hybridized with particle swarm optimization and the cuckoo optimization algorithm (COA) to fine-tune their hyperparameters. The results showed that the hybrid LSTM-COA model generated superior performance, achieving the lowest prediction error on the blind test well data (root-mean-square error = 2.1534 m3/day). Shapley additive explanation (SHAP) analysis identified bottomhole temperature as the most influential feature. This study’s novelty lies in integrating multi-well data, hybrid optimization, and explainable artificial intelligence into a single workflow, offering a practical tool for fast and reliable OPR forecasting to guide field development and optimize production in mature reservoirs
Текстовый файл
AM_Agreement
DOI:10.1016/j.fuel.2025.136847