The Korea Institute of Geoscience and Mineral Resources conducted depressurization experiments at various gas hydrate saturation conditions, using mini pressure cells, to determine critical gas hydrate saturation. Their results indicated that it is necessary to carry out a numerical simulation in connection with the experiments to determine the critical gas hydrate saturation at numerous conditions that could not be incorporated into the experiments. In this study, a numerical simulation model, which simulated a mini pressure cell, was developed to verify the results of the aforementioned experiments. The changes in the fluid density and computer tomography values showed a similar trend. The validated numerical simulation model was used to determine the dissociation behaviors and productivities for various gas hydrate saturation levels (10–80%). As the gas hydrate saturation increased, the fluid permeability decreased. The gas hydrate dissociation was slowed by this correlation, and the production of gas and water was delayed at 50–60% gas hydrate saturation. Thus, the critical gas hydrate saturation was determined to be 50–60% for the experimental conditions of the field production test at the Ulleung Basin, East Sea, Korea. A pressure difference was noticed between the production and injection points due to the pressure propagation being delayed at critical gas hydrate saturation. Verifying the results of the experiments proved useful to understanding the dissociation and multiphase flow patterns in various depressurization scenarios.
9 November 2022 in Journal of Petroleum Science and Engineering
In this study, we propose a novel workflow to predict the production of existing and new multi-wells. To perform reliable production forecasting on heterogeneous shale formations, the features of these formations must be analyzed by classifying the formations into various groups; the groups have different production characteristics depending on the key factors that affect the shale formation. In addition, the limited data obtained from nearby existing multi-wells should be used to estimate the production of new wells. The key factors that affect shale formation were derived from the correlation and principal component analysis of available production-related attributes. The production of existing wells was estimated by classifying them into groups based on their production characteristics. These classified groups also identified the relationship between hydraulic fracturing design factors and productivity. To estimate the production of new wells (blind wells), we generated groups with different production characteristics and leveraged their features to estimate the production. Probabilistic values of the group features were entered into the input layer of the artificial neural network model to consider the variation in the production of shale formations. All the estimated productions exhibited less error than the previous analytical results, suggesting the utilization potential of the proposed workflow.
31 December 2022 in Journal of the Korean Society of Mineral and Energy Resources Engineers
Insufficient data in the early stages of production results in uncertainty of predicted values on applying decline curve analysis (DCA) using production data of shale reservoir. Additionally, there are limits in production-related factors, such as hydraulic fracturing or well completion conditions, that cannot be considered. In this study, Long Short-Term Memory (LSTM) neural networks were constructed to predict shale gas production by assuming the production time. It was discovered that prediction models could simulate production decline for new wells by learning the decline behavior of multiple shale wells. Moreover, the error could be reduced by using production-related factors for true vertical depth, gross perforated interval, stage count of hydraulic fracturing, and initial production rate, as additional input was confirmed. It may be possible to predict production based on LSTM at the early stages of shale gas production, where it is challenging to apply DCA, through the results of this study.
31 December 2023 in Journal of the Korean Society of Mineral and Energy Resources Engineers
Various studies have used decline curve analysis (DCA) to predict shale gas production, with recent proposals focusing on artificial intelligence-based machine learning techniques. Future production is generally predicted using production history as input data, but few studies have examined how to improve predictive performance by processing specific data from given data. This study used Long Short-Term Memory (LSTM) neural networks to predict the production of shale gas, and predictive performance was improved by extracting and using decline characteristics to reflect the decline trend of shale gas. Future predictions of shale gas production based on Long Short-Term Memory neural networks can be improved by processing data on decline characteristics from production history and using them as input data.
1 August 2019 in Journal of the Korean Society of Mineral and Energy Resources Engineers