Lithofacies classification of a geothermal reservoir in Denmark and its facies-dependent porosity estimation from seismic inversion. (September 2020)

• Record Type: Journal Article
• Authors: Feng, Runhai; Balling, Niels; Grana, Dario
• Is Part Of: Geothermics. Volume 87(2020)
• Publication Date: 2020-09
• Subjects: Geothermal reservoir characterization -- Markov priors -- Artificial Neural Networks -- Lithofacies classification -- Porosity prediction -- Seismic inversion results; Hydrogeology -- Periodicals; Geothermal resources -- Periodicals; Énergie géothermique -- Périodiques; GEOTHERMAL ENGINEERING; GEOTHERMAL ENERGY; GEOTHERMAL EXPLORATION; Geothermal resources; Hydrogeology; Periodicals; Electronic journals; 621.44
• DOI: 10.1016/j.geothermics.2020.101854
• Languages: English
• ISSNs: 0375-6505

Abstract

Highlights: A novel system combining ANN and HMM is proposed to classify lithofacies of a potential geothermal reservoir in Denmark. Depositional rules and complex data distributions are considered at the same time. The classified lithofacies are then used as constraints for the prediction of porosity. Different regression neural networks are trained and applied within each type of lithofacies. Abstract: Characterization of geothermal reservoirs is an important step for exploration and development of geothermal energy, which is reliable and sustainable for the future. Based on the inversion results of seismic reflection data, lithofacies and porosity are predicted beyond well locations on a potential geothermal reservoir in the north of Copenhagen, onshore Denmark. To classify the lithofacies, a new system of Artificial Neural Networks-Hidden Markov Models is proposed to consider the complex spatial distribution of rock properties and the intrinsic depositional rules. Artificial Neural Networks can overcome the common Gaussian assumption for the distribution of rock properties. At the same time, the transition matrix in Hidden Markov Models provides the conditional probability for the lithofacies transitions along the vertical direction. After classification, the resulting lithofacies are used to constrain the porosity prediction, in which the Artificial Neural Networks is trained and applied within each type of lithofacies, as a regression process. The novelty of this approach is in the integration of statistics and computer science algorithms that allows capturing hidden and complex relations in the data that cannot be explained by traditionally deterministic geophysical equations. This workflow could also improve the prediction accuracy and the uncertainty quantification of the porosity distribution given rock properties.