Search

Conventional galaxy mass estimation methods suffer from model assumptions and degeneracies. Machine learning (ML), which reduces the reliance on such assumptions, can be used to determine how well present-day observations can yield predictions for the distributions of stellar and dark matter. In this work, we use a general sample of galaxies from the TNG100 simulation to investigate the ability of multibranch convolutional neural network (CNN) based ML methods to predict the central (i.e. within 1−2 effective radii) stellar and total masses, and the stellar mass-to-light ratio (M */ L). These models take galaxy images and spatially resolved mean velocity and velocity dispersion maps as inputs. Such CNN-based models can, in general, break the degeneracy between baryonic and dark matter in the sense that the model can make reliable predictions on the individual contributions of each component. For example, with r -band images and two galaxy kinematic maps as inputs, our model predicting M */ L has a prediction uncertainty of 0.04 dex. Moreover, to investigate which (global) features significantly contribute to the correct predictions of the properties above, we utilize a gradient-boosting machine. We find that galaxy luminosity dominates the prediction of all masses in the central regions, with stellar velocity dispersion coming next. We also investigate the main contributing features when predicting stellar and dark matter mass fractions (f *, f DM) and the dark matter mass M DM, and discuss the underlying astrophysics. [ABSTRACT FROM AUTHOR]

SUMMARY: Due to the strong heterogeneity of the reservoir, it is difficult to extend the classical reservoir parameter prediction model to the new work area. Traditional geological methods rely on experience, while machine learning algorithms lack consideration of reservoir heterogeneity. This makes it impossible for these methods to achieve good predictive performance in new work areas. In this paper, we discuss designing a multilayer deep transfer learning network (MDTL). The network enables accurate prediction of reservoir parameters in new areas based on core and logging data from mature areas. For the architectural design, we establish a network structure that realizes data interaction in different work areas. For network parameter optimization, we design a multilayer maximum mean discrepancy loss function. It guides network training to learn feature knowledge about data in different work areas. For logging data processing, we explore a support vector machine method to remove abnormal noise in logging data. We apply MDTL to predict shale gas porosity and total organic carbon content in southern Sichuan. Compared to state-of-the-art reservoir parameter prediction models, MDTL achieves the best performance in new well areas. The network integrates the idea of multilayer transfer learning, reduces the influence of reservoir heterogeneity and effectively ensures the prediction accuracy. [ABSTRACT FROM AUTHOR]