An ANN Based Constitutive Model for Interbedded Hydrate-Bearing Sediments

Interbedded hydrate-bearing sediments (IHBS) are usually formed by horizontally overlapped layers of different thickness. To ensure the safety of hydrate gas production, a constitutive model characterizing this multi-layer structure is required. Since the thickness of each layer is only a few centimeters, which is much smaller than the minimum unit size of numerical simulation, it is necessary to consider IHBS as an anisotropic continuum to improve the calculation accuracy. Because of the diversity of layered occurrence forms and hydrate saturation distribution, it is difficult to establish a hand-craft constitutive model. This study states a novel approach to establish constitutive model by adopting ANN. First, the COMSOL software is used to generate sufficient artificial datasets of representative volume elements (RVEs) simulation. The Mohr–Coulomb model is used as the material model. Then, an ANN neural network is trained based on the artificial datasets to quickly and efficiently predict the constitutive response of RVEs. Finally, the prediction accuracy of the ANN based model is tested by synthetic data.