Recurrent convolutional neural network for the surrogate modeling of subsurface flow simulation
This work addresses uncertainty quantification in subsurface flow simulations for researchers and engineers, but it is incremental as it combines existing techniques (SegNet and ConvLSTM) for a specific domain.
The authors tackled the problem of expensive Monte-Carlo simulations for subsurface flow uncertainty quantification by developing a surrogate model using a recurrent convolutional neural network, which improved performance over previous methods when handling time-series simulation data.
The quantification of uncertainty on fluid flow in porous media is often hampered by multi-scale heterogeneity and insufficient site characterization. Monte-Carlo simulation (MCS), which runs numerical simulations for a large number of realization of input parameters , becomes infeasible when simulation cost is expensive or the degree of uncertainty is large. Many deep-neural-network-based methods are developed in order to replace the numerical flow simulation, but previous studies focused only on generating several snapshots of outputs at the fixed time steps, and lack to reflect the time dependent property of simulation data. Recently, the convolutional long short term memory (ConvLSTM) is utilized to deal with time series image data. Here, we propose to combine SegNet with ConvLSTM layers for the surrogate modeling of numerical flow simulation. The results show that the proposed method improves the performance of SegNet based surrogate model remarkably when the output of the simulation is time series data.