Data-Space Inversion With a Recurrent Autoencoder for Naturally Fractured Systems

Data-space inversion (DSI) is a data assimilation procedure that directly generates posterior flow predictions, for time series of interest, without calibrating model parameters. No forward simulation performed in the process. DSI instead uses prior generated by performing O(1000) simulations on geomodel realizations. Data parameterization useful framework as it enables representation correlated time-series quantities terms low-dimensional latent-space variables. In this work, recently developed based recurrent autoencoder (RAE) applied with real naturally fractured reservoir. The parameterization, involving use neural network and an autoencoder, able to capture important correlations data. RAE training accomplished using results 1,350 An ensemble smoother multiple (ESMDA) provide samples. modeling work much more complex than considered previous studies includes 3D discrete fracture realizations, three-phase flow, tracer injection production, complicated field-management logic leading frequent well shut-in reopening. Results reconstruction new (not seen training), both RAE-based simpler approach principal component analysis (PCA) histogram transformation, are presented. shown better accuracy these reconstructions. Detailed then presented particular “true” (which outside ensemble), summary provided five additional models consistent ensemble. These again demonstrate advantages challenging reservoir case.