Abstract
Construction of predictive reservoir models invariably involves interpretation and interpolation of limited available data and imperfect modeling assumptions that introduce significant subjectivity and uncertainty into the modeling process. A major source of uncertainty comes from the description of the geologic continuity that controls reservoir connectivity and, hence, governs the main fluid displacement patterns and the production efficiency. In particular, identification of anisotropy and heterogeneity in reservoir property distributions (e.g., permeability or porosity) under geologic uncertainty can be quite challenging. Using recent developments in sparse signal processing, we develop a prior model identification (PMI) approach using a mixture modeling framework that utilizes production history to identify and correct errors and biases in prior reservoir continuity (i.e., variogram) model. The method begins by generating a diverse "dictionary" of initial geologic continuity models that reflect a wide range of variability and uncertainty in the prior knowledge about reservoir connectivity. Given this diverse geologic dictionary as a mixture model and the production history, we implement an adaptive sparse reconstruction algorithm to automatically identify incorrect models of reservoir continuity (variogram parameters) during history matching iterations and replace them with more consistent prior model components to improve the production data match. The novel sparse signal recovery algorithm implements a model selection procedure that discriminates against incorrect prior models by heavily penalizing them and promotes models with significant contribution to matching the production data. Unlike conventional history matching techniques that assume a given prior model, this adaptive sparse PMI implementation provides a systematic mechanism for identifying consistent global continuity models from production data in addition to constraining the local variability in the property map to match the observed data. As such, it has significant practical implications for minimizing the subjectivity in prior model specification and the overall history matching workflows. We use two numerical experiments to illustrate the effectiveness of the sparse PMI for detecting and correcting errors in the initial specification of the prior reservoir continuity model. The results clearly demonstrate the suitability of the approach for reliable identification of reservoir connectivity under significant geologic uncertainty.
