Truncated conjugate gradient and improved LBFGS and TSVD for history matching
| Autor: | Gustavo T. Pfeiffer, Flávio Dickstein, Paulo Goldfeld, Renan Vicente Pinto |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
Mathematical optimization
Covariance matrix 010103 numerical & computational mathematics 010502 geochemistry & geophysics 01 natural sciences Computer Science Applications Computational Mathematics Reservoir simulation Computational Theory and Mathematics Factorization Conjugate gradient method Applied mathematics 0101 mathematics Computers in Earth Sciences History matching 0105 earth and related environmental sciences Mathematics |
| Zdroj: | Computational Geosciences. 22:309-327 |
| ISSN: | 1573-1499 1420-0597 |
| DOI: | 10.1007/s10596-017-9694-4 |
| Popis: | We propose a new algorithm for solving the history matching problem in reservoir simulation, truncated conjugate gradient (TCG), which involves a model reparameterization based on the factorization of the prior covariance matrix, C M = L L T . We also revisit the LBFGS algorithm, framing it into the same reparametrization, introducing M-LBFGS. We present numerical evidence that this reparameterization has an important regularizing impact on the solution process. We show how TCG and M-LBFGS, as well as TSVD, can be implemented without the need of actually computing the factor L. Our numerical experiments, including the PUNQ-S3 and the Brugge cases, indicate that TCG and M-LBFGS are effective schemes for history matching. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full text from SpringerLink