Seminar| Institute of Mathematical Sciences
Abstract: Oil and gas reservoirs are typically highly heterogeneous, but the detection methods are very limited. Therefore, the geological model parameters are highly uncertain. In consequence, the production prediction based on reservoir simulation is often inconsistent with the observed data. Therefore, histroy matching, i.e., the inversion of formation parameters based on production history, is a necessary step in reservoir engineering for accurate production forecast. However, due to the high computational costs of reservoir simulation, the existing inversion algorithms are often for MAP, while uncertainty quantification based on MCMC is still of low efficiency. To solve this problem, the surrogate modelling and the efficient inversion algorithms are studied. First, surrogates are built based on flow equation as well as statistics. Second, an efficient uncertainty quantification algorithm combining simultaneous perturbation stochastic approximation and variational Bayesian inference is established. Furthermore, a PINN algorithm for solving Darcy equations in highly heterogeneous reservoirs is studied, but the computational efficiency still needs to be improved.