A Closer Look at Non-Uniqueness During Dynamic Data Integration into Reservoir

Abstract

Abstract Characterizing heterogeneous permeable media using dynamic data such as transient pressure, tracer or multiphase production history typically requires the solution of an inverse problem. It is well known that such inverse problems are ill- posed in general and cannot uniquely constrain the detailed variations in reservoir properties. In this paper, we systematically examine the nature of the objective function during multiphase production data integration to explore the source of the non-uniqueness and the impact of some proposed remedies using prior information, structural analysis or smoothness conditions. The results are compared with ensemble inference from multiple realizations of permeability field generated from dynamic data without using any prior information. We show that continuous nonlinear trade-off between parameters is the major source of non-uniqueness during dynamic data integration. Another source is the presence of isolated, distinct local minima. Our results clearly demonstrate the danger associated with using dynamic data in isolation and the potential pitfall associated with the dictum finer is better'. We show that during reservoir characterization, there is a trade-off between high resolution and minimizing uncertainty and a compromise must be made. Finally, recognizing the general lack of prior information during reservoir characterization, the relative merits of different approaches to remedy the ill-posed behavior are examined. P. 155