I guess I am scratching my head a bit at this one.... You want to go a research paper in the field of reservoir simulation, in which you base your tests on simulation models.....

In other words, you will be assuming that simulation models can already well describe the scenarios you intend to simulate (otherwise their comparison with simulation tests will be flawed), which limits you to perhaps areas of increased efficiency rather than lack of understanding

Some areas where efficiency could be increased may be;
Methods to auto-tune simulation runs dynamically
More efficient methods of upscaling

Just a thought, but understand that many (I'm guessing) far more experienced minds than you would have looked at these already, so be realistic in your goals.