The aim of this project is to find out the correlation between the quality of the history match to that of a forecast. It is generally assumed that the better a model fits the past performance of a reservoir the better it will predict the future performance. In this thesis we challenge this assumption and aim to demonstrate the circumstances under which an apparently good history match leads to a poor prediction. An initial investigation of the relationship between the quality of a forecast and that of the history match was made using the IC fault model. This was then extended to a systematic study of a 2D reservoir model with log normally distributed permeability fields to demonstrate the fact that this correlation between the qualities of the history match and the forecast depends on various factors like water breakthrough times, extent of heterogeneity and the duration of history match and forecast period. Generally, it can be said that if both the base and compared case have their water breakthroughs within history match period, the quality of the forecast is good. It is also very interesting to find out that not always good history match leads to a good forecast and bad match leads to a bad forecast but sometimes depending on the forecast duration a moderately bad history match can lead to a good forecast.
The quality of the match can be correlated with the ratio of the breakthrough times of the base and the compared cases. We have fitted curves describing the relation between these two as a function of heterogeneity level. Heterogeneity has been controlled by changing the standard deviation of the log normally distributed permeabilities around a mean value.
During the third phase we have developed a 3D reservoir model with 5 vertical layers and log normally distributed permeability fields. We repeated the methods applied for the 2D model but only at two different heterogeneity levels to save the computational time.
We found that Quantitative results might be case specific but qualitative results are similar for both 2D and 3D models. The main result is to find a quantitative prediction of how the quality of the forecast deteriorates and how rapidly one can go from the situation where the quality of the forecast is good to where it is very poor.