Development of techniques for quantifying pressure and saturation changes from 4D seismic data applied to a North Sea field

Abstract

This thesis addresses the facilitation of 4D seismic data interpretation by its conversion to the pressure and saturation domain. I develop three techniques for the quantification of pressure and saturation changes from 4D seismic data. Each technique addresses a specific demand on the reservoir monitoring cycle. The first is a fast track approach designed to provide quick approximated solutions that aid in qualitative interpretation of the 4D seismic data. The second provides quantitative estimations to the pressure and saturation changes using a deep neural network trained on synthetic data. I present a study on the use of synthetic training datasets and conclude that using reservoir simulation results to populate the training dataset can improve inversion results considerably. The third method is a stochastic Bayesian inversion that uses reservoir simulation results as prior information, providing estimations to the pressure and saturation changes as well as their related uncertainties. All three methods are applied to a real 4D seismic dataset with multiple monitoring surveys, from the Schiehallion field. I compare the inversion results from each of the presented methods showing how each has their own value and can be used in different stages of the reservoir monitoring cycle. I show that the more robust solutions are necessary to carry out specific quantitative tasks such as seismic history matching. On the other hand, I show that it is possible to conduct valuable qualitative interpretations at a fraction of the turnover time, with the use of quick and simple inversion solutions. Hence, I conclude that all three inversion techniques explored here have their own value and can be applied at different stages of the 4D seismic interpretation cycle, to achieve specific objectives. I close by developing a method for seismic history matching of reservoir models that is performed in the pressure and saturation domain. This method uses the Bayesian inversion results as the 4D seismic historical data and the Bayesian uncertainty estimations as spatial localisation weights in the seismic data objective function. This study concludes that 4D seismic data assimilation can be improved with the use of the decoupled seismic information from the seismic inversion. It also shows the importance of having a good estimation to the uncertainties in the inversion results, when data assimilation is performed using inverted pressure and saturation maps.