Seismic history matching with saturation indicators combined with multiple objective function optimization

Time-lapse seismic data offers additional information about the reservoir which can be used in history matching. Observed seismic data is often available as a relative measure of pressure and saturation changes and it can be difficult to compare directly to modelled data. Usually we need to calibrate the observed data or otherwise scale it for use in the misfit function. An alternative is to use a qualitative measure but this means that the production and seismic misfits can no longer be compared quantitatively. In this paper to solve this we examine the use of multiple objective functions to find better models that match the data.
We use a qualitative measure of the change of acoustic impedance in both the model and observed data domains to determine how much change takes places during production. Then we compare regions of change as a measure of similarity. With production data we use a conventional sum of squares misfit. A stochastic algorithm is used to find the best solutions. We modify the neighbourhood algorithm so that we accept models that will best fit either seismic or production misfits in a multi-objective fashion.
We demonstrate the approach on a synthetic model where the answer is known. We compare the results using a more conventional sum of squares misfit with one which uses multiple objective functions. We show that both the NA finds a better set of solutions that match both sets of observed datasets. We quantify the speed at which this occurs.
History matching with time-lapse seismic data adds additional constraints to conventional approaches with just production data. Models better reflect observations and are more reliable for forecasting. This leads to better decision making and planning in reservoir development.