Faster convergence in seismic history matching by dividing and conquering the unknowns
Abstract
The aim in reservoir management is to control field operations to maximize both the short and long term recovery of hydrocarbons. This often comprises continuous optimization based on reservoir simulation models when the significant unknown parameters have been updated by history matching where they are conditioned to all available data. However, history matching of what is usually a high dimensional problem requires expensive computer and commercial software resources. Many models are generated, particularly if there are interactions between the properties that update and their effects on the misfit that measures the difference between model predictions to observed data.
In this work, a novel 'divide and conquer' approach is developed to the seismic history matching method which efficiently searches for the best values of uncertain parameters such as barrier transmissibilities, net:gross, and permeability by matching well and 4D seismic predictions to observed data. The ‘divide’ is carried by applying a second order polynomial regression analysis to identify independent sub-volumes of the parameters hyperspace. These are then ‘conquered’ by searching separately but simultaneously with an adapted version of the quasi-global stochastic neighbourhood algorithm.
This 'divide and conquer' approach is applied to the seismic history matching of the Schiehallion field, located on the UK continental shelf. The field model, supplied by the operator, contained a large number of barriers that affect flow at different times during production, and their transmissibilities were largely unknown. There was also some uncertainty in the petrophysical parameters that controlled permeability and net:gross. Application of the method was accomplished because it is found that the misfit function could be successfully represented as sub-misfits each dependent on changes in a smaller number of parameters which then could be searched separately but simultaneously. Ultimately, the number of models required to find a good match reduced by an order of magnitude. Experimental design was used to contribute to the efficiency and the ‘divide and conquer’ approach was also able to separate the misfit on a spatial basis by using time-lapse seismic data in the misfit. The method has effectively gained a greater insight into the reservoir behaviour and has been able to predict flow more accurately with a very efficient 'divide and conquer' approach.