Combining Experimental Design with Proxy Derived Sensitivities to Improve Convergence Rates in Seismic History Matching

Abstract

Abstract An assisted history matching method has been developed where simulations are quantitatively compared to observed 4D seismic and production data and then updated in an objective manner. The approach uses experimental design to initialise a stochastic parameter search but also to derive a proxy model of the misfit between observations and predictions. The proxy model is then used to guide the stochastic search algorithm to speed up convergence. Updating of the proxy model periodically during the search improves convergence further. The approach was applied to the Schiehallion UKCS field where we use 6 years of production along with six 4D seismic surveys. The proxy model, a quadratic regression equation, was derived for the combined misfit of both production and seismic data. Transmissibility of a number of barriers and faults as well as permeability and net:gross (NTG) were modified to improve the history match. The approach improves the efficiency of finding the best models. Experimental design reduces the number of models to initialise the stochastic approach by a factor of 10. Continuous updating of the proxy not only leads to three times faster convergence but also enables better models to be found. History matching without experimental design or proxy derived gradients reduced the production misfit by sixty six percent while our new method gained a further ten percent reduction. We improved the exploitation of misfit information without compromising exploration of the parameters. History matching is a very important activity during the development and management of petroleum reservoirs. Obtaining matched models is fundamental to ensure reliable future forecasts, and give an improved understanding of the geological description and the dynamic behaviour. Improved recovery factors can then be achieved by accurately locating infill wells and maintaining production longer.