Seismic History Matching Using a Fast-Track Simulator to Seismic Proxy

Abstract

In this paper we propose a proxy model based seismic history matching (SHM), and apply it to time-lapse (4D) seismic data from a Norwegian Sea field. A stable proxy model is developed for generating 4D seismic attributes by using only the original baseline seismic data and dynamic pressure and saturation predictions from reservoir flow simulation. This method (MacBeth et al., 2016) circumvents the petro-elastic modelling with its associated uncertainties and also the need to choose a seismic full-wave or convolutional modelling solution, which are used in conventional simulator to seismic (sim2seis) modelling. The method is tested on an offshore field case study from the Norwegian Sea. In this study we firstly perform a check on the validity and accuracy of the proxy approach following the methodology of (Falahat et al. 2013) as a guide. The results confirm linear superposition between the pressure and saturation effects controlling the seismic data. Next a quasi-history matching is set up-here simulation model realisations are selected by random assignation of the key parameters to define a walk through solution space. After this, both the sim2seis and proxy modelling approach are compared for each realisation against a known reference case. The results show a mean seismic error of lower than 5%, which indicates the possibility to utilise a fixed proxy to model the 4D seismic. Finally, the full seismic history matching loop is implemented, where the sim2seis and the proxy-driven SHM are launched to find the optimal solution for our field. A particle swarm optimization (PSO) algorithm is applied as the optimisation tool, and only seismic data are used in the objective function. In both cases the algorithm converged after 30 iterations, and the optimal solutions of the two schemes are comparable. It is observed that the full sim2seis and proxy-driven SHMs are only marginally different, implying that solution space is similar in both cases. We also observe that in either case, matching to seismic data only can improve the production match. A unique feature of this study is the application of a seismic modelling proxy in the SHM scheme. Despite its relative simplicity, the approach is found not to bias the optimal solution of the more conventional SHM where the full physics of seismic modelling is applied. Meanwhile, this approach can save over 60% of the total computing time compared with the normal procedure, and this helps significantly to achieve a rapid and effective seismic history matching and better define uncertainty with a larger number of realisations.