Oil & Gas - Maximising Recovery Programme (formerly SHARP) IOR Views

Introduction

Management of an oil or gas field depends upon accurate reservoir characterisation and prediction. Better decisions can be made when seismic data, geological modelling and history matching of flow simulations are fully integrated. Time-lapse (4D) seismic is gathered routinely in many fields and captures the effect of changes in fluid properties (e.g. pressures and saturations) spatially.  Much more information is then available for history matching than is conventionally at hand from sparsely separated wells.

To bring 4D seismic into the modelling workflow, we have developed a quantitative approach to Seismic History Matching (Figure 1). We generate multiple models, which are converted to predictions of seismic data (i.e. maps or volumes of attributes such as RMS amplitudes and their changes) as well as well production. The predicted data is then compared to the equivalent observed data and the resulting misfit is used to construct new models.  Our automatic history matching method improves on the classical approach, where the engineer manually adjusts parameters in the simulation model. We also use a global optimisation method to improve on gradient-based methods, which are good at finding local minimum misfits but can fail to find the global maximum. Finally sufficient models are generated to enable uncertainty analysis of reservoir parameters, saturation and pressure changes as well as the rock physics parameters.

The project is now in its second phase, which started in March 2005. The first phase ran from May 2002 to July 2004.

Figure 1: Seismic History Matching workflow - beginning with an initial ensemble of models, the workflow may be passed through dozens of times generating thousands of representations of the reservoir.

Phase I Results

Phase I was carried out with sponsorship from seven companies (Amerada Hess, BG Group, BP, ChevronTexaco, Kerr McGee, Shell and Total) secured through the UK DTI ITF programme. During that time, our approach was applied to a UKCS dataset.

Some of our results include:

• We derived a good match between predicted and observed seismic allowing for noise and some anomalies in the data (Figure 2). The match to well data was also improved, particularly at the injectors.
• We found that the static geological model should be constrained to the base line survey, in particular where net to gross is obtained from RMS amplitudes.
• Several parameterisation schemes were applied including an object model for channelised turbidites as well as geostatistical methods such as the pilot point method with Kriging and also simulated annealing.
• Simultaneous updating of several parameters was possible including permeabilities, fault transmissibilities and the petro-elastic transform parameters.
• History matching using a synthetic case showed that seismic data was essential to find a unique match.
• Improved calculation of the seismic data covariance matrix, reducing the calculation time by 90%.

For more details see Stephen et al., 2005. Sponsors of Phase I may obtain the final project report also.

Figure 2: Comparison of (a) observed and (b) predicted 4D seismic attributes for a sector of our field study reservoir after history matching. Both datasets have been normalised to the monitor surveys and represent the difference between surveys taken prior to and after the first year of production. Red indicates softening of the reservoir due to pressure build up or gas evolution while blue indicates stiffening due to pressure drop or water saturation increase. Vertical injectors I1, I2 and I3 are indicated as blue circles. Horizontal producers P1, P2 and P3 are indicated with red lines and red circles.  Producer P1 is inactive in this period.

Phase II

Our Phase I dataset has been augmented with recent seismic and production data. The new data is of higher quality and we are investigating its impact on history matching. We are also working on a new dataset which offers alternative seismic attributes with a different flow response and which must be modelled in volumes rather than maps.

We are continuing research under two main themes.

Value of Seismic

We are investigating the degree to which 4D seismic improves:

• spatio-dynamic reservoir characterisation.
• the predictive capability of models including fluid displacement and pressure changes; essential for reservoir management and well placement.
• planning of future surveys to maximise information.

Model Updating

The choice of geo-model and the update mechanism is important in the history matching workflow and we are studying:

• appropriate geo-modelling and updating strategies, comparing conventional engineering schemes versus those that are more geologically robust.
• the impact of seismic data quality and the model error on updating.
• the rate of convergence of history matching towards better simulations as a function of the above.

The following sponsors are thanked for their contribution to Phase II of the project: BG-Group, BP, Shell and Total as well as the DTI OG-MRP programme. We interact with sponsors regularly to discuss the fields we are studying and hold steering meetings every six months. Sponsors receive presentations from the meetings and advance viewing of forthcoming papers electronically. We also deliver copies of research code that has been developed during the project.

We continue to seek sponsors for Phase II and welcome enquiries. Please see www.pet.hw.ac.uk/research/shm/index.htm for more details on the project.

Reference

Stephen, K. D., Soldo, J., MacBeth, C. and Christie M. 2005. Multiple Model Seismic And Production History Matching: A Case Study. SPE94173, Proceedings of the 14th Europec Biennial Conference held in Madrid, Spain, 13-16 June 2005.