4DSEIS: Assimilating 4D Seismic Data: Big Data Into Big Models
Computer models of oil and gas reservoirs are used to predict future production and to predict where pockets of remaining oil might be located. Improving the reliability of these predictions by incorporating the maximum amount of information from seismic and well production data is the focus of this project. The result will be improvements in reservoir management and field development with subsequent reduction in CO2 footprint.
One key to improving the ability to forecast future events is to ensure that the models are consistent with historically observed behavior. It is generally necessary to adjust parameters in a computer model so that the predicted seismic data and production data agree with the data that was actually observed. Calibration of the model is challenging because that the amount of data that is provided by repeated seismic surveys can be exceptionally large, and calibration of a large reservoir flow model can be very difficult in that case. We will develop methods for calibrating large reservoir models to seismic and production data such that errors in forecasts are reduced and uncertainty is properly quantified. Additionally, both production and seismic data are subject to measurement errors and neither the reservoir flow model nor the seismic model is perfect. This, also, makes calibration difficult because overfitting noisy or biased data to an incorrect model can result in biased predictions that are too confident. We plan to develop methods for assessing the quality of calibrated reservoir models and forecasts on large problems. We will also develop methods for identifying the sources of imperfections in the reservoir model: missing parameters, missing processes, and parameters for which the prior uncertainty is too small. Finally, we will develop recommendations for a standardized work flow for 4D seismic history matching that results in reduced model error and better forecast ability.
Methods developed in this project will be tested on data from real fields, including the Edvard Grieg Field and the Norne Field.