Estimation of reservoir parameters such as porosity and permeability from production history and time-lapse seismic data is still challenging. When the object is a large hydrocarbon reservoir with a large number of unknown parameters, the problem will be even more difficult and needs to have geologically realistic solutions especially in adjoint-free gradient-based methods.

Replacing the original set of unknown parameters with a lower dimension group that captures the most important features of the field will be a proper procedure to speed-up the optimization procedure. This work proposes a method for parameter reduction by using sensitivity of principal components (PC's) to the observation data and combination of principal component analysis (PCA) and discrete cosine transform (DCT).

DCT is a robust, fast and efficient method that does not require heavy computing. However, the difficulty of including geological as well as bound constraints in some optimization algorithms can lead to solutions that are geologically unacceptable. PCA has the facility to implement both types of constraints. Then it can reduce the number of parameters and geologically acceptable results may be ensured. Reduction of the number of parameters will be done in two steps, first by PCA and then by DCT. Resulting in much fewer parameters and still provide solutions with the high number of accuracy from a geological point of view.

This paper evaluates the performance and advantages of combining these methods and choosing most sensitive components for parameter reduction in parameter estimation. The methodology is applied in noise-free and noisy semi-synthetic models inspired by real field data, based on data from the Norne Field offshore Norway. The results show that using most sensitive principal components to the observation data and combining it with DCT can be a useful tool for speeding-up this type of optimization routines.

You can access this article if you purchase or spend a download.