A novel extension of assisted history matching to honour geological constraints

Most gradient-based history matching (HM) tools consider only the minimum and maximum geological constraints in the algorithm and the applied calculations. As a result, the crucial link connecting geological, petrophysical and reservoir engineering data is diminished or, at times, completely lost. Therefore, an iteratively created model suffers significantly from the subsequent inconsistencies, e.g., porosity/permeability relations do not honour their petrophysical constraints. This leads to questionable admissibility of the entire model and, thus, calls for the necessity of removing or minimalising these inconsistencies. This dissertation focuses on finding an innovative solution to suggest better and plausible results. It investigates a new workflow that improves the geological consistency and suggests automation of the process of assisted history matching with a strong consideration of different geological constraints with respect to the different rock type definitions. The new workflow is provided as an external tool using available libraries in Python that can be applied as an extension for existing assisted history matching workflows. The rock typing coupled with the adjoint method is proposed to maintain the relationship between model parameters to different rock types, which are then iteratively updated during the history matching procedure. The rock types have different porosity and permeability ranges, relative permeability curves, and different connate water and residual oil saturation. The rock type is changed with corresponding parameters at the grid-block level based on the porosity and horizontal permeability change. The so-called rock-typing extension of the history matching workflow allows parameters to be modified co-dependently according to the rock type definition after the permeability adjustment suggested by adjoint-based sensitivity calculations. As proof of the concept, the simulation was carried out for the synthetic model with the same parameter distribution with and without the extended workflow. The results show obvious improvements in history matching quality in terms of geological consistency with fewer iterations or within the same amount of iterations with favourable objective function (OF) values. The simulation output achieves the target observed production profiles driven by the automatic joint correction of the saturation functions, including the initial saturations due to the rock type adjustments. Properly including the porosity-permeability correlations, priorities and model-specific details, the rock type can bind all the geological properties together, allowing consistent parameter changes for an improved history matching process. Since the base case is a product of geostatistical modelling, the level of certainty needs to be accounted for. Therefore a statistical distance is applied, which is the Mahalanobis distance. The Mahalanobis distance is associated with each rock type; it helps to guide the validation and correction step and determines the appropriate rock type based on the underlying statistical information. It also serves as the basis for the calculation to prove the concept. Overall, the novel approach has successfully managed to improve the geological consistency of the models during the history matching process, thereby improving the quality and reliability of the reverse simulation. Moreover, this extension includes workflow automation. It is an excellent practical standalone achievement that can potentially reduce the previously required work hours for the iterations to find the right set of data.