UPThèses

This thesis proposes a sequential inversion methodology for calibrating highly-heterogeneous reservoir models on well test data. The design and the application of that methodology are performed on an experimental hydro-geological site settled on a karstic and fractured limestone aquifer located near to Poitiers, France. The methodology involves two successive steps that are: first, the inversion of facies petrophysical properties; second, the gradual deformation of the facies distribution. By proceeding this way, the gradual deformation method, applied both globally and locally, improves the distribution of facies while keeping the previously optimised petrophysical properties. The fairly good capability on the resulting model to predict well responses allows to consider the gradual deformation as an efficient and robust method to find a facies geo-statistical realization matching at best flow data constraints. Alternative implementation of the sequence above are studied, by simply changing the schedule of the calibration steps. Efficiency and numerical performances of the methodology are also assessed by changing the number of gradual deformation parameters. Finally, it is investigated on the possibility to improve calibration by means of alternative flow modelling approaches. Whereas the adoption of a dual-medium model does not change significantly the simulated pressures, a more accurate modelling of conductive bodies by using a refined grid improves the prediction of short-time well responses. Further works could tentatively address the gradual deformation of object-based models for mimicking flow features of karstic and/or fractured heterogeneous reservoirs.