| Title | History Matching of Interference Tests in the Dogger Formation – Paris Basin: Comparison of a Bayesian Framework and a Multi-Objective Optimization Approach |
|---|---|
| Authors | Thomas SCHAAF, Delphine PATRIARCHE, Patrick EGERMANN, Tiphaine FARGETTON |
| Year | 2020 |
| Conference | World Geothermal Congress |
| Keywords | reservoir characterization, interference test, history matching, Dogger |
| Abstract | Getting reliable production forecasts is a key aspect of any geothermal project. A preliminary and mandatory (but not sufficient) step is to history match the associated geothermal numerical model(s) to existing observations and data. Our application cases are low enthalpy projects, located in the Paris Basin and exploiting the Dogger formation. Typical development plan consists of a doublet (a pair of one injector and one producer wells) which long-term sustainable flow rates and thermal breakthrough timing should be thoroughly assessed. The Dogger aquifer has been extensively developed in the 70-80’s, leading to 40+ geothermal doublets still in operation and most of the time densely implemented close to each other. This proximity of injectors and producers might lead to unexpected pressure and thermal interferences, potentially jeopardizing the long-term sustainability of the existing system or any new nearby development. The reservoir characterisation of the Dogger fractured limestone should include all available static and dynamic observations, from well logs to wells test and production data. An interference test between five doublets in a prospect zone to the South-East of Paris is used to better characterise and assess the hydraulic connectivity in this area. The static and dynamic models are built through an integrated and flexible workflow linking a geomodelling software together with a thermal reservoir simulator. Two alternative history matching approaches are applied and compared. Both of them capitalize on an extensive use of advanced Design of Experiments techniques for uncertainty space sampling, reliable proxy-models computations, global sensitivity analysis and optimization techniques. The first approach is based on the Bayesian framework for which a proxy-model of the likelihood function is computed through an iterative process. The a posteriori distributions of the uncertain parameters might then be derived from the a priori ones through the Bayes theorem. The former distributions define the history matched samples. The second approach is based on multi-objective functions optimisation, typically ending with a Pareto front in a two functions application case. Both approaches are cutting-edge history matching techniques, with comparable numerical costs and allow a better reservoir characterisation of the study area. Also, both of them do have specific advantages – like leaving to the decision maker the compromise to make on alternative competing history matching criteria - which would be described and discussed in details. |