2026-08-02T13:39:42Zhttp://doidb.wdc-terra.org/oaip/oai

oai:doidb.wdc-terra.org:67252026-06-10T10:00:03ZDOIDBDOIDB.GFZ

false4DOIDB.GFZ 10.5880/GFZ.1.5.2018.008 Jatnieks, Janis Janis Jatnieks GFZ German Research Centre for Geosciences, Potsdam, Germany Section 1.5 Geoinformatics, GFZ German Research Centre for Geosciences, Potsdam, Germany De Lucia, Marco Marco De Lucia 0000-0002-1186-4491 GFZ German Research Centre for Geosciences, Potsdam, Germany Section 3.4 Fluid Systems Modelling, GFZ German Research Centre for Geosciences, Potsdam, Germany Sips, Mike Mike Sips 0000-0003-3941-7092 GFZ German Research Centre for Geosciences, Potsdam, Germany Section 1.5 Geoinformatics, GFZ German Research Centre for Geosciences, Potsdam, Germany Dransch, Doris Doris Dransch GFZ German Research Centre for Geosciences, Potsdam, Germany Section 1.5 Geoinformatics, GFZ German Research Centre for Geosciences, Potsdam, Germany GFZ Data Services 2019 automated machine learning regression surrogate models reactive transport geochemistry EARTH SCIENCE > SOLID EARTH > GEOCHEMISTRY EARTH SCIENCE SERVICES > DATA ANALYSIS AND VISUALIZATION > STATISTICAL APPLICATIONS Jatnieks, Janis Section 1.5 Geoinformatics, GFZ German Research Centre for Geosciences Sips, Mike Section 1.5 Geoinformatics, GFZ German Research Centre for Geosciences Software 10.1016/j.egypro.2016.10.047 https://git.gfz-potsdam.de/sec15pub/Surrogate_playground_core/tree/master 1.0 GNU Lesser General Public License v2.1 Surrogate playground is an automated machine learning approach written for rapidly screening a large number of different models to serve as surrogates for a slow running simulator. This code was written for a reactive transport application where a fluid flow model (hydrodynamics) is coupled to a geochemistry simulator (reactions in time and space) to simulate scenarios such as underground storage of CO2 or hydrogen storage for excess energy from wind farms. The challenge for such applications is that the geochemistry simulator is typically slow compared to fluid dynamics and constitutes the main bottleneck for producing highly detailed simulations of such application scenarios. This approach attempts to find machine learning models that can replace the slow running simulator when trained on input-output data from the geochemistry simulator. The code may be of more general interest as this prototype can be used to screen many different machine learning models for any regression problem in general. To illustrate this it also includes a demonstration example using the Boston housing standard data-set.