For those with an interesting in reservoir characterisation as used in the oil & gas- and the geothermal sector we want show some results of a machine learning study we finished this weekend.
Recently, we did a deterministic evaluation of a finely laminated deep marine turbidite sandstone reservoir for OMV Norway. The study includes two wellbores – one with core cover parts of the reservoir and one without. We came to the conclusion that the deterministic results regarding sand volume and sand properties are associated with large uncertainty.
So we analysed the hydraulic flow units (HFU’s) based on core data and parameterized the core images and created three facies – Shale, Silt and Sand. Subsequently we used the Self Organising Maps (SOM) module of Interactive Petrophysics (the petrophysical tool we have chosen at WellPerform) to train and predict the HFU’s and facies away from intervals with core data.
The results are almost overwhelming good. This tool is doing a great job discriminating the three facies albeit many of the sands are thin way beyond log resolution.
The work was undertaken by our petrophysicist Søren A Christensen and we will soon make a couple of articles and post them here.
