BP's data lake

摘要

BP presentations at the SPE's digital energy (DE) event and PPDM's annual Houston event demonstrate the growing role of open source software in high-end upstream information management. At DE, Mohamed Sidahmed presented work performed leveraging the 'R' statistical programming language to 'augment' operations monitoring by mining unstructured drilling reports. Unstructured, textual, hitherto the 'missing link' in the information workflow, contains valuable information on the root causes of deviations from plan and help address 'inadequate reaction to real time changes.' R-based text analytics leverage the collective knowledge stored in BP's Well Advisor, looking for interesting patterns. Visual representations (word clouds) integrate existing surveillance systems and can provide early warning of, for instance, pump failure. More fancy techniques such as 'latent Dirichlet allocation' helps identify precursor events hidden in the data. Reports with similar content can then be attached to the root causes of non productive time and rarer high impact events. Data driven learning is now embedded in BP's CoRE real time environment.