What is Southwestern’s data strategy in the downturn?
We are looking at a year without drilling so focus has moved to our production activity including a lookback at old data to see how to improve drilling and completion when things startup again. Almost all departments are now looking hard at data, even stuff that has practically never been looked at before. Our data managers are getting hammered with requests. One of our first responses to the new interest in data intensive computing was exposing much of our data in Spotfire. Making accessible this formerly siloed information from geological prognosis, geosteering, completion and production has been a big win for us. We consider Spotfire like Excel on steroids. It provides access to data from multiple sources with tabular and pivot table analytics.
What exactly do you mean by analytics?
For us today this means analysis, spotting trends and doing data clustering. We are in a transition stage. We have not yet embarked on predictive analytics or machine learning.
How is data stored?
With help from Petrolink we store real time Witsml data from our wells. Witsml data is deconstructed and stored in a large relational, SQL database. The idea is to be able to find the ‘pacesetter’ well and see why it was successful. We have been storing Witsml data since 2012 as part of our vision for land-based drilling information aggregation, proactive real-time systems, and post-drill analysis.
So how is the data stored? As tables, traces, XML in blobs?
Currently all the data is in the database and can be accessed via Witsml or by direct SQL query. But direct data access is challenging and Witsml is not conducive to analytics and machine learning. We are working architectural solutions to resolve both the query lag and on Witsml variabilities.
Have you considered using blobs?
We could us blobs or a NoSQL type data store. But right now we are working on accessing data in the existing store. We are also working on a similar project with production and operations data. We want to be able to do analytics on live data for production optimization with machine learning and using predictive analytics to scheduling downtime/maintenance/workovers.
You are storing lots of time series data. Did you consider PI?
PI carries a heavy commitment and we are determined to stay light and nimble. But we are in no rush, we are still looking for the best technological fit.
And Petrolink’s NoSQL repository?
It is an option. But in the first instance we are looking for a permanent storage option for cleansed data from the field. We are working on problems like ‘how many ways can you spell ‘block height!’ PetroLink’s realtime focus is through a relational (SQL) schema.
Are you planning for a data lake?
I don’t like the term but we sure have lots of data to archive. We have scada, Witsml, geosteering – you name it. And we got lots more data with the Chesapeake asset acquisition.
© Oil IT Journal - all rights reserved.