Using Machine Learning to Improve How Energy Gets to Market

By collecting data and utilizing machine learning, the midstream oil and gas industry is improving operational efficiency.

Getting energy to market is a challenge for all energy companies ranging from oil producers to electricity generators.  Transporting and processing the natural gas resources that are produced by exploration and production companies is one of the main focus areas of the midstream oil and gas industry.  While historically oil and gas companies have been criticized for being slow to adapt to technology, the recent prolonged low-price commodity market has caused the industry to turn to digital solutions in order to improve operational efficiency [1].

DCP Midstream, with over 50,000 miles of natural gas gathering pipeline and over 60 gas processing plants, is a major player in the midstream segment of the oil and gas industry and is utilizing digital technology in the day to day operations of doing business [2].  In this year’s first quarter earnings call, the DCP Midstream Chairman, President & CEO expressed reluctance to let innovation pass by the industry and explained the company’s plans to utilize data to help corporate decision making and asset optimization [3].

DCP Midstream has already taken monumental steps towards aggregating and interpreting company data. The first step for any organization to start implementing machine learning is the thoughtful collection of data.  Oil and gas operational data is often available but has historically been localized to rural plants.  To collect data into a central repository, DCP Midstream entered into an enterprise agreement with OSIsoft to utilize their PI System to aggregate data from companywide assets [2][4]. Aggregated data is then analyzed to translate operating conditions into financial performance and made readily available to personnel at all levels of the organization.  In addition, DCP Midstream installed an Integrated Collaboration Center (ICC) as a central location where data could be viewed, collectively interpreted, and informed decisions could be made with input from stakeholders across the organization. [2][4]

With the infrastructure in place for centralized data collection and interpretation, DCP Midstream started to utilize the data to look closely at specific equipment performance and maintenance.  The use of machine learning in equipment maintenance programs takes information from sensors embedded within machinery and looks for patterns in the data that indicate the machine needs attention before a failure event occurs [5]. Predictive analytics using machine learning allows for a precise maintenance program that predicts when failures might occur and provides sufficient time for preventative action [5].  For DCP Midstream, compressors are an integral piece of equipment used to transport and process natural gas.  DCP Midstream’s utilization of equipment data collection and specialized analytics has improved compressor reliability and saved costs on equipment parts [2][4].

In addition to the maintenance application specifically mentioned by DCP Midstream, it would be interesting to incorporate existing and forecasted weather data into the predictive data analytics.  This would allow for plant operating parameters to be optimized considering upcoming external weather conditions. Extreme heat and freezing cold can significantly impact operating parameters of natural processing plants.  Utilizing forecasted weather data could allow for predictive operating conditions to be determined and implemented in advance instead of making reactive changes.  Integrating predictive weather data into operator schedule assignments could also help to ensure operations staff are not deployed in advance of an impending snowstorm thereby avoiding the danger of adverse road conditions.

Predictive data analytics and machine learning could also be developed to help guide DCP Midstream’s overarching strategy and improve customer relations.  Machine learning algorithms could be developed to anticipate geographic locations and well performance of customer drilling to provide advance guidance on where infrastructure will be needed, and the capacity required.  This information could be used to anticipate bottlenecks in pipeline gathering systems and predict upcoming capital expenditure necessary to keep up with production.  These types of predictive analytics could serve as a great competitive advantage for DCP Midstream to capture customer production and exceed customer expectations of midstream infrastructure speed to market.

Utilizing digital technology has already improved DCP Midstream’s business and is likely to further guide operations in the future.  As machine learning is further deployed, how will DCP Midstream empower operations staff with data without decreasing headcount?  How will the organization implement a robust feedback loop from operations staff to ensure that data is accurate and not skewed by instrumentation and sensors in need of repair?

(708 words)

[1] Paul Hart, “What’s Next?,” Midstream Business, Volume 8, Issue 1, (Jan/Feb 2018): 26-34, ABI/INFORM via ProQuest, accessed November 2018.

[2] Tauna Rignall, “DCP Midstream – Enabling Business Transformation with the PI System: The DCP 2.0 Journey” PowerPoint Presentation, 2018,, accessed November 2018.

[3] Wouter T. van Kempen, Chairman President & CEO of DCP Midstream LLC, remarks made at ‘DCP – Q1 2018 DCP Midstream LP Earnings Call,’ May 8, 2018 3:00PM GMT.  From edited transcript provided by Thomson Reuters Streetevents,, accessed November 2018.

[4] Tauna Rignall, “DCP Midstream – Enabling Business Transformation with the PI System: The DCP 2.0 Journey” presentation given at OSIsoft PI World Conference, Barcelona, Spain, 2018,—Enabling-Business-Transformation-with-the-PI-System–The-DCP-2-0-Journey/, accessed November 2018.

[5] Grainger Editorial Staff, “Artificial Intelligence Takes on Preventative Maintenance,” Grainger KnowlegdeCenter, January 24, 2018,, accessed November 2018.


Machine Learning Reduces Dropouts at the University of Arizona


From Auto- tune to Auto- compose

Student comments on Using Machine Learning to Improve How Energy Gets to Market

  1. Really fascinating to see how machine learning can be used within the oil and gas industry, as use case I had not yet thought of! I particularly liked your evidence for its importance in helping optimise for weather impacts as I can imagine this mitigates a lot of frustration from employees. Your initial question of how to use large datasets and not reduce headcount raised some thoughts. Is this something which you believe could cause negative buy-in from employees should there be a potential to reduce headcount? Are there other ways in which large datasets and machine learning could help in order to make decisions about future plants and linking in changing historical oil prices? A couple of thoughts but overall thought this is a very impactful piece, thank you.

Leave a comment