OGE Header

Machine learning applied in U.S. oil and gas shale fields

A machine-learning solution allowed Denver-based Bonanza Creek Energy to monitor data drawn from its SCADA system to detect increasing risk of volatile organic compounds (VOC) or carbon-dioxide emissions.

05/21/2018


Figure 2: Self-service machine learning for engineers makes it easier to implement and use. Courtesy: FalkonryA machine-learning solution that can be used by subject-domain experts allows Denver-based Bonanza Creek Energy to monitor data drawn from its supervisory control and data acquisition (SCADA) system to detect increasing risk of volatile organic compounds (VOC) or carbon-dioxide emissions. In doing so the system identifies the potential emissions source and allows the situation to be addressed before emissions occur, said Martin Lohmann, production operations manager, Bonanza Creek Energy.

Bonanza Creek's operations are focused in Colorado's Wattenberg Field. Every well the company drills is extended horizontally and is fracture-stimulated. The company's strategy, its managers say, is to apply advanced technologies for more efficient and responsible extraction of oil and natural gas resources.

It's important to address the potential for emissions effectively and pre-emptively. "The one thing you don't want to do," said Lohmann, "is shut down wells just coming into production, which could lead to reservoir damage."

To take an example, if a vapor control system exceeds the permitted level of emissions, as defined by a given production output, per the regulations, operators must either shut down production for a period of time or pay hefty fines.

"To avoid fines against credits, the system identifies what VOCs are involved and what steps, if taken, would lead to emissions." Lohmann said.

Live, streaming well data is monitored by a SCADA/telemetry system, and soft-alarm ticklers sent to operators alert them that unless there is an intervention, a given state is likely to occur, Lohmann said. Pressures, levels, and temperatures are monitored, involving about 30 instruments at each production location, with signals that vary from 30 seconds to 15 minutes based on the criticality of the equipment involved.

"We have the tribal knowledge and experts eminently qualified to deal with these types of situations, but what if they occur at 3 a.m. in the morning? Moreover, we don't want rooms full of engineers to take on the time-consuming task," said Lohmann.

The solution, Falkonry LRS, uses multi-variate time-series data to discover patterns, recognize conditions, and predict critical events in industrial operations.

At Bonanza Creek, ambient air temperature was identified as an important factor when gauging the possibility of emissions. The results over time could influence the company to move to tankless systems, since it is with the storage tanks that the greatest possibility of emissions arises.

Lohmann is confident additional machine learning applications will be coming on line soon at Bonanza Creek Energy.



click me
click me