Origin Energy employs AI for well risks
An artificial intelligence system can predict the risk of faults in over 2200 wells.
Origin Energy has developed an artificial intelligence system for its sprawling Queensland coal seam gas operations that allows it to predict the risk of faults in over 2200 wells, as part of a goal to boost production at its Australia Pacific LNG project.
The Sydney-based energy producer, which operates the gas fields for APLNG, has been trying to improve its understanding of why certain wells would struggle to restart after the gas system had been closed for either maintenance or an unplanned shutdown.
Its data scientists fed over 4700 historical well “events” into a machine learning model which then ranks 27 different features based on the chances they would contribute to a well failure. Typically, coal particles gunked up inside the well can be a common problem, along with the mix of gas and water in a well and how long the unit has been producing.
After shutdown, the average risk of a well failing upon start-up was 13 per cent, Origin data showed. Prior to the machine learning tool being rolled out, its own engineering experts selected wells for shutdown and lowered this number to 8 per cent.
The AI system, developed internally at Origin, is now able to select wells with an average 5 per cent risk of failure, according to Aleta Nicoll, who runs Origin’s Brisbane central control room which manages the flow of APLNG gas.
“Now if we have a downstream impact we run the tool and it would tell us which wells to shut off because they’re the lowest risk of failure wells and the least likely to cause us an impact,” Ms Nicoll told The Australian.
With 2205 operated wells under the watch of the Brisbane control room, an added perk is the ability of the AI system to analyse the entire supply network simultaneously, compared with a worker who at best might be able to assess 30 wells a day. Some 880 individual data points are now fed into the control system for each well every 30 seconds.
“We can use this model to calculate the risk score and likelihood of a well failing or not failing,” Ms Nicoll said. “One of the problems in the past we have tried to solve is getting an engineer to understand how the well behaves, which involves getting deep into the science behind the well performance. In this methodology, we said ‘don’t worry about the exact science of a well but let’s just use the data that we’ve got and let the data tell us if we will or won’t fail’.”
Origin’s system was first rolled out in January before a major shutdown of the APLNG export plant set for May, in a bid to give Ms Nicoll and her team the ability to identify which wells to shut should less gas be required for processing. While that maintenance was delayed due to COVID-19, work is now under way running checks on the export facility, with the control room able to switch wells on and off remotely using some of the insights gained from data crunching.
“With coal seam gas you normally get the well on and leave it on and we try hard not to take wells off,” she said.
“But when we do need to go through that process, we can now identify the lowest risk and manage the process more effectively.”