It’s no secret that artificial intelligence (AI) and machine learning (ML) hold the power to revolutionize industries. Using this technology, businesses have increased revenue, improved operations and solved complex problems.
So what promise does AI hold for the energy industry? Read on for examples of how companies are currently using AI to reduce GHG emissions:
Methane is the main component of natural gas. As a potent GHG, with over 70 times the climate warming impact of carbon dioxide over a 20-year timespan, methane is responsible for about 25% of the global warming we see today.
In oil and gas (O&G) production, methane is released into the atmosphere intentionally (e.g. venting from equipment during maintenance) and unintentionally, referred to as ‘fugitive emissions’ (e.g. equipment leaks). As the main component of natural gas, it’s also highly valuable; limiting methane leaks not only cuts GHG emissions, but is also a cost-saving measure.
On January 1, 2020, the federal government implemented new methane regulations for fugitive emissions, which require the upstream O&G sector to implement leak detection and repair (LDAR) programs to stop natural gas leaks. LDAR programs now require leak inspections three times per year, with corrective actions taken when leaks are found.
Global emissions monitoring company GHGSat recently demonstrated a method that uses ML algorithms to identify emission risks and leaks on the ground using satellite imagery. This gives operators the ability to identify patterns and trends over time, and make data-driven decisions quicker.
Profire Energy Inc., a technology company that provides solutions for industrial combustion appliances for the O&G industry, participated in two phases of Amii’s REMI program. Within the program, the team identified and explored a high-value and low-risk opportunity: to improve burner control by using ML technology to detect out-of-tune burners. Out-of-tune burners waste natural gas and increase GHG emissions -- as Profire’s install base is 75,000 systems, the savings potential of such a technology is significant. In addition, if this technology works, it could also be applied to an estimated 1.7 million wells in North America.
In the current market, there is no objective or cost-effective way to continually monitor and identify an out-of-tune burner for natural gas heaters. Field operators must make manual observations of burner performance based on flame colour, shape, noise, manual combustion analysis and other field observations. This is both costly and time-consuming and results in suboptimal burner management, wasted natural gas and increased GHG emissions.
Profire’s proposed solution involved installing additional low-cost sensors to the burner control system and streaming the data to a cloud server where an ML-based application would automatically predict out-of-tune burners. Field operators would then be notified to schedule priority maintenance to tune the burner thereby reducing wasted natural gas and lowering GHG emissions.
Unplanned critical equipment failures -- ranging from submersible pumps, wind turbines, electric generators, or natural gas compressors -- impact reliable operations, costs and GHG emissions.
Reliable operation of critical equipment often includes continuous monitoring and data collection. ML prediction models can be trained using historical operations data of failure events and normal operation; companies can then use these models to detect anomalous sensor readings and predict equipment failures before they happen.
Predictive maintenance using ML is the next step towards best-in-class reliable business operations. According to a 2017 Deloitte position paper on Predictive Maintenance: “On average, predictive maintenance increases productivity by 25%, reduces breakdowns by 70% and lowers maintenance costs by 25%”.
Building construction and operation generates nearly 40% of global GHG emissions -- an issue that energy management programs are now trying to tackle.
Using data on weather and building occupant’s activities, companies are now able to predict building energy use. As explored by the 2021 paper Prediction of office building electricity demand using artificial neural network by splitting the time horizon for different occupancy rates, two separate prediction methods are used to predict thermal and electrical loads of buildings:
With conventional building energy forecasts, there is often a gap between forecasts and measured energy use. The problem gets even more complicated when taking into account the unique activities of different building types; people will use office buildings differently than hotels, shopping centres and so on.
Overestimating or underestimating energy use can raise costs and GHG emissions. For example, if a building’s power use is above what was forecasted, it would need to purchase electricity on the spot market, which is more expensive and typically carbon-based.
Predicting building energy use more accurately with ML would reduce such instances, allowing building managers to balance operational performance with reducing energy costs and GHG emissions.
Are you ready to be a leader in AI adoption in the energy sector?
Nov 23rd 2022
Amii researchers present their work in the fields of reinforcement learning, natural language processing, data optimization and more at the 2022 Conference on Neural Information Processing Systems.
Nov 21st 2022
The position will help develop tools that apply statistical analysis and machine learning to determine a characteristic structure within multiple early warning signals driven by a disease outbreak.
Nov 16th 2022
On Sept. 2, Robert Paproski — CTO of Nanostics — presented "Lessons Learned Developing Predictive Models for Healthcare" at the AI Seminar.
Looking to build AI capacity? Need a speaker at your event?