Search
  • Avatar Innovations

Digitization in the energy industry – the machine learning revolution

By Mike Henry, Burnet, Duckworth & Palmer LLP


In researching for this blog, I reached out to Brendan Bennett, a Reinforcement Learning Researcher at the University of Alberta, for his thoughts on how emerging digital technologies may be deployed in the energy industry. Brendan and I discussed how some recent landmark accomplishments in artificial intelligence might soon make their way into the energy industry.


Digital innovation in commercial spheres has largely been a story of improving efficiency and reliability while reducing costs. In the energy sector, these innovations have been a result of oil and gas companies doing what they do best: relying on talented engineers to improve on existing solutions. Improvements have quickly spread across the industry, bringing down costs and making processes more efficient.


I recently co-authored an article on the future of Artificial Intelligence in the Canadian Oil Patch, which discusses a number of examples of current innovations, including AI-powered predictive maintenance, optimized worker safety, and “digital twin” technology for better visualization of construction projects and formations. Looking forward, network effects, improving sensors, and algorithmic advances will continue to increase the rate of innovation and prevalence of new tech in the energy industry.



The most common example of network effects can likely be found in your pocket or in your hand right now. Because of the network effects of the smartphone, every new smartphone purchase increases the value of everyone else's smartphones by a little bit. Coupled with economies of scale in production, this means that the cost of these devices falls, while the value they provide increases. Some may view this as a virtuous cycle.


This same effect can be seen with sensors deployed in the oil and gas sector. Advances in technology and widespread use are pushing down the cost of sensors. This allows for more sensors to be deployed in a given application, creating a more complete and reliable data set when all measurements are taken together. Algorithms trained on larger, more comprehensive data sets can produce leaps in efficiency that were previously impossible.


DeepMind, an artificial intelligence research laboratory with a research office in Edmonton, recently combined prolific sensors with its own machine learning capabilities to reduce the cooling bill at Google's data centres by up to 40%.[1] Cooling is one of the primary uses of energy in a data centre; the servers running services like Gmail and YouTube generate a massive amount of heat. Given that Google already runs some of the most sophisticated energy management technology in the world at its data centres, an energy savings of almost half is astounding.


The same combination of plentiful sensors and advanced machine learning will soon be applied throughout the energy value chain, and promises to deliver those same astounding results. Accurate sensors providing clear insight into power use relative to a variety of factors will soon allow power grids run by machine learning algorithms to more accurately predict periods of peak demand, and provide the energy to satisfy demand with dramatic efficiency. These systems could also be designed to optimize for multiple variables, providing low cost power while also minimizing CO2 emissions.


More abstractly, AlphaFold, another project from DeepMind, employed deep neural networks to model protein folding, providing a solution to a 50-year-old grand challenge in biology.[2] The protein-folding problem has baffled biologists for decades. Cyril Levinthal, an eminent biologist, estimated in 1969 that it would take longer than the age of the known universe to describe all of the possible configurations of a typical protein through brute force calculation, an estimated 10300 possible configurations. AlphaFold's deep neural network can predict the configuration of a protein with stunning accuracy, in less time than standard complex experimental methods.


A similar approach might be applied to the problems of resource extraction and mapping of geological formations. Feeding the neural net with massive amounts of information generated from sensors that are cheaper and more plentiful in the oil and gas industry may lead to improvements in production efficiency. Further, the ability to map and test within the digital playground of these advanced neural nets may help producers avoid undesired consequences to human health and to the environment.


These advanced AI technologies will fundamentally change the way we explore for and develop our natural resources. Organizations like Avatar Innovations, which work with some of the province's leading entrepreneurs to bring innovations into the energy space, will be pivotal in helping Alberta lead the way in the development of these technologies.


I'd like to thank Brendan for his help with this post. If you're interested in learning more about his research into reinforcement learning, check out his website. If you have any questions on digital transformation in the energy sector please reach out to Mike Henry.

[1] https://deepmind.com/blog/article/deepmind-ai-reduces-google-data-centre-cooling-bill-40