The mission of this group is to bring together utility professionals in the power industry who are in the thick of the digital utility transformation. 


The Complicated Path to Widespread AI Adoption in the Energy Industry

image credit:
Sayonsom Chanda's picture
Technical Lead Plexflo

Sayonsom Chanda, Ph.D. is a smart grid engineer and entrepreneur based in Boulder, Colorado, USA. He is the founder & tech lead of Plexflo - a Utility-focused EV infrastructure planning...

  • Member since 2021
  • 5 items added with 2,703 views
  • Mar 18, 2021

The energy industry has redefined what it means to be a human being over the last two centuries. From the era of James Watt's steam engines, energy has always been the harbinger of change, the hotbed of innovation, and a garden for genius to grow. As its complexity increased, the energy industry - notably, the electric utility industry - was the first to adopt computing technology to simulate power flows across the counties and countries and continents and achieve grid stability. When Artificial Intelligence and its various applications became known, it seemed the energy industry would go all-in, once again, to adopt the latest technological advances and then take them further. 

On the contrary, the power and energy industry's approach towards AI adoption has been more cautious. Despite AI's recurrent groundbreaking news, remarkable advances from AlphaGo to Amazon's relevant purchase recommendations and silly Snapchat filters, AI's applications in critical infrastructure industries, such as the energy industry, are limited due to three main reasons: (1) Explainability (2) Complexity of Contexts (3) Old-school tools still work.

Like their peers in other industries, professionals, and contractors in the energy industry are keen to improve productivity. Ninety-two percent of all supervisors in a utility-focused survey suggest that they are actively searching for tools that enhance their co-workers and team members’ employee satisfaction and productivity. 

Artificial Intelligence is machines that can detect patterns on behalf of a human and sets up natural-language based communication channels between man and machine. In other words, AI is supposed to be an inexhaustible, perpetually productive, and peak-performing proxy of the human mind - and deployable to places where it is unsafe or unnecessary for humans to go. 

AI works by mimicking the physiological tapestry of what's known and visible about our brains - the neural networks. Many computing nodes (also called neurons like their human counterparts) fed with data adjust their preset initial parameters (called 'weights')  in their nodes in response to the input data. This process continues until the combination of all their parameters adequately resembles or represents input data features. When these computing nodes have been exposed to tonnes of similar input data and are allowed to form their parameter combinations, they begin to detect trends on how to arrive at the correct weights, that is, those parameter combinations that define the neurons.  Human data-labelers approve this combination of parameters, and this process is called 'supervised learning' - still the most common way of building AI. There's a whole new domain of GANs (generative adversarial networks) - where machine endeavors to train itself - still require human hand-holding and feedback loops. 

Since arriving at these neural weights (i.e., the representative parameter combinations to describe an input) are governed by iterations and random adjustments - AI's mind is an impenetrable black box. Like a savant soothsayer, AI can attempt to foretell the future, but it can never explain the why behind its answers. Would you bring a 'beware the Ides of March'-totting soothsayer to operate the power grid, no matter how accurate he might be?

The lack of explainability is a significant reason for a slow pace in widespread AI adoption in heavily regulated industries, such as the energy industry. Regulators and law-makers need to know what's inside the black box because they are the ones liable for every loss, answerable for any adversity. As human decision-makers, we all prefer complete, understandable, unambiguous information and develop rationale and reasoning around it. 

The power and energy industry is a universe on its own. It is an entanglement of precision systems, lock-stepped, and synchronized to serve the rest of the world. It has taken centuries of human enterprise and genius to build the energy industry today. It is robust and reliable, though very sensitive to perturbations. AI's lack of transparency on how it works scares off practitioners in the power industry. Also, it means that AI will not function within the enormous diversity of contexts within which energy professionals work. There needs to be an AI tailored to every role and responsibility of an energy industry worker.

AI doesn't scale as advertised. AI is fundamentally only accurate as its training data sources and scenarios and use-cases extracted. Thus, models trained to work in the Arctic cold deserts will not fly in the sub-Saharan energy networks. Human data-labelers' pre-existing biases, their experiences (or limitations thereof) can seep into the training process. Data might be over-sampled or under-sampled. The natural language processing that transcribes meeting notes or translates Mandarin to Swahili - does not parse "CAN not working - stop production now," where CAN stands for control area network in the car manufacturing industry. Nor does it understand that a line worker wants specific types of cable configurations when she requests for "Osprey" and "Dove." 

Another major bottleneck in AI's integration with the energy industry is unclear boundaries between automation and artificial intelligence. Many industry-workers - especially workers in more manual and physical-work roles -  fear that robots, with their mechanical arms, AI-minds, above human capabilities, and of course, unsolvability - will take billions of jobs away. Unions harbor their fears and slow down the adoption of new processes. There needs to be a shift in the mindset - that AI is an empowering tool, like a car or a pen, that can uplift humans - by being an extension of their brain. It must also be highlighted that AI - and all its ever-evolving data-labeling tasks - has created more manual jobs in the computer industry instead of eliminating them. 

The energy industry thrives on many decades-old tools, such as the digital spreadsheets invented in the 1970s, by Dan Bricklin and his associates at Harvard. It took labor to build the well-crafted macros and dashboards of many utility industry projects, alongside their bookkeeping functionality.  Much of the industry's data analytics is locked in traditional spreadsheets and relational, locally-stored databases - that have delivered phenomenal performances over the last three decades. 

It is not fair to paint a picture that the energy industry is entirely at odds with the latest and greatest artificial intelligence. There are tools in place across multiple departments with deep learning-driven tools that analyze phone calls and sales patterns. Deep learning is also actively used by a small number of employees working on energy markets. 

However, there is greater scope for AI than what the industry already explored to date. It is time for a new kind of data analytics paradigm to become more popular within the energy industry.

In light of the ongoing COVID-19 crisis, the energy industry saw surplus supply and a change in usage patterns to which it had to adapt painfully. Utility crews had to deal with a tedious hurricane season, with socially distant, limited staff. AI could have played a role in prioritizing essential tasks for more satisfied employees. Artificial Intelligence-driven vegetation management can not only reduce unplanned power outages but can also reduce millions of dollars in operations and maintenance costs while bolstering customer satisfaction. 

Given the widespread changes in our grid infrastructure, databases will not only fail to extract value from the 21st-century data streams, but they might fail to function. The industry has seen record numbers of sensors, solar panels, electric cars, stronger ransom, Internet-of-Things, and Advanced Metering Infrastructure data floods through the 5G networks to the utilities. 

The number of parallel and orthogonal advances - whether it is in computing or data processing, or sensors - is hard to keep up with, especially if one is an energy professional, which itself is very demanding. The simultaneous profusion of new technologies makes us wonder - which to choose appropriately, a sort of buyer's dilemma. 

Thus, there is a need to streamline the AI applications for energy industry users. Such focussed AI for energy professionals can be called Energy AI. The complex DevOps and strenuous data science can be automated and tucked behind-the-scenes. A user from the energy domain can use AI without becoming a cloud-computing cyber-security expert while losing sight of why she began working on the cloud-computing project in the first place. 

Energy AI is artificial intelligence demystified, attenuated to energy-related problems, and accessible for power & energy professionals. Every business that is dependent on energy can use Energy AI confidently to prognosticate power interruptions, achieve peak productivity as individuals and teams, and in general, improve continuity of power supply to all downstream customers and business processes. 

AI does not have extraterrestrial origins. On the blue marble, AI was born here to solve real-world challenges that limit human abilities - whether personal or professional. It is now upon us, across several professions, to look beyond AI's silver bullet expectations and think beyond insurmountable computer science chops; one needs to leverage AI. 

How the energy industry adopts AI today will shape human lives for the next hundred years. Like steam engines led to power plants in the James Watt-era, who knows how the silly Snapchat's AI filters can one day correctly predict a power outage months in advance, helping a hospital avert a backup power crisis and save a life. 

Matt Chester's picture
Matt Chester on Mar 18, 2021

AI doesn't scale as advertised. AI is fundamentally only accurate as its training data sources and scenarios and use-cases extracted. Thus, models trained to work in the Arctic cold deserts will not fly in the sub-Saharan energy networks. 

I think this underscores how AI should really be thought of for grid functions-- not as a replacement to do it all, but as a tool that enhances the ability of the human workers. Combining worker expertise and AI tools will achieve better results than relying on just one or the other ever will. 

Sayonsom Chanda's picture
Thank Sayonsom for the Post!
Energy Central contributors share their experience and insights for the benefit of other Members (like you). Please show them your appreciation by leaving a comment, 'liking' this post, or following this Member.
More posts from this member

Get Published - Build a Following

The Energy Central Power Industry Network is based on one core idea - power industry professionals helping each other and advancing the industry by sharing and learning from each other.

If you have an experience or insight to share or have learned something from a conference or seminar, your peers and colleagues on Energy Central want to hear about it. It's also easy to share a link to an article you've liked or an industry resource that you think would be helpful.

                 Learn more about posting on Energy Central »