The North American Electric Reliability Corporation recently released a white paper that examines artificial intelligence and machine learning in real-time system operations.
While AI, machine learning (AI/ML) and data science have been studied and developed for decades, technological growth and public attention have recently ballooned, the white paper noted.
“This has led to a tremendous amount of research and new solutions in the marketplace, affecting nearly every aspect of work and personal life. Innovations and discourse continue to evolve rapidly as excitement grows and investment and research branch into new areas.”
“Within the realm of real-time electric power operations, there is also recognition of the increasing complexity and complicatedness of the bulk power system given ongoing changes, with several new use cases that stretch the assumptions of the system (e.g., increasing concerns around cyber aspects, excess solar flowing onto the transmission system, significant load growth for electric vehicle charging, growing power requirements for AI/ML, cryptocurrency mining on blockchains, and other data center operations).”
The BPS “is the backbone of North America’s energy infrastructure. It is crucial for security and economic stability across both the continent and the nation and underpins our daily lives,” the white paper notes.
Managing the real-time reliability of the system “requires control room operators to possess increasing levels of cognition, attention, vigilance, knowledge, and abstract reasoning, invariably leading many to consider new AI/ML solutions. Because the BPS is the planet’s most complex sociotechnical system (a system involving complex humans and complex systems with complex interactions between them), many considerations are needed to minimize the risks to the system.”
This document is intended for decisionmakers, regulators, and end users of these technologies, particularly in real-time operations.
“It is not helpful to assert whether these technologies should be used in real-time operations, as surveys and interviews of key stakeholders throughout the industry show that this is already happening. The ‘genie’ cannot be put back in the bottle (and this document does not assert that it should),” the white paper said.
Instead, the document “provides guidance on the kinds of questions that one should ask about these technologies to thoroughly understand what they are capable of and what kinds of changes are needed to implement them properly.”
Previous technologies to come to market have fallen into typical patterns, leading to initial “bumpy” implementations with unexpected risks or adverse events, the white paper said.
The document “offers a path for real-time operations -- in which such adverse events are intolerable -- in an attempt to ensure that AI/ML technologies can be implemented in a way that maximizes the chances of a successful, reliability-enhancing deployment.”
There is strong recognition that many organizations across the industry are already considering AI/ML applications and making a variety of decisions from actively trying to avoid them to embracing them, the white paper said..
The document focuses on currently available technologies that are built, trained, and deployed to deal with specific situations "and would not work outside the domain of their training (e.g., a solar generation predictor could not be relied upon to predict wind generation), generally called narrow AI (or sometimes, weak AI)." This includes recent areas of rapid growth, including the ability to generate new content (with Generative AI algorithms such as GPT).
The document is divided into several sections:
- An overview of the technologies currently within the AI/ML space and distinctions between the types;
- An overview of the human factors associated with real-time control room operators working alongside AI/ML and the kinds of preparations that are needed to increase the likelihood of desirable outcomes;
- An overview of an anonymized survey and interviews from decisionmakers within the industry to provide a snapshot of the kinds of use cases and considerations that these entities are investigating
- An overview of AI/ML technologies and the roles that they can play in real-time operations
- An overview of cybersecurity risks that AI/ML technologies pose to the reliable real-time operations of the BPS
There is a history of new, advanced technologies initially failing, "not because of technological failures specifically but because of an insufficient focus on the interactions between humans and the system."
The paper advocates that these new systems can avoid those early pitfalls if they are properly designed, implemented, and used.
Implementing AI/ML systems into real-time operations "requires a strong relationship with the humans to allow the humans to effectively maintain, operate, and question the accuracy of the systems’ responses, both in real-time operations and in facilitating the identification of new ways of developing, testing, and deploying the systems to reduce the risks of novel errors."
These systems should not be implemented "under a mindset that sees 'humans as hazards' and seeks to avoid or disintermediate human involvement but rather one that seeks to bring the strengths of humans and systems together to achieve even higher levels of effectiveness and reliability," the white paper said.
"Similarly, successful implementation also requires recognition that, as support/aid tools, AI/ML systems will provide operators more mental bandwidth to monitor and strengthen the reliability of the system-- but not if that bandwidth is removed through the addition of more switch-tasking and distractions in the system operators’ daily work."
