What generative AI can do for the power sector
The power industry is no stranger to the concept of artificial intelligence (AI) and its subsets machine learning and deep learning. The industry has been applying neural networks and expert systems for decades. It uses machine learning methods for forecasting and optimization, data processing, and detecting anomalies in equipment and operations.
A new era is unfolding, however, as advances in the scope and capabilities of AI technologies combine with ever-greater computing power and digital connectivity to create opportunities across the industry’s value chain.
Some of the opportunities are outlined in the new DNV report New Power Systems1 that discusses present and future applications of AI in the sector under the broad headings ‘generative AI’ and ‘discriminative AI’.
Generative AI is a catch-all term for models used to generate new data. These models learn the pattern of data they receive and then generate new data similar to what they have been trained on. The models include Large Language Models like ChatGPT.
Discriminative AI is a broad way to describe models used for making predictions and drawing inferences based on existing data. They identify patterns and use them to make decisions but, unlike generative AI, do not generate new data.
Generative AI now and in future
Consumer AI is where most of the hype about artificial intelligence is encountered, in particular the use of tools such as ChatGPT and Dall-E. This generative AI is being used to create content – text, videos, images, audio, and so on. It is also being applied in chatbots for customer service and organizational and industry help forums, and may replace how we currently search the web.
Generative AI’s applicability to power systems is currently limited to low-risk applications like productivity enhancement tools for staff (e.g. Microsoft Copilot) and customer service applications (e.g. chatbots). It is also in early research and development.
However, as tools and methods evolve to deal with challenges associated with generative AI, industrial applications will develop over time.2
In a sector which collects, stores, and relies on massive quantities of historical and real-time data of varying quality across its value chain, the growing ability of generative AI to learn and adapt in real-time suggests it could come to play a central role in maintaining the reliable, safe, and secure supply of power.
Managing power sector complexity
AI will become essential for maximizing flexibility and reliability of electricity networks as electrification spreads; grids expand and connect to more generation sources including intermittent renewables; more energy storage becomes grid-connected; virtual power plants become more common; power trading becomes more sophisticated and diverse; smart metering proliferates; and, as consumers also become suppliers of electricity, from electric vehicle batteries, rooftop solar, and so on.
Generative AI will thus be an increasingly important enabler of net-zero infrastructure electrification for decarbonizing heating from homes and industry and for transportation such as electric vehicles.
Generative AI in action in the power sector
Many players in the industry are already benefitting from using generative AI.
For example, researchers at the National Renewable Energy Laboratory (NREL) in the US have reported using invertible neural networks to improve the wind turbine blade design process 100 times quicker than through conventional methods.3
MIT’s Laboratory for Information and Decision Systems4 has shown that generative models trained on existing data can create additional, realistic data to augment limited datasets or substitute for sensitive datasets. The MIT researchers say that stakeholders can then use these models to understand and plan for specific what-if scenarios beyond what could be achieved with existing data alone. They suggest, for example, that data generated this way can predict the potential load on the grid if 1,000 more households adopted solar technologies, how the consequent load might vary during the day, and other parameters essential for future grid planning. The MIT research is being applied in a project aiming to support rural electric utilities and energy technology startups to mitigate risks involved in deploying the new technologies.
Pacific Northwest National Laboratory’s ChatGrid project in the US provides a generative AI-driven tool for grid visualization allowing engineers to pursue innovative designs while simulating various operating conditions.5
Towards fully automated electricity supply chains?
Conventional artificial intelligence is a familiar tool used in grid planning, line routing, and transformer placement. It is now frequently suggested that generative artificial intelligence could place within the power sector’s grasp the dream of fully automated electricity supply chains. The idea is that AI would orchestrate data analysis and operations across the entire chain, from grid planning, engineering, and design to asset management and grid operations.
Full automation could clearly be a game-changer for the effective, reliable, safe, and cyber-secure management of smart grids. However, it is generally thought that this vision is a long way off from becoming reality given the current state of generative AI and the challenges involved.
Full automation will also depend on collaboration between AI researchers, electrical engineers, industry professionals, and policymakers for furthering the adoption of these technologies.6
Challenges in using AI in the power sector
The biggest challenge to generative AI enabling full automation is that power grids operate under strict physical constraints imposed by the need for reliability and the laws of physics.
For one thing, today’s generative AI can ‘hallucinate’ – just make things up – when it does not know the answer to a question on the basis of the data it has been trained on or can access easily.
Generative AI is also ‘black box’ technology whose inner workings are opaque and not easily understandable to those deciding on its deployment and who have a duty to reduce operational, business, safety, regulatory, and reputational risks. They need assurance that a decision will not lead to widespread power blackouts.
There are also data quality and data format issues to consider, especially given the degree of collaboration implicit in the notion of fully automated, integrated value chains. Then there are questions of data privacy and cyber security, which are increasingly regulated for, such as in the EU AI Act (2024).
Last but not least are the high costs of designing, testing, and deploying generative AI and ensuring that the skills for using it are available in a market where they are in short supply.
Creating power sector trust in AI
DNV can provide the assurance needed to maximize and accelerate the use of generative AI to benefit the power sector while safeguarding grid reliability, integrity, safety, and security. The company has invested extensively since 2021 to develop recommended practices (RPs) providing guidance for organizations and industries including energy throughout the digitalization journey.
The RPs are guidelines and best practices for procurement, development, and operation of AI and other digital solutions. They include the recently published DNV-RP-0671 AI-enabled systems assurance, the world’s first RP on how to ensure the trustworthiness of AI-enabled industrial systems.
The company’s roles and services related to digital trust include, among others, risk assessments; training and self-assessment; verification, validation, and testing.
References
1 DNV (2024) New Power Systems report
2 Avci, Ezgi. (2023). GENERATIVE AI IN ELECTRICITY DISTRIBUTION: A QUALITATIVE EXPLORATION. Pressacademia. 10.17261/Pressacademia.2023.1788.
3 Glaws, A., et al. (2022). Invertible Neural Networks for Airfoil Design. AIAA Journal, Vol. 60, No. 5, May 2022
4 MIT.(2024). Generative AI for smart grid modeling. Massachusetts Institute of Technology. News release [online] 26 February 26, 2024
5 PNNL. (2024). ChatGrid™: A New Generative AI Tool for Power Grid Visualization. Pacific Northwest National Laboratory. News release [online], 22 February 2024
6 Richter, L., et al. (2022). Artificial Intelligence for Electricity Supply Chain automation. Renewable and Sustainable Energy Reviews, Vol. 163, 2022, 112459, ISSN 1364-0321
9/24/2024 12:39:00 PM