While the increasing energy demands of AI are worrying, some strategies can also support make power grids cleanser and more efficient.
Artificial intelligence (AI) has depicted headlines lately for its rapidly growing energy demands, and mainly the increasing electricity usage of data centers that permit the training and deployment of the recent generative AI models. But it’s not all bad news — some AI tools have the potential to reduce some kinds of energy consumption and allow cleaner grids.
One of the most promising applications is using AI to optimize the power grid, which would enhance efficiency, increase resilience to intense weather, and allow the incorporation of more renewable energy. To learn more, MIT News spoke with Priya Donti, the Silverman Family Career Development Professor in the MIT Department of Electrical Engineering and Computer Science (EECS) and a fundamental investigator at the Laboratory for Information and Decision Systems (LIDS), whose work targets on applying machine learning to optimize the power grid.
Q: Why does the power grid require to be optimized in the first place?
A: We need to maintain an exact balance between the amount of power that is put into the grid and the amount that comes out at every moment in time. But on the requirement side, we’ve got a some uncertainty. Power corporations don’t ask clients to pre-register the amount of energy they are going to use beforehand of time, so some estimation and prediction must be achieved.
Then, on the deliver aspect, there’s commonly some variation in costs and fuel availability that grid managers need to be aware of. That has become an even larger trouble because of the incorporation of energy from time-various renewable sources, like solar and wind, in which uncertainty in the weather may have a major effect on how much power is available. Then, on the same time, relying on how power is flowing within the grid, there’s a some power misplaced by resistive heat on the power lines. So, as a grid operator, how do you make sure all that is working all the time? That is where optimization comes in.
Q: How can AI be maximum beneficial in power grid optimization?
A: One way AI can be useful is to use a combination of historical and real-time data to make more specific predictions about how much renewable energy will be available at a certain time. This ought to lead to a cleaner power grid by permitting us to handle and better use of these resources.
AI could also assist tackle the complex optimization troubles that power grid operators should solve to balance supply and demand in a way that also reduces costs. These optimization troubles are used to determine which power generators should generate power, how much they produce, and when they have to produce it, as well as when batteries need to be charged and discharged, and whether we can use flexibility in power loads. These optimization issues are so computationally costly that operators use approximations so they can solve them in a viable amount of time. But these approximations are often wrong, and while we combine more renewable energy into the grid, they’re thrown off even farther. AI can assist by using offering more correct approximations in a faster way, which may be deployed in real-time to support grid operators responsively and proactively manage the grid.
AI could also be useful in the planning of next-generation power grids. Planning for power grids needs one to use massive simulation models, so AI can play a huge role in running those models more correctly. The technology also can assist with predictive upkeep by detecting where anomalous behavior at the grid is likely to take place, lowering inefficiencies that come from outages. More widely, AI can also be applied to boost up experimentation targeted toward developing better batteries, which would permit the incorporation of more energy from renewable sources into the grid.
Q: How should we think about the pros and cons of AI, from an energy sector perspective?
A: One essential thing to remember is that AI refers to a heterogeneous set of technologies. There are different types and sizes of models that are used, and one of a way models are used. If you are using a model that is trained on a smaller quantity of data with a smaller range of parameters, that is going to consume much less energy than a large, general-motive model.
In the context of the energy sector, there are plenty of locations wherein, in you use these software-unique AI models for the applications they are meant for, the cost-advantages tradeoff works out in your prefer. In these cases, the applications are enabling advantages from a sustainability perspective — like integrating more renewables into the grid and assisting decarbonization strategies.
Overall, it’s vital to think about whether the types of investments we are making into AI are clearly matched with the advantages we want from AI. On a societal level, I think the answer to that question right now is “no.” There is a lot of development and enlargement of a particular subset of AI technologies, and these aren’t the technologies that will have the most important advantages throughout energy and climate applications. I’m not saying those technologies are useless, however they are enormously resource-intensive, whilst also now not being responsible for the lion’s share of the benefits that could be felt in the energy sector.
I’m excited to develop AI algorithms that admire the bodily constraints of the power grid so that we are able to credibly deploy them. This is a hard trouble to solve. If an LLM says something this is slightly wrong, as humans, we can commonly correct for that in our heads. But if you make the same magnitude of a mistake when you are optimizing a power grid, that can cause a large-scale blackout. We need to construct models differently, however this also offers an opportunity to advantages from our knowledge of ways the physics of the power grid works.
And more broadly, I think it’s essential that those of us in the technical community put our efforts towards to fostering a more democratized system of AI development and deployment, and that it’s completed in a way that is aligned with the needs of on-the-ground programs.
