Artificial intelligence has grabbed the headlines lately rapidly increasing energy demandsand especially climbing power consumption of data centers enabling the training and use of the latest productive AI models. But it’s not all bad news – some AI tools have the potential to reduce other types of energy use and enable cleaner grids.
One of the most promising applications is using AI to improve the power grid, which can improve efficiency, increase resilience to extreme weather, and enable the integration of renewable energy. To learn more, MIT News he spoke to Priya DontiSilverman Family Career Development Professor in MIT’s Department of Electrical Engineering and Computer Science (EECS) and principal investigator in the Laboratory for Information and Decision Systems (LIDS), his work focuses on using machine learning to improve the power grid.
Question: Why does the power grid need to be developed in the first place?
A: We need to maintain a precise balance between the amount of energy put into the grid and the amount coming out at every point in time. But on the demand side, we have some uncertainty. Electricity companies do not ask customers to pre-register the amount of energy they will use in advance, so some estimation and forecasting has to be done.
Then, on the supply side, there are certain variations in the cost and availability of fuel that grid managers must respond to. That has become an even bigger problem because of the integration of energy from renewable sources that change over time, such as solar and wind, where the uncertainty of the weather can have a big impact on how much energy is available. Then, at the same time, depending on how the power flows in the grid, there is some power lost due to heat resistance in the power lines. So, as a grid operator, how do you make sure all of that works all the time? This is where optimization comes in.
Question: How can AI be more helpful in running the power grid?
A: Another way AI can be useful is using a combination of historical and real-time data to make accurate predictions about how much renewable energy will be available at a given time. This can lead to a cleaner energy grid by allowing us to better manage and use these resources.
AI can also help address the complex problems the power grid must solve in order to balance supply and demand in a cost-effective manner. These optimization problems are used to determine which electric generators should produce power, how much they should produce, and when they should produce, as well as when the batteries should be charged and discharged, and whether we can use flexibility in power loading. These optimization problems are so computationally expensive that operators use approximations to be able to solve them in the most feasible amount of time. But this balance is often wrong, and when we integrate renewable energy into the grid, it is thrown even further. AI can help by providing more accurate estimates in a fast, real-time manner to help grid operators react and manage the grid proactively.
AI may also be useful in planning next-generation power grids. Planning for power grids requires one to use large simulation models, so AI can play a major role in using those models effectively. The technology can also help in predictive maintenance by detecting when abnormal behavior in the grid is likely to occur, reducing inefficiencies caused by outages. More broadly, AI could also be used to accelerate experiments aimed at creating better batteries, which would allow the integration of more energy from renewable sources into the grid.
Question: How should we think about the pros and cons of AI, from the perspective of the energy sector?
A: One important thing to remember is that AI refers to a diverse set of technologies. There are different types and sizes of models used, and different methods used for models. If you use a model trained on a small amount of data with a small number of parameters, that will use much less power than a large, general-purpose model.
In the context of the energy sector, there are many areas where, if you apply these application-specific AI models to targeted applications, the cost-benefit trade-off works in your favor. In these cases, the applications enable benefits from a sustainability perspective – such as adding more renewables to the grid and supporting decarbonisation strategies.
Overall, it’s important to think about whether the kinds of investments we’re making in AI actually match the benefits 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 expansion of a particular set of AI technologies, and these are not technologies that will have major benefits for all energy and climate applications. I am not saying that these technologies are useless, but they are incredibly resource-intensive, while they are also not responsible for the lion’s share of potential benefits in the energy sector.
I am excited to develop AI algorithms that respect the physical constraints of the power grid so that we can use them reliably. This is a difficult problem to solve. If LLM says something slightly wrong, as humans, we can usually correct that in our heads. But if you make the same mistake in size when configuring the power grid, that can cause a major blackout. We need to build models differently, but this also provides an opportunity to benefit from our knowledge of how the physics of the power grid works.
And more broadly, I think it’s important that we in the technology community focus our efforts on promoting a more democratic system for AI development and deployment, and that it is tailored to the needs of downstream applications.
