Madeline Liao, ML Engineer
Q&A
Q: What kind of machine learning problems do you work on at Gridmatic?
The main ML challenge we tackle is time series forecasting. We focus on predicting things like wind and solar generation, electricity prices, and demand—essentially, anything that fluctuates over time in the power grid. What makes our work especially interesting is that we make probabilistic forecasts of these quantities. We don’t just care about the median prediction of what might happen tomorrow; we care about the fifth percentile or the 95th percentile event. This means that we have to forecast probability distributions, which is a much more difficult problem, but also (in my opinion) a lot more interesting to work on!
Q: What makes probabilistic forecasting so important in electricity markets?
In order to make risk-informed decisions, we need to understand all possible scenarios–not just the average scenario–of what electricity demand and prices might be in the future. We need to anticipate rare but impactful events, which makes probabilistic forecasting essential. Accurately predicting these long-tail events is crucial to the reliability and robustness of our operations.
Q: How does weather factor into all of this?
Weather is very predictive of electricity behavior. You can imagine that if there’s a huge heat wave, many people will be turning on their ACs at once, which will cause demand to spike and thus electricity prices to spike as well. So if we can improve our usage of weather forecasts, or improve our own weather forecasts, we can in turn improve the prediction quality of the quantities that we care about downstream. This is a huge area of research in the ML community, so I’m really excited about going in that direction.
Q: What kind of ML models do you work with?
We work primarily with generative models, which are a hot topic in the broader ML research community right now. It’s exciting because we’re constantly reading new papers, experimenting with the latest methods, and figuring out how to apply cutting-edge research to real-world grid problems.
Q: How closely does your work align with the academic research community?
Very closely! A lot of the open problems in ML research—especially around generative modeling and uncertainty quantification—are exactly the kinds of things we’re solving here. So there’s this constant back-and-forth between what’s happening in research and what we’re building at Gridmatic. It makes the work feel very current and connected.
Q: What’s the most exciting part of your job?
I love that our work sits at the intersection of machine learning, energy, and real-world impact. While I enjoy working on deeply technical ML problems, it’s important to me that those problems are aligned with a mission that I care about. This is exactly what I get to work on here at Gridmatic!