Simply put, to make sense of what we observe. If we understand the hidden mechanisms that drive a system, we can anticipate how it will respond when conditions change—and, crucially for researchers and policymakers, when we intervene.
Take the electricity grid as an example. It is shaped every second by millions of decentralized decisions made by consumers, businesses, and power grid operators. Yet when planning the power grid, we still use models that rely on stylized, oversimplified assumptions about how these actors behave. When those assumptions are inaccurate, simulation models may misestimate the long-term value of engineering investments and policy interventions.
My research bridges that gap. I study how the real-world behaviour of emerging electricity-sector actors shapes the future of our power system.
Let me give you a quick example.
In my recent work on AI data centers, I developed a calibrated, bottom-up model [Read the paper]. I found that when batch workloads, such as training jobs run on university computing clusters, are colocated with inference workloads from services such as ChatGPT, overall electricity demand appears much smoother from the grid’s perspective. At the same time, the data center still produces substantial short-term power ramps.
Why does that matter?
Smoother overall demand means that the grid may require less load-following reserve to manage normal fluctuations. However, persistent short-term ramps mean that system operators still need fast-ramping resources, such as batteries, to maintain reliability.
Let me give you another example.
Studies of electricity demand in Germany, the United States, and Australia show that price responsiveness can vary over time and across hours of the day. However, these studies do not always explain the mechanisms behind this variation.
I observed a similar time-varying pattern in Bitcoin-mining demand response [Read the paper]. In recent work, I traced it to a clear breakeven-cost mechanism [Read the paper]. When mining revenues are low, an increase in electricity prices makes more machines unprofitable, so demand falls rapidly. However, when mining revenues are high, the same price increases have much less effect on demand.
If grid operators assume that this demand is always flexible, they risk overestimating the response available during a critical system period.
In both examples, failing to account for these internal operational and economic states could cause power grid operators to misestimate the need for ancillary services, leading to higher electricity costs for the consumers.
This brings me to the core of my work.
Rather than treating electricity demand as a static load profile or a simple demand curve, I look inside these systems to understand the mechanisms that generate their demand. I then integrate these models into power-system simulations to examine how individual decisions propagate through the grid and influence planning, reliability, and electricity markets.
These insights help us develop better electricity markets, planning methods, and control strategies. I then use causal inference to test whether these interventions achieve their intended outcomes.
My ultimate goal is to develop a framework in which decentralized decisions support a reliable, efficient, and socially beneficial power system.
