Causal Machine Learning for Data Center Heat Recovery

What is it about?

The paper is about finding smarter ways to reuse the heat produced by data centers, which are big energy users in our digital world. Instead of letting that heat go to waste, it can be used to warm up homes - making them more sustainable. To do this well, we need good models that can predict how heat flows and how changes affect the system. Traditional simulations are slow, and regular machine learning doesn’t handle changes very well. That’s where causal machine learning comes in—it focuses on cause-and-effect and can better predict what happens when we make adjustments in the system. In the work we built a small, real-life data center with a heat recovery system to test these ideas. By running experiments and collecting data, we compared different traditional Machine Learning (ML) and causal ML methods. The results show that causal ML is better for predicting the effects of interventions compared to traditional approaches.

Featured Image

Why is it important?

We explore how to make data centers more sustainable by reusing the heat they generate. By building a small, real-world tested, we run controlled experiments and compare different AI methods to model heat generation in a real world scenario. This helps us understand ho causal ML can be a useful tool for modelling physical systems and guiding smarter, greener data center operations.