Mines professors using math to optimize energy flow

Getting from one point to another isn’t always a straight line. The most direct path might not be the quickest, and the quickest way forward might have unexpected barriers or delays.

So how do you ensure delivery of a material from one place to another is both efficient and successful?

Samy Wu Fung and Daniel McKenzie, both assistant professors of applied mathematics and statistics at Mines, have been thinking about this question for the last three years, working together in the Mines Optimization and Deep Learning (MODL) group. While there are powerful optimization algorithms already set up to tackle processing flow paths, they aren’t always able to take unexpected real-world barriers into account. Fung and McKenzie are working on combining the strengths of existing optimization algorithms with machine learning models, so outcomes are both ensured and the most efficient.

Fung gave the example of navigating downtown traffic. Finding the quickest path could be done using an optimization algorithm, but what about events like baseball games or construction that can impede the flow of cars? Machine learning models can use past information to handle these kinds of situations, while optimization algorithms provide guarantees on the path prediction.

“Optimization models will ensure success, but machine learning algorithms use historical data to take into account variability,” McKenzie said. “Integrating the two finds the quickest path and can guarantee arrival.”

For their research, they’re working on applying optimization models to a challenge that is crucial to everyday infrastructure—the power grid. 

Currently, machine learning models are a popular mechanism to predict flow through a power grid. But these models don’t always consider physical constraints. How do you guarantee that certain lines will not become overloaded?

“By integrating optimization algorithms into deep learning models, you have fast ways to distribute power while at the same time ensuring that voltage constraints and demand for electricity are met,” Fung said.

Fung and McKenzie see their next steps as working with domain experts, specifically electrical engineers, on how to best apply their models to the real world, in ways that improve everyone’s day-to-day lives. 

Fung said, “We’ve been developing the tools here, but applying our work to something realistic, like the power grid, is really the ultimate goal.”