Plant Optimization

Coal plant’s AI drives down emissions, boosts efficiency

Global management and consulting firm McKinsey & Company published a case study detailing how it helped Vistra develop a machine-learning model to improve the efficiency and emissions of the Martin Lake Power Plant in Rusk County, Texas.
Coal plant’s AI drives down emissions, boosts efficiency
Martin Lake Power Plant (By Larry D. Moore, CC BY 4.0, https://commons.wikimedia.org/w/index.php?curid=80251398)

There’s plenty of hype surrounding AI— no matter the industry. But clear applications are emerging from the clutter, and power generators are getting a taste of the technology’s potential.

One of the largest generators in the U.S., Vistra, tapped McKinsey & Company to develop a machine-learning model to improve the efficiency and emissions of the coal-fired Martin Lake Power Plant in Rusk County, Texas.

The effort began when Vistra wanted to build and deploy a heat-rate optimizer (HRO) for the plant. The company worked with McKinsey data scientists and machine learning engineers from QuantumBlack AI to build a “multilayered neural-network model,” or an AI-powered algorithm that learns about the effects of complex nonlinear relationships.

The team fed the model two years of plant data to see which combination of external factors and internal decisions could produce the optimal HRO for any given time. External factors included temperature and humidity, and internal decisions included variables that operators can control.

It wasn’t a “one-and-done” solution, though. Vistra’s team continued to provide guidance on how the plant worked and identified data sources from sensors, which McKinsey said helped its engineers refine the model by adding and removing variables to see how the heat rate changed.

Through the training process and “introducing better data,” the models eventually made predictions with 99% accuracy or higher. After running the model through a series of real-world tests, the engineers turned the model into an “AI-powered engine.” After implementing the engine, the plant’s operators received recommendations every 30 minutes on how to improve the plant’s heat-rate efficiency.

“There are things that took me 20 years to learn about these power plants,” said Lloyd Hughes, Vistra’s operations manager. “This model learned them in an afternoon.”

With higher efficiency came more carbon reduction. Martin Lake was running more than 2% more efficiently after three months of operating with the machine-learning tool, which McKinsey said resulted in savings of $4.5 million per year and 340,000 tons of abated carbon.

Following the success at the Martin Lake Power Plant, Vistra distributed the AI-enabled HRO to another 67 generation units across 26 plants, which resulted in an average of 1% improvement in efficiency, McKinsey said, in addition to more than $23 million in savings.

Overall, Vistra’s AI initiatives have helped the company avoid around 1.6 million tons of carbon per year, McKinsey said.

Read the full case study here.