The temperature of rivers is something most people only think about if they plan to go swimming, kayaking, or spend a day fishing. Few consider how it could potentially impact their electricity bill.
Power plants use river water for their cooling towers. A seemingly small increase in river water temperature can lead to a measurable decrease in plant efficiency and a significant cost to power plants if the temperature change was unexpected.
Automated sensors provide power-plant operators with essential observations for river temperature monitoring. However, they only provide information about the current state of the water temperature, without any insight into future conditions. The sensors are also vulnerable to disruptions caused by things like debris, lightning, and communication failure that can lead to temporary data gaps or outages.
There is a need for methods that can sustain reliable forecasting using potentially incomplete or unavailable sensor data to help limit downtime at power plants and maintain continuity during periods of sensor disruption.
A collaborative research group within the Tickle College of Engineering, led by Industrial Systems Engineering Professor Anahita Khojandi and Environmental Engineering Professor Jon Hathaway, has been tasked by the Tennessee Valley Authority (TVA) to try and find a solution.
The group developed a novel model that integrates future physics-based insights with historical data to improve forecasting prediction accuracy. RetroSight and ForeSight Ensemble Model (ReForM) is intended to deliver highly accurate river temperature predictions over extended periods of time.
“There are various water temperature forecasting models out there. However, they cannot necessarily forecast the temperature well over extended periods of time. The novelty of our work is in using ‘future’ data from physics-based models and combining them with ‘historical’ sensor data using machine learning models to build much more accurate forecasting models,” Khojandi said. “Physics-based models are very costly to produce, but you are not reproducing any of that. You’re just leveraging them in our machine learning model and giving our models a little bit of foresight into the future.”
Improving Accuracy, Efficiency
Khojandi and Hathaway began working with TVA three years ago for their applied data science class. They didn’t have a funded project, so they pitched ideas to TVA to acquire real-world data for the class.
TVA informed them about the issue it was having with sensors in rivers failing at times, causing last-minute adjustments such as reducing output or purchasing power from external sources, which comes at a significant financial cost.
Driven by concerns about global warming and its detrimental effects on river ecosystems, governments have enacted stricter regulations on river temperature to safeguard aquatic life. The policies require companies like TVA to ensure their operations do not elevate river temperatures beyond safe ecological thresholds.
During periods of high temperatures, cooling towers don’t work as well. This makes it harder for power plants to get rid of extra heat. Sometimes the plant needs to make less electricity or use more power to keep things safe. This means the plant uses more energy and runs less efficiently, which can potentially lead to substantial operational losses.
A more accurate forecasting system like the ReForM can enable power plants operators to plan ahead and arrange for replacement power from other external sources much cheaper than if the need arises last-minute. That can result in millions of dollars in savings during a typical summer.
“If one of these incidents happen, they have to shut down. This is a major burden, not just shutting down, but restarting the system,” Hathaway said. “It’s something they really, really want to avoid.”
Real-World Impact
The TCE sensor project, which was partially funded by an AI Tennessee seed grant, was so successful that it extended into a journal article based on a case study the applied data science class performed on the Buffalo River in Middle Tennessee.
“Students end up taking the class because they like to see how they can use this real-world data to solve a problem,” Hathaway said. “They get so excited about it that, even if it’s not their thesis, they end up publishing the work because the data is so rich and valuable, and the problems are so interesting, and TVA is so engaged.”
The collaborative class project displays how TCE successfully fulfills the university’s land-grant mission with hands-on research and practical teaching that benefits the state while educating the future workforce.
“The fact that research that happens at UT is useful and can result in real-world impact is very rewarding,” Hathaway said. “Sometimes there is a disconnect between research and real-world applicability. But I think it’s a good example of how the work at UT and even just AI in general can be impactful and useful for everyday life.”
Contact
Rhiannon Potkey ([email protected])