Depending on who you are, there are many contenders for the most important building on a university campus—the football stadium, the chancellor’s office or the coffee shop. But the cornerstone of any university is actually the central utilities plant, the industrial heart that distributes electricity, steam, hot water, and chilled water needed to run air conditioning systems and research facilities at multiple buildings on campus.
Victor Zavala
Managing utility plants, however, is enough to boggle the sharpest minds. Plant managers must predict future energy needs from all the buildings and decide how much water, natural gas and electricity to buy and exactly when to turn the boilers up or turn down the cooling towers. Even tougher to manage are unforeseen fluctuations and events, like a spike in electricity prices or an unexpected cold snap that has everyone cranking the heat up on a spring day.
Most current control systems don’t take these types of uncertainties into account, leading to economic losses as well as safety and utility supply issues. That’s why Ranjeet Kumar, a spring 2020 PhD graduate in the Department of Chemical and Biological Engineering at the University of Wisconsin-Madison, and Associate Professor Victor Zavala designed a new stochastic model predictive control system for central utility plants, outlined in the June 2020 issue of the Journal of Process Control.
Central plants are used by large colleges and universities, military bases, medical centers and large corporate campuses. Though they are complex and costly to operate, they create efficiencies that over time save money. Currently, most complex decisions on when to purchase energy are made by logic-based control systems and people doing their best to forecast future energy needs and energy prices. However, they don’t often take into account potential disruptions or extreme events.
To understand how to optimize the control system, Kumar modeled the utilities plant of a typical university campus derived from data collected by his industrial collaborator, Milwaukee-based Johnson Controls, which develops control systems for large utility plants. He then ran simulations to find optimal strategies to maximize efficiency and minimize cost while mitigating energy demands and market uncertainties.
“Managers have to make a plan for each hour of every day for how much load a boiler or cooling tower will have,” Kumar says. “Our goal was to design a control architecture that deals with multiple timescales and uncertainties that energy systems encounter.”
Ranjeet Kumar
Currently, the most sophisticated control architectures use what is called a deterministic model to decide when and how much energy to buy. Typically, these systems look at average prices and past energy use to make their decisions and ignore uncertainty. But Kumar found that these models failed to properly capture unforeseen disruptions, resulting in economic losses for drawing too much power or an inability to keep up with demand for hot water or heating.
Kumar’s model, however, is stochastic, meaning that it is designed to explicitly anticipate uncertainties by capturing multiple possible scenarios when making decisions.
“It was not obvious that there would be many benefits. But the electricity market is subject to a lot of uncertainty. The prices are fluctuating all the time and have limited predictability,” says Zavala. “And there’s uncertainty on the side of the buildings because the use of electricity and hot water is uncertain. So, plant managers are in between a rock and a hard place.”
According to the results, there could be a big benefit. By incorporating uncertainty into the control system, the stochastic model was able to save about 7.5% over the deterministic model, which in the case of a large university utility system translates into hundreds of thousands of dollars.
The new control model isn’t just for universities. Large manufacturers and chemical plants that must also balance energy costs and demand could also benefit from the stochastic model. In fact, Kumar is moving on to a position in the chemical industry, which could benefit from his ideas.
Zavala says this research is a great of example of industrial cooperation. “One of the things that I like about this particular project is that it shows how industry and academia can benefit from cooperating,” he says. “It’s nice to highlight the synergy; our collaborator proposed a real-world problem that current control technology cannot solve and it gave us an interesting problem to research. I think it would be nice to keep pursuing those types of collaborations.”
Victor Zavala is the Baldovin-DaPra Associate Professor in the Department of Chemical and Biological Engineering at UW-Madison.
The research was supported by funding from National Science Foundation award NSF-EECS-1609183.