Almost daily, scientists release another study documenting the growing influence of human-caused global warming. Driven by ever-increasing amounts of greenhouse gases such as carbon dioxide in the atmosphere, mostly from coal, oil and natural gas we burn for energy, the Earth’s average surface temperature has risen by almost 1ºC since the late 19thcentury, with much greater increases predicted ahead.
With this seemingly inexorable rise in the concentration of atmospheric carbon dioxide, researchers have been exploring ways to meet our growing energy needs with clean renewable energy sources. In fact, nearly two-thirds of newly installed global power generation in 2018 came from renewable sources, with distributed installations of solar photovoltaic panels contributing to this energy transition.
L to R: Recent PhD graduate Fiodar Kazhamiaka and his co-supervisor Professor Catherine Rosenberg. Professor Srinivasan Keshav, Fiodar’s other co-supervisor, was unavailable for this photo. Dr. Keshav’s photo appears below.
Dr. Kazhamiaka is currently a postdoctoral researcher at Stanford University, a position he started in Janaury 2020.
“Photovoltaic or PV panels produce power that can be used in homes, buildings, machines — anything electrical,” explains Fiodar Kazhamiaka, a recent PhD graduate from the Cheriton School of Computer Science’s Information Systems and Science for Energy lab. “PV panels can be deployed in solar farms — large fields of panels producing enough electricity to perhaps power a town — or in smaller, distributed installations at the scale of rooftops. A home with PV panels installed on the roof can use the energy produced by the panels to offset the electricity bought from the grid.”
PV panels have potential to significantly lower the carbon footprint of electricity production, and in many places the cost of solar power has dropped below the price of fossil fuels such as coal. But despite growing demand for green energy, the cost of the energy storage systems that often need to be used alongside PV panels has limited their widespread adoption.
The main challenge in using this source of clean energy is its variability. A PV panel produces electricity only when the sun is shining. Even a cloudy day can significantly reduce the amount of electricity they produce, so PV panels are often used with banks of lithium-ion batteries that are charged and discharged to help match the electricity supply from the PV panels with a household’s demand. For a typical North American home, tens of thousands of dollars would have to be spent on PV panels, battery modules, and peripheral hardware to significantly reduce the need to buy electricity from the grid.
The high capital cost of a PV panel and storage system is prohibitive for many people, but the expense could be offset by the value they provide to homeowners, especially if their size and operation are optimized.
“Some utility companies let you sell electricity, whatever amount your panels are generating you’re not using,” Fiodar explained. “A decade ago these feed-in rates were quite high, but they’ve dropped enough in Ontario that selling electricity back to the grid isn’t as attractive as it once was. As rates drop, storage can make more sense — storing excess electricity in batteries for you to use later. But whatever makes the most sense — selling electricity or storing it for later use here — we want to design a system that provides the greatest value to users.”
To this end, Fiodar’s doctoral research focused on practical methods to increase the economic value of residential PV-storage systems. “We treat the system as an investment vehicle. We look at the return on investment of the PV-storage system to see if there’s something we can do on the design side — carefully choosing how large a battery a user should buy and how many PV panels to install — and on the system operation side — how to run the system — so it generates the most money for the user.”
Every deployment of a PV-storage system is different, Fiodar said. “It’ll depend on which grid-pricing policy the utility company uses. It’ll depend on how much electricity you use — what your load profile looks like over a typical day — and it’ll depend on where you install the panels and how much sunlight they capture. If you’re trying to maximize the return on investment over so many variables, there isn’t going to be a single solution applicable to every operating environment.”
Among the biggest challenges is that the data needed to optimally size and operate a PV-storage system typically isn’t available when the system is installed. If a user knew how much energy the panels will produce over time and what the typical household load patterns looked like it would be much easier to optimize the system’s design and operation, but typically this data is not available. And even if the data underlying current conditions were available, usage patterns can change dramatically over time.
“Say I buy an electric vehicle several years after I install a PV-storage system in my house. If I’m now charging my electric car at home my residential energy use will increase dramatically. We simply don’t know how our energy use will change over the lifetime of the batteries and PV panels, which can be decades long, but we do know that energy use will likely change over their lifetime.”
On the design side, Fiodar worked with his supervisors and Professor Yashar Ghiassi-Farrokhfal at Erasmus University in the Netherlands to develop algorithms for sizing PV-storage systems. The algorithms make use of limited data to compute the most appropriate system size — the one with the highest value — by using the available data to characterize a statistical distribution of what the future operating environment will look like.*
For optimizing system operation, the key is identifying the parameters that the flexibility of the battery can take advantage of to reduce costs and increase revenue. Time-of-use electricity pricing, as we have in Ontario, is a good example of a parameter that could be optimized for operation by a PV-storage system.
Waterloo North Hydro, the utility that supplies electricity to residents and businesses in Waterloo, Woolwich and Wellesley, charges 20.8 cents per kilowatt-hour (kWh) during peak usage times, 14.4 cents per kWh during mid-peak times, and 10.1 cents per kWh during off-peak times during the winter. The utility charges different rates and has different on-peak, mid-peak and off-peak periods during the summer. With rates varying depending on the time of day and the time of year, it’s not always a straightforward exercise for a user to determine when it’s best to use electricity the PV panels are currently generating, when it’s best to store the energy in batteries to use later, and when it’s best to do some combination of the two.
To help solve this problem, Fiodar developed several control solutions for PV-storage systems, the most sophisticated of which can learn on its own and adapt to changes in the operating environment.
“We developed three control methods that don’t necessarily require data when the system is deployed because they’re going to learn over time. As soon as the system is installed, it begins to collect data, and in one of our control methods this data is used encode a control policy in a neural network. The neural network is periodically retrained as more and more fresh data is collected by the system, allowing it to adapt to changes in the operating environment, so it’s always operating close to optimal performance.”
Even when optimized, a PV-storage system is arguably still a substantial expense that might deter many potential buyers, but if a strong financial argument can be made even people who don’t consider themselves green may buy one.
“For widespread adoption, a PV-storage system has to make sense economically. Everyone understands money and that’s the Holy Grail of sustainable energy production — finding a way to provide clean energy that competes with or is cheaper than what’s available now.”
* This research, in an article titled “Comparison of Different Approaches for Solar PV and Storage Sizing,” will be published in IEEE Transactions on Sustainable Computing, in a special issue on energy systems.
