It goes without saying that solar investors want a good cost and performance analysis before deciding whether to pump money into a project. What many may not realize is the numbers they get often are superficial and too basic, said researchers from Argonne National Laboratory.
In a paper published in Energy & Environmental Science last month, the authors crunched numbers to create a cost analysis that showed in more detail a range of scenarios for power production and costs throughout the expected lifetime of a solar power plant. The analysis also tells the likelihood that each scenario will take place.
“People are taking these guesses and not capturing the uncertainties associated with them,” said Seth Darling, an Argonne researcher and lead author of the paper, which looks at levelized cost of energy calculations for utility-scale projects. “They take their best guesses based on existing data and make projections about what those numbers will be.”
Levelized cost of energy (LCOE), expressed in cents per kilowatt hour (kWh), takes into account not only the capital cost of building a project, but also all the operating and maintenance expenses over time (such as the length of a power purchase agreement). It doesn’t include the profit a plant owner wants to make.
Banks or other project financiers typically hire consulting firms to generate LCOE analyses to help them make investment decisions. Some large developers also have internal teams doing the same. The LCOE numbers aren’t just valuable for developers and bankers, they also are useful for policy makers, particularly given the solar energy industry’s reliance on government incentives.
Energy Secretary Steve Chu recently launched the SunShot initiative to put money into projects that will help drive down the LCOE to $0.06 per kilowatt hour by the end of the decade, a rate that will be cost competitive with power from fossil fuel sources without a need for government subsidies.
The paper, “Assumptions and Levelized Cost of Energy for Photovoltaics,” pointed to Spain as an example of how simplistic LCOE calculations can contribute the collapse of a booming market. The country became the largest photovoltaic installation market in 2008 when developers anticipated a sharp decline in government-set solar electric pricing and a national cap of 500 megawatts for 2009.
“For the PV industry, LCOE analysis failed most spectacularly in Spain in 2008, when too many projects were developed using best case assumptions regarding panel failure rates and other performance factors. A more thorough analysis of the uncertainties associated with these assumptions could have prevented substantial losses,” according to the paper, which was co-authored by a researcher from Northwestern University and an analyst from Gartner.
What Darling and his fellow researchers have found is that analysts typically plug in just one number for each data field, the result of which leads to a simplistic view of a power plant’s power output. A LCOE analysis takes into account factors such as the amount of sunlight, sunlight-to-electricity conversion rate of the solar panels, anticipated degradation rate of the photovoltaic materials, and the cost of borrowing money.
Every firm may have its own model and software for doing LCOE calculations. The National Renewable Energy Laboratory offers what it calls the Solar Advisor Model for doing such analysis. The model does allow a deeper analysis by letting you vary one number for each data field to get a range of results, Darling said. But there are better ways to project cost and performance, he added.
“You have to give a range when you give LCOE estimates. When you talk to people in the industry, they will say, ‘This is the number,’ especially finance guys. We want to say: don’t do that,” said Alfonso Velosa, a Gartner analyst and co-author of the paper.
In the paper, the researchers used a well-known method called Monte Carlo to show what a more detailed LCOE analysis will look at for each of the three hypothetical projects in different locations. Instead of, say, putting in the average value for the solar resource at one location into the model, the researchers first come up with a range of possible numbers for a given location by culling decades of data. Then, Monte Carlo calculations randomly select numbers from different sets of numbers. By doing these Monte Carlo calculations many times over, they came up with not just a range of LCOE numbers but also information showing the probability of achieving certain results. Each analysis also could shed light on which factors lead to more uncertainties in the outcome.
The work requires more assumptions and calculations, which paint a more complicated picture of a project’s potential cost and performance.
“It makes things less certain, but it was always less certain, you just didn’t know how uncertain it was before,” Darling said. “I hope it gives people more input and helps them think about ways to reduce the uncertainties.”
Using a good mathematical model isn’t the only key to producing a more comprehensive analysis. Good data also are important. Unfortunately, there isn’t a whole lot of operational data from commercial solar power plants because the market has only begun to take off in the United States. Operational data tend to be proprietary anyway. Data on weather, solar irradiance, solar panel efficiencies are more readily available, but the quality of data can be inconsistent, depending on who is collecting and providing them.
Larger developers and owners such as SunPower, SunEdison, First Solar, Sempra and Florida Power & Light can carry out more sophisticated analyses by drawing data from the large-scale projects they have installed. Some universities and national labs, such as NREL and Sandia, are collecting data from their own field trials. Chevron announced a test bed of photovoltaic technologies in California last year. ProLogis, which leases industrial spaces worldwide, including rooftops for solar power projects, also launched a test site in Colorado last year.
Argonne also is planning a test site at the Illinois Tollway’s headquarters outside of Chicago, Darling said. The idea is to collect data about how solar panels operate in the Midwest and make that data available. The site will test panels using crystalline (mono- and multicrystalline), amorphous-silicon, cadmium-telluride and copper-indium-gallium-selenide. Weather stations will be set up to collect data on temperatures, humidity, solar irradiance and others. The installation will use microinverters, said Darling. Selections for the solar panels and microinverters haven’t been finalized.