So the moment has finally arrived – my first PhD manuscript has been accepted for publication in Solar Energy Journal!
This is a great accomplishment for me as a young scientist and I couldn’t be more pleased.
The paper forms a core component of my PhD thesis/dissertation – the use of PV data as the primary input to solar forecasting routines. The method is important, because it confronts a key challenge in the Australian electricity market: namely, that we have no real idea how much electricity the 1.4 million rooftop photovoltaic energy systems are producing at any given time.
Most small-scale PV systems are not actively monitored by energy distributors in Australia.
How can this be? Simply because most of them are not actively monitored by energy distributors in Australia. Entities such as ActewAGL here in Canberra know how much energy your solar system has produced each time they read the meter – but that is only done every three months! The day-to-day, hour-by-hour, minute level variations in solar power output are unknown. And although that isn’t causing any serious problems yet – that is soon likely to change.
Canberra is a great example. With 14,000+ PV systems installed and some estimates from ActewAGL insiders estimating the total capacity of rooftop solar at nearly 70MW – solar energy is getting very close to being too large of a contributor for distributors and energy markets to ignore!
So what’s the solution?
We need a low cost, easily implementable solution for estimating the total power production of these 14,000+ generation sites. After we do that, then we need to produce solar forecasts to predict their future contributions. And my idea is to use the solar panels themselves as the primary mechanism for doing so.
The concept of my paper is simple – the best solution for this problem is to use a subset of the existing PV systems to generate estimates of the power production at the rest.
The main challenge here is that each PV system is very different. It points its own direction, has its own tilt, is made of up of different number and type of modules and inverters and experiences its own levels of shading and soiling.
Fortunately, I’ve devised a way to handle most of these issues.
KPV: The Clear-Sky Index for Photovoltaics
The concept is called KPV: The Clear-Sky Index for Photovoltaics. It operates by removing these system level characteristics from a PV systems power output time series – effectively normalizing the data. So instead of a having a 3kW system facing NW at a tilt of 30 degrees reporting its power output in Watts, it now reports a value between 0 and 1. 1 = clear sky, 0 = completely opaque skies (no sunlight).
How is this done? By dividing the measured power output by a clear-sky simulation of the system in question.
We can generate this clear-sky performance estimate through two steps:
(1) First, we have to prepare an estimate of the clear sky radiation arriving at the plane-of-array (POA). This is generated via a clear-sky radiation estimate from the Esra clear-sky model (which by the way is the best simple clear sky model for use in Australia – more on that soon). This is then transposed to a tilted surface by the Reindl transposition formulae. Now we have the radiation arriving at the solar array.
(2) Second, we create a ‘virtual array’ using the Sandia Performance Model and Inverter Performance Model routines, which are the core components of the System Advisor Model (SAM). This virtual array is created to match the real-world array as closely as possible. Once this array is designed, we input the POA irradiance and VOILA – a clear-sky estimate of the PV system’s power output.
Divide the measured power output (PVMEAS) by the clear-sky power output (PVCLR) and you have the Clear-Sky index for Photovoltaics: KPV:
This an extremely useful, very simple calculation that removes the diurnal and seasonal cycles and the system level nuances (namely its orientation, power rating and tilt). You can see it work in this image:
And here is the coolest part...
Using the KPV value at one system (KPV_1 = PVMEAS_1/PVCLR_1), we can make a quick, computationally cheap estimate of a nearby PV system’s power output (PVMEAS_2). The only requirement is that we can draw the theoretical clear-sky curve for that system (PVCLR_2). With the clear sky power output method I’ve described above, we can do that! As long as we know the basic properties of the second PV system (it’s power rating, orientation/tilt, etc).
We can make that estimate via:
And it works quite well! The only real challenge remaining is the clouds – which is always the challenge with solar forecasting/estimations. The method works very well under uniform cloud conditions and under clear skies – but mixed cloud conditions, well we need to come up with some solutions for that (the focus of upcoming work of mine).
The key here is that it works particularly well with large, collective ramp events. And that is great, because those are the events we’re most concerned with. Individual systems or small groups or PV systems ramping up and down aren’t (usually) problem. But cases where an entire city drops out or ramps up, well those could be! We’ll discuss that more in the future, but for now I’ll leave you with a an image of this method being used during such an event:
Welcome to the world, Mr. KPV calculation. I have a hunch you are going to be quite useful!
You can [download the pre-print version here].
It will likely have some further changes to it before the final version, which will appear in the journal in the coming months.
Also coming soon, an R package that allows anyone to use the KPV tool!