Pyranometer with modbus connection

[J.P.M.]

I am listening carefully J.P.M. thank you! I really like that quote also: "All clear skies are different" - there is a lot of nuance to this science I guess. The POA sensor may be more appropriate for what we are trying to do now that you mention the conversion from GHI to POA. This is what we are trying to do and out assumptions: when you look at solar generation over a day for any given site, there are many computable reasons as to why you got so many kilowatt hours out of the system. The time of the year, the time of the day, the sky conditions, the temperature of the air, the temperature of the panels, the wind direction, the shading from a tree 60m away to the southwest, the kinds of clouds, the pitch and orientation of the array, the brand of module, the brand of inverter, the humidity, the barometer readings, the way it is wired, etc.etc. Let's call these the "context" of the energy production. So while some of these can give you very exact reasons (lower/higher conductivity of the module cells, angle of the sun to POA) why you would get a certain number of watt-hours, there are probably many that add a lot of variance to that figure. And because some elements of the context depend on other elements of the context, it's complex. For the average solar PV user (I have a bog-standard 1kW SMA Sunny Boy flushmount multi-crystalline system for example) I don't think they have too many easy ways to judge all these things or get a simple forecast of what tomorrow looks like in terms of watt-hours. Yes there are becoming available. Certainly the NREL labs and NIST and other science foundations are going to have more sophisticated models. So our goal is not to do that, but provide a "point and shoot" version of energy data analysis that uses the newly available neural networking capability (that is all over the news these days) and apply it to this basic challenge. The theory is that by presenting a neural network with a set of inputs that describe the "context" as well as the answer of how many watt-hours per produced, it will naturally weight those inputs in ways to zero in on that answer. Then, you can approximate a "context" for tomorrow let's say and feed that to the trained network- and it should come up with some estimates. Clearly it will need iteration and possibly 12 months of data and maybe years of experience to get it really right, but early testing on this is looking promising. my question to you as an expert on this science is: do you think there is any utility for the average person to get some decent numbers, some reasonable idea of what tomorrow or this week looks like in terms of generation from their specific array? For example - you mention wind direction and module temperature right at the site, not remote by 100m, is that part of the "context" going to be critical in getting closer to our goal? If so we need to put a thermistor on our test panels, and look at integrating a device like the Davis weather stations that we can locate near the array. Also: do you recommend NOT using GHI thermopile in this case but going directly to a POA device, possibly silicon based device, mounted in the plane of the array? Thanks, John

I'll be busy this late A.M./early P.M. measuring array output and efficiency and thus a fouling estimate on this clear day, but briefly, and for a lot of reasons, IMO only, the short answer to your question is no. As for predicting future daily output, that'll be limited by the accuracy of the weather forecast. Anyway, given net metering, to the average homeowner, particularly with net metering, that's usually of little importance or concern, except for nerds like me.

After that, NREL has graced humanity with PVWatts, and it's [pretty versatile if you delve into its capabilities. While not a predictor, it is a useful indicator of possible long term array output.

Because I'm mostly a Luddite, your goals and how you describe them are unclear to me. So, given that what you seem to be after is, IMO only, of questionable use to Joe/Jane 6-pack. and because most serious researchers much more knowledgeable than me are probably beyond what you seem to be looking to do, I'm not sure your goals fit in with a need somewhere.

Respectfully,

[J.P.M.]

I am listening carefully J.P.M. thank you! I really like that quote also: "All clear skies are different" - there is a lot of nuance to this science I guess. The POA sensor may be more appropriate for what we are trying to do now that you mention the conversion from GHI to POA. This is what we are trying to do and out assumptions: when you look at solar generation over a day for any given site, there are many computable reasons as to why you got so many kilowatt hours out of the system. The time of the year, the time of the day, the sky conditions, the temperature of the air, the temperature of the panels, the wind direction, the shading from a tree 60m away to the southwest, the kinds of clouds, the pitch and orientation of the array, the brand of module, the brand of inverter, the humidity, the barometer readings, the way it is wired, etc.etc. Let's call these the "context" of the energy production. So while some of these can give you very exact reasons (lower/higher conductivity of the module cells, angle of the sun to POA) why you would get a certain number of watt-hours, there are probably many that add a lot of variance to that figure. And because some elements of the context depend on other elements of the context, it's complex. For the average solar PV user (I have a bog-standard 1kW SMA Sunny Boy flushmount multi-crystalline system for example) I don't think they have too many easy ways to judge all these things or get a simple forecast of what tomorrow looks like in terms of watt-hours. Yes there are becoming available. Certainly the NREL labs and NIST and other science foundations are going to have more sophisticated models. So our goal is not to do that, but provide a "point and shoot" version of energy data analysis that uses the newly available neural networking capability (that is all over the news these days) and apply it to this basic challenge. The theory is that by presenting a neural network with a set of inputs that describe the "context" as well as the answer of how many watt-hours per produced, it will naturally weight those inputs in ways to zero in on that answer. Then, you can approximate a "context" for tomorrow let's say and feed that to the trained network- and it should come up with some estimates. Clearly it will need iteration and possibly 12 months of data and maybe years of experience to get it really right, but early testing on this is looking promising. my question to you as an expert on this science is: do you think there is any utility for the average person to get some decent numbers, some reasonable idea of what tomorrow or this week looks like in terms of generation from their specific array? For example - you mention wind direction and module temperature right at the site, not remote by 100m, is that part of the "context" going to be critical in getting closer to our goal? If so we need to put a thermistor on our test panels, and look at integrating a device like the Davis weather stations that we can locate near the array. Also: do you recommend NOT using GHI thermopile in this case but going directly to a POA device, possibly silicon based device, mounted in the plane of the array? Thanks, John

Later:

Having reread your post again, several times. I'm of the opinion that your outfit and I are not on the same page. I come at solar energy as one originally trained in undergraduate physics, then, many years later and completely separately, educated and retrained to be a mechanical engineer.

Not a knock, but we seem to do our engineering in different ways. What you call context, I and other most engineers usually call input variables or design parameters.

This stuff is not rocket science. Most of it can be verified in your back yard with a few simple instruments, an understanding of basic physics, and a copy of Duffie and Beckman. A copy of a text such as " An Introduction to Solar Radiation" by Muhammad Iqbal can also be helpful.

With all possible candor, I think you're late to the party. I also think you're biting off more than you can chew with what appears to be your level of technical sophistication with respect to renewable energy at this time.

- From a practical and everyday standpoint, I'm not sure being able to predict solar weather a few days in advance is something most homeowners worry about. At the other end of the scale, the big solar farms already have that capability. In addition, the big farms are usually located in areas where solar output is much easier to predict - it's usually sunny.
- Measuring POA directly may be useful, but most analysis, both in the open literature and the proprietary stuff will use GHI as a reference and convert to POA. There is a lot of software to do a conversion that will include albedo as well. I've written a bunch as have others. Thus, for comparison or other purposes beyond panel or device efficiencies, you'll be working backwards to GHI, which, because of history, is not as easy as the other way around - the correlations are nowhere near as available or accurate.
- All of what you mention and more is the type of stuff that's usually accounted for in the better models. Take a look at something called SAM from NREL. It's PVWatts on steroids and it's pretty hard to beat for flexibility, not to mention price.
- Also look at something called "Solaranywhere" from Richard Perez and the boys in Albany. It sounds like what you may have a mind to be doing with respect to irradiance data is what they are already in the business of doing.
- I've got what I believe is a fairly accurate model of an array's output based on the parameters you mention and a few others. It originally started out as a model for solar thermal flat plate analysis in the late '70's and was originally written in FORTRAN IV, Then, about 10 years ago, it morphed into an EXCEL model for PV. Its output compares quite favorably with SAM and another model called TRNSYS and, most importantly, with actual measurements from my array (which is what I was doing earlier today, and do most days around 20 min. past solar noon on days when the sky is completely clear.) Such models are fairly easy to come up with once you get the basics down.
- The variables you consider hard to judge get a lot more manageable when approached in the disciplined ways of engineering, at least the way I learned them. The basic challenge in that context is to understand the fundamentals and use them to understand a process, and then extrapolate new ways of understanding and then describing what's happening. Again, not rocket science.
- I emphatically claim no more expertize than anyone else, but to respond to your question: Many reasonably reliable ways to predict a solar array's output already exist. I and many others have done such modeling, and the open literature has many such models and methods available. All you need do is look. IMO, it seems you need no more than a literature search which, given what and how you write, seems hasn't been done yet. But specifically, and again IMO only, wind data is probably the single most difficult variable to deal with due to its variability and unpredictability. Someday, instrumentation will be better. Ultrasonc anemometers are a big improvement, but they are styill inadequate to the task.
- As for panel temp. measurements: Contact thermometry is usually done by people who have little understanding of the requirements, limits and error introduced by simply sticking a thermometer or electronic device like a thermistor or thermocouple on a panel. Long story, but the bottom line best and easiest way to measure a panel's temp. is with an infrared thermometer that can accurately correct for surface emittance. Contact methods just screw up the panel temp. and they need insulation on the air side which will only screw up the panel temp. more. Liquid process lines have what are called thermowells for the very reason that simple taping or clamping of temp. measurement devices gives readings somewhere between ambient and process temps as f(ratio air film heat transfer coeff./contact surface thermal conductivity).
- For irradiance measurements, I'd use a true pyranometer operated according to mfg. instructions oriented in the horizontal plane and convert the data to POA as I do every time I need to know what the POA is on my array.

Good luck.

[jwrgorman]

Thanks J.P,M. I hear what you're saying in terms of us approaching the problem differently. That being said, I agree that it's good to learn from solutions that have already be implemented, and I can see that you're advising me to read up on the existing literature and proven algoithms, and have a lot of experience with this stuff.

I don't have a good feeling yet if there are a lot of consumer-grade monitoring services that will tell you in a visual format on your solar monitoring dashboard what your likely production is going to be tomorrow, and chart that prediction over the day on the same interface as your historical one, regardless of the type of inverter(s) you use. It's true what you say that not everyone with a residential sized array would be interested in this information, but I am among the small few of people perhaps who do. That's one reason we're trying to build it, and anyway always good to learn something in the process... I am certainly learning!

Thanks for the advice - interesting about the wind aspect to the calculations, I have been looking into that. Plus the panel temperature - I don't know whether I will be able to afford the kind of instrument that you describe. I think with certain IoT-plus-Machine-Learning schemes, people think they can derive value from a whole bunch of cheap sensors - possibly in some cases that's true, say sleep-quality apps for your phone or mass produced fitbit devices. But garbage-in-garbage-out applies in most cases. For things like air quality or turbidity in water quality, I really think you need the good quality devices to get data that's worth anything. So, I wonder where that tradeoff between sensor price and data quality impacts solar generation forecasting. Thanks, John

[J.P.M.]

Thanks J.P,M. I hear what you're saying in terms of us approaching the problem differently. That being said, I agree that it's good to learn from solutions that have already be implemented, and I can see that you're advising me to read up on the existing literature and proven algoithms, and have a lot of experience with this stuff.

I don't have a good feeling yet if there are a lot of consumer-grade monitoring services that will tell you in a visual format on your solar monitoring dashboard what your likely production is going to be tomorrow, and chart that prediction over the day on the same interface as your historical one, regardless of the type of inverter(s) you use. It's true what you say that not everyone with a residential sized array would be interested in this information, but I am among the small few of people perhaps who do. That's one reason we're trying to build it, and anyway always good to learn something in the process... I am certainly learning!

Thanks for the advice - interesting about the wind aspect to the calculations, I have been looking into that. Plus the panel temperature - I don't know whether I will be able to afford the kind of instrument that you describe. I think with certain IoT-plus-Machine-Learning schemes, people think they can derive value from a whole bunch of cheap sensors - possibly in some cases that's true, say sleep-quality apps for your phone or mass produced fitbit devices. But garbage-in-garbage-out applies in most cases. For things like air quality or turbidity in water quality, I really think you need the good quality devices to get data that's worth anything. So, I wonder where that tradeoff between sensor price and data quality impacts solar generation forecasting. Thanks, John

You're welcome. Overall, do as you wish and good luck, but I honestly and without any rancor believe that if you intend to make a contribution to the renewable energy field you need to expend some effort in finding out more about the basics of the subject and what's already been done as well as more technical information and education about how to do what you want to achieve.

FWIW, and only as one person's experience, I've never had anyone voice, nor have I read of anyone or any group with a need to know what tomorrow's solar potential might be. Climate's what you expect and weather's what you get.

Example: Tomorrow is expected to be sunny and ~ 22 C. for a high temp. I can model tomorrow's expected array output now based on hourly NOAA predicted weather and cloud cover for my zip code if I choose and call that a solar array output forecast. If NOAA published minute granularity data (which they don't because it would be absurd) I could use that too.

Point is, if the weather at my site matches what NOAA predicts, I can successfully predict tomorrow's (or any day's) solar output for my array, probably within a few % or so.

So what ? Anyone with a decent solar model can do the same. If tomorrow's weather mirrors the forecast, I might get a little closer in my prediction than some other models because after 4+ years of watching it like a hawk, I've got my array dialed in tighter than ma Barker's poop chute. But, the point is, I (or anyone with access to existing software, or stuff they can write) can already and quite easily do what you seem to be looking for. As I wrote, you're late to the party.

Additionally, most solar energy analysis is concerned more about long term output, say on a monthly or annual time frame. Predicting output a few days or a week into the future may have some merit, but because accuracy of any model's output is limited by the accuracy of a weather forecast, and short term predictions of output being notoriously inaccurate due to the stochastic nature of weather, most models are limited to relying on what are actually better described as long(er) term climate trends rather than daily forecasts.

See the TMY manual for a decent description of the TMY logic. There are other weather data bases available. For example, I've constructed my own weather database using 30 yr. hourly averages of weather data from NOAA and a clear sky irradiance model that's pretty close to something called the HDKR model (See D.& B.). With that, and my Davis data, and array output data in 5 min. increments I've compared the last 4 yrs. of array output and found that it pretty much matches what NREL's SAM gives me once adjusted for weather data bases. Long, boring story.

All that's nothing special or little more than tinkering by someone with more time on their hands than brains. Others have done similar and done a better job.

IMO, and respectfully, you can do the same, but it seems a waste of needless reconfirmation. Suit yourself, but I'd suggest you spend some time looking and I bet you'll agree. What you'll find and learn is what's already been done, and maybe get some of the background necessary to understand why what you seem to be looking for is mostly redundant and probably ill conceived at that before the attempt. You'll learn more and maybe add to the body of knowledge instead of simply copying it for purposes that few if any will find useful.

Take what you want of the above. Scrap the rest.

Respectfully,

[jwrgorman]

Alright - well I do think that the need to know tomorrow's solar potential for a small residential system is not just my interest alone, there are some strong economic reasons why that information is going to be very important as PV adoption expands. But it sounds like you and many others already know how to do that. Let me do some reading and get back to you in a few weeks. Thanks, John

[J.P.M.]

Alright - well I do think that the need to know tomorrow's solar potential for a small residential system is not just my interest alone, there are some strong economic reasons why that information is going to be very important as PV adoption expands. But it sounds like you and many others already know how to do that. Let me do some reading and get back to you in a few weeks. Thanks, John

I don't do people's thinking or exploring for them, but I bet the more you find out, the more you'll come to understand that there is more information and help available than you know or that most residential users will probably ever take the time and make the effort of explore much less understand.