PVsyst's forum

Welcome on PVsyst's forum.
It is currently Tue Aug 04, 2020 1:28 am

All times are UTC+02:00




Post new topic  Reply to topic  [ 15 posts ]  Go to page 1 2 Next
Author Message
PostPosted: Tue Jul 30, 2013 4:44 pm 

Joined: Thu Feb 28, 2013 12:48 am
Posts: 16
Location: Charlottesville, VA, USA
I understand that in the 6.10 revision new shading losses for diffuse irradiation have been added for tracking systems. According to the FAQ viewtopic.php?f=30&t=694post on this topic, it seams that this is the first time in any PVSyst version that tracking systems have included a system loss for the diffuse component. Since I have upgraded to the 6.10 version, I have the opportunity to compare the results from simulations run prior to 6.10 to those run with 6.10 and I have some questions and concerns.

I have found that in two example single axis horizontal tracking systems with backtracking, one in north-central US and one in southwest US, the "Near shadings: irradiance loss" that is now included in the loss diagram is -2.4% for the southwest site and -2.7% for the north-central site (all system details and losses were unchanged from 6.08 to 6.10). This is much higher than I expected from the explanation provided in the FAQ post. Both sites coincidentally have a 31% diffuse ratio (diffuse/GHI).

Why is this loss so high? For another example system I ran in 6.10 for a 25 degree fixed tilt system in the northeast US I only get a "Near shadings: irradiance loss" of -2.4% for a system with a higher GCR (46% vs. 33%) and a higher diffuse/GHI ratio (45%).

I am concerned that this new loss is a very significant change to prior versions of PVSyst and thus should be considered carefully. As someone who reviews many PVSyst reports from dozens of users on a regular basis I can say that this change presents a very significant challenge to the multitude of users of the PVSyst tool. There are hundreds of MW of tracking systems out there that were modeled in PVSyst prior to version 6.10 and hundreds more will continue to be modeled by users who have not upgraded. This presents a real challenge to people who analyze PVSyst reports and system performance issues. Is the suggestion that the industry no longer consider any tracking system production modeled prior to version 6.10 because they are erroneous?

Applying an additional >2% loss to our systems is not something that is good for the industry. We need to have full confidence in this change. We need to better understand how to apply this issue to systems modeled prior to 6.10. Millions of dollars are at stake with changes like this. How much confidence do we have in the new modeling method? Are there studies to support this change? Are there papers we can reference? Have third party firms reviewed this change?

PVSyst is an excellent tool and one I value very highly. I really appreciate all the work that goes into the creation and maintenance of this powerful tool. With the success of this tool comes a significant responsibility however so I would ask that major changes like this be supported with more transparency and support.

Thank you.

Jason Fisher
Charlottesville, Virginia, USA


Top
   
PostPosted: Wed Jul 31, 2013 10:43 am 
User avatar

Joined: Mon Apr 16, 2012 7:29 pm
Posts: 1658
I understand your bewilderment.

With the PVsyst simulations, our objective is always to approach the reality at best. But this involves models, and models are never the reality.

Now PVsyst is considered by some people as a reference, and many assume that "this is the truth". This situation is quite independent on our willingness.
We never pretended to become such a reference, and we cannot assume any responsibility about the full reliability of the results in any case.
In particular, this should not prevent us to do modifications in the simulation, if we judge it necessary, even if it is not in accordance with previous simulations.
A software which doesn't evolve is dead.

In PVsyst, the estimation of the shading effects on the diffuse is a model. It is based on the hypothesis that the diffuse is isotropic, and performs an integral of the irradiance with shading factor over all directions of the sky vault "seen" by the collectors. A similar hypothesis and calculation is applied to the albedo contribution, involved in the transposition models.
This model is from my own, implemented in Pvsyst since the beginning (1993). I proposed a research project for assessing this hypothesis by an experimental work, but I couldn't obtain financial support: this question didn't have any interest for anybody!
I don't know any publication studying the effect of shadings on the diffuse part. I don't know how this question is treated in other software. However this is a crucial question as it usually represents half of the shadings loss (or more) in most cases. If someone has some references, please communicate them to me.

The calculation of the tracking systems was introduced in PVsyst more than 10 years ago. The calculation of the shading on diffuse was treated in the same way as for the fixed planes, i.e. by integrating over the shading factor table. This was indeed an erroneous assumption, but I was not aware of that until about one year ago. Correcting this represented a not megligible work, that I just performed for the version 6.1. I could'nt guess that this would imply such a high loss before doing the full calculation (see our FAQ How is calculated the shading loss factor in tracking systems).

But this is quite plausible if you have a look on my publication "Optimization of row-arrangement in PV Systems, shading loss evaluations according to module positioning and connexions". This shows that in row arrangement the shading factor on diffuse is proportionnal to the plane tilt; then in tracking systems we have high tracker's tilts in the morning and the evening. Moreover the limit angle is usually rather high (increases with the tilt when the pitch is constant). The situation will still be worse with lower GCR, as when the distance between rows increases, the maximum tilt allowed by the backtracking will also increase.
Attachment:
File comment: Shading factor on diffuse in row arrangement, as computed with Pvsyst hypothesis.
It is a geometrical characteristics, independent on the meteo data and site.

Shading_Factor_On_Diffuse.png
Shading_Factor_On_Diffuse.png [ 17.81 KiB | Viewed 21213 times ]
The loss is also due to the albedo contribution: the transposition model yields an albedo contribution proportionnal to (1 - cosi)/2 (i = tilt angle), with about 1% loss at 30° tilt, increasing more than linearly. This contribution is completely lost in big arrays: only the first row "sees" the albedo contribution.

NB: This calculation error appears essentially in the calculation of backtracking systems. The old calculation with "normal" tracking is much less affected: according to my first observations (to be confirmed) the result with the new calculation is rather similar to the old one: the shading factors (global of for each position) compensate each other.

NB: This work about shadings with tracking systems was undertaken after the publication "Increased Energy Production of First Solar Horizontal Single-axis Tracking PV Systems without Backtracking", Lauren Ngan and al, First Solar, San Francisco 94105.

Now the economical impact of such uncertainties in the simulations essentially acts on the previsions of the financial balance, not in the real yield. PVsyst has never created nor lost any kWh by itself !
This can have some implications on contracts, or technological choices like choosing Backtracking. However the backtracking choice may easily be reverted to "normal operation" without significant costs, if this appears to be better.


Top
   
PostPosted: Wed Jul 31, 2013 8:35 pm 

Joined: Thu Aug 02, 2012 3:38 am
Posts: 105
I agree with André, PVsyst can't be kept static because changes might make the results different. If someone wants to use PVSyst as a standard that is up to them but that does not mean that now some kind of vetting process needs to happen before any changes are made to determine if the changes are for the better. The software has to evolve because our understanding of how to properly simulate PV systems will continue to evolve. Does that mean that simulations from previous version are now wrong? That depends on the changes that were made and what you consider "wrong". If they find a better way to simulate something does that make the previous way wrong? No, it just means that the previous way was not as good. If they find a bug that gives wrong answers and fix it then yes, the previous way gives the wrong answer.

The final determination is always going to be how well any of the software simulators allow someone to estimate the future output of a PV system and change is going to happen because this is an evolving field. In retrospect some changes might make the software better, some might make it worse, but it is up to the makers of the software to make the changes they think are necessary.


Top
   
PostPosted: Thu Aug 01, 2013 1:40 am 

Joined: Thu Feb 28, 2013 12:48 am
Posts: 16
Location: Charlottesville, VA, USA
Thanks Andre as usual for your well reasoned and thoughtful explanation of the good work you are doing with PVSyst. The intention of my original post is not to demand answers but rather to spark debate so I appreciate you adding to the discussion. My comments around the financial impact of these changes is (like it or not) based in the reality of our market right now. I completely understand your position that this is a modeling tool and not a tool to predict the future! The reality though is that the PV market only grows with the adoption of the technology by people who are not as familiar with these systems as you (who eclipses all of us with your knowledge), or I, or my good friend Marvin. Developers, investors, builders, owners, and maintenance firms all look to performance modeling tools for answers to the tricky question of how these systems are expected to perform. Those of us in the position of explaining these tools must be able to reduce a great deal of complexity into clear and defensible explanations of the inputs and outputs of these models. We need to have a high degree of faith in these tools and for me, that faith comes from understanding them as best I can.

That is what was on my mind when starting this thread, how do I explain this change? How do we discuss previous simulations? How accurate are these assumptions when compared to what we see in the field?
The same debate happens everyday with other simulation details whether they be on voltage drop, soiling, etc.
I realize that you cannot tell us what values are "true" and what are "false" but in this case, this change to the tool has produced a very significant change in the model's results which for me merits extra consideration and explanation.

Despite your very comprehensive and clear explanation I still have concerns over how the new algorithm is being applied. I would not attempt to challenge your equations but I would like to be able to understand the results better, and thus explain them better. Today I modeled a new horizontal tracking system in the northeast US. Using PVSyst v6.1 the simulation resulted in a -3.7% Near Shadings loss (0% loss reported in pre 6.1 versions). I then modeled a fixed tilt array of the same size and equipment, using the same Meteo resource and system losses. The 0 degree single axis tracker was 0.33 GCR and the fixed tilt array was 25 degree tilt and 0.43 GCR. Both were modeled in the 3D scene. The Tracker used backtracking and the fixed tilt modeled electrical losses according to strings. The result for the fixed tilt was -2.9% Near Shading losses. Lower losses than the backtracking single axis 0 degree tracking system.

I am still confused by this. It simply does not make intuitive sense to me. A backtracking horizontal tracking system spends a significant portion of the most productive part of the day and year at very low tilts, which based on the little I know about this subject would suggest that any diffuse losses would be extremely small during this time, just like when we model low tilt fixed tilt systems (i.e. 10-15 degrees). Furthermore when each tracking column is tilted (thus blocking some diffuse from it's neighbor) it is in the lower irradiance periods of the day when the total irradiation of both diffuse and beam is of lower total significance to the daily system production. In contrast a fixed tilt system has a geometry that never changes so diffuse irradiation blockage from neighboring sheds is an effect which occurs throughout the day. It just doesn't add up to me.

Another way to look at this question is by looking at the Global incident in the collector plane differences between these two systems. My GHI for this site was 1371 kWh/m2. The horizontal tracking system with backtracking added 25.1% to this value. The 25 degree fixed tilt (0 azimuth) added 15.0%. This is a difference of 10.1% additional irradiation of the tracker, marginal but understandable in this region. When I deduct the near shading losses of -2.9% for the fixed tilt and -3.7% for the tracker I now only get a incremental improvement of 9.3%, with all system losses I only see +9% energy harvest for horizontal tracking at this site. Incidentally at a second site I modeled in the desert southwest doing the same comparison using v6.10, I got similar results, near shading losses for the horizontal tracker were about 1% higher then they were for a fixed tilt system and the resulting net energy gain was about 19%. This is a lower gain than is typically seen in the field. Thus my concern. Is this tool becoming overly conservative?

I am not suggesting that PVSyst should remain static, I really appreciate the continuous improvement you have shown in the development of this tool. It is a great tool not just to model system production, but to explore various designs and even to illustrate complex PV theory. If my thread does spur some discussion and investigation into this effect, I might suggest that when planning major changes like this in the future, you consider launching them as a beta version for a period of time prior to full "upgrade". This would alert users to consider upgrading immediately more carefully. I know you give us the option of parallel installation and at times I have done that, that method does get messy though for file management. I admit I have formed a habit to routinely upgrade my PVSyst software since your fixes and improvements are much appreciated. This change caught me off guard and so I am now trying to develop faith that the tool is predicting the "most likely" performance of a tracking PV system, I will avoid using the term "most accurate"! ;)

Thanks for the discussion. I hope we can continue it.

Jason Fisher, Charlottesville, VA, USA


Top
   
PostPosted: Thu Aug 01, 2013 11:24 pm 

Joined: Thu Aug 02, 2012 3:38 am
Posts: 105
I can understand how you feel Jason. We don't like unknowns and trying to figure out how much a PV system is worth over the next 25 years is a big unknown with a lot of money riding on it. Personally I think a lot of people oversell these energy predictions by giving into people's desire to have something that can predict the future. If I'm selling the results with an error bar of zero then a 3% change from one version to another is a big deal, if my error bar is 10% then a 3% change is factored into the expected variation. Do people like that 10% error bar? No they don't, but that does not mean I can make it smaller just to make them happy, it is what it is. If we are not using error bars when we are presenting the results of these applications then we are setting ourselves up to fail.

As for having an intimate understanding of how these applications operate, it's immensely complex. I have just enough understanding of it to know how much is going on in the backend. I would not expect someone not in the field to be able to understand it anymore than I could expect a non-engineer to understand complex electrical engineering subject matter. So to be able to provide a complete explanation of what changed in the algorithm between version 6.09 and version 6.1 would, to start, require an in depth understanding of how the whole process of converting irradiation data into energy available to a module takes place.

I think it's asking a lot to expect the PVSyst team to basically teach all of us users how to model PV system performance in detail enough to justify to us how and why the internal models change. I have no problem depending on people who are leaders in their field of science evolving our understanding of how this all works. I have to depend on them just like people have to depend on me for electrical engineering decisions. I become an expert in my field so my clients don't have to. André is an expert in his field so I don't have to be. Now if another expert in the field comes along and says André is doing it all wrong then we have something to worry about.

So if we are not going to have a down in the weeds description of what changed and why then what do we tell people who are depending on the program results? We tell them there has been a refinement in the way the results are calculated that will tend to lower the energy production prediction over that from earlier versions of the program and the change should provide more accurate results, with the same error bars. How will it effect systems that were profiled in the past? It won't, once a system is built the only thing that matters is real production verse what the contract requirements are. If we have systems in the pipeline and still have time to rerun the sim and update the contract then we can do that and see how the new numbers change the outlook. Once the contract is signed then it does not matter what application changes take place, the only thing that matters at that point is real production verse what is in the contract. You're on your own about the version issue. What version you used for the sim is between you, the client, and the contract when you stand before the judge.

Great discussion. I look forward to more.


Top
   
PostPosted: Fri Aug 02, 2013 3:40 pm 

Joined: Thu Feb 28, 2013 12:48 am
Posts: 16
Location: Charlottesville, VA, USA
I have tried to make clear that I am not asking for an explanation of the detailed algorithms and inter-functionality of the PVSyst tool, nor am I suggesting that this tool should be expected to be a tool to predict the future. I view it as counter productive to simplify this discussion to those points.

Let me more briefly summarize the important issues that I am trying to discuss in my previous posts on this subject:

1) A major change was made to the tool from version 6.08-6.10 (6.09 was immediately replaced by 6.1). By "major" I mean that it impacted previous simulations of tracking systems (at least the horizontal SA trackers I have modeled with 6.1) by between 2-3% in annual production. My personal feeling is that this version change should not have been released as an "upgrade", this change in shading modeling is a fundamental change from the original v6.0 release, not a bug fix, so in hindsight I feel that a parallel installation should have been encouraged at a minimum. This change has made it impossible to rerun any tracking simulation performed in any version prior to 6.10 without significant alterations to the original model. By making this comment I am trying to encourage the developers of PVSyst to consider these impacts for future changes. This is not a complaint forum so that is not what I am doing here.

2) The new losses for tracking systems labeled "Near shadings: irradiance loss" that account for adjacent row/column diffuse radiation shading seems rather high to me compared to other array geometries at the same geographic location, using the same meteo file, and the same system losses. I am not asking to see the formulas but rather am asking what we have as users of the tool to better understand this effect. Third party papers, studies, white papers, etc. are examples. The best would be one or more independent studies comparing these new PVSyst methods to other methods and better yet, actual field performance. As the industry has developed over the years, other issues have generated similar discussions and independent analysis of modeling software's assumptions, including PVSyst.

3) Numbers generated by PVSyst (and other tools though PVSyst is clearly the most widely used) are used by the market to assess the value of PV systems so as much as possible the industry should be aligned on these numbers. For this discussion the "accuracy" of this tool is relevant I feel. By "accuracy" I mean how well does it model arrays deployed in the field that match the variables as input in the tool and the meteorological conditions. In other words has someone seen the level of diffuse shading loss on backtracking single axis trackers that are now built into the tool? Most probability analysis performed by independent consultants hired by investors include uncertainty factors for irradiation and weather resources, but also for the modeling tool (perhaps 5% on average). Is there an argument to be made that these changes should now lower that uncertainty for the PVSyst tool? I think we all understand that there will be a range or error band to all energy modeling tools but as we know, investors tend to focus on worst case. Most people accept PVSyst based numbers a representing a P50 number. If our tools become overly conservative this will keep trending the value of PV systems lower. That is just a general comment by me that I ask others to consider, I do not have a particular ask of the developers other than to continue to support their users as much as possible to understand the numbers generated by the tool.

Jason Fisher, Charlottesville, VA, USA

_________________
Jason Fisher
Charlottesville, VA, USA


Top
   
PostPosted: Fri Aug 02, 2013 10:53 pm 

Joined: Thu Aug 02, 2012 3:38 am
Posts: 105
Well, when it really comes down to it I think they are providing the application they describe on the website, the description being kind of non-specific anyway. I agree with you that the software release needs to be more transparent. Looking at the list of changes for this version it's hard for me to tell that a major change was made in the modeling calculations that would result in a change to the results for a large number of systems. PVSyst should have done a better job of highlighting this so people could evaluate the change and how it might effect the work they do.

As for the rest I can understand where you are coming from but that's just not the service PVSyst is providing. Normally this would result in competing products to choose from but this is such a narrow market segment and the choice of what program we use is often dictated by non-technical clients that our hands are often tied.

So what do we tell the financial folks when they see this difference? How about, "This is the program you are making us use, the program developers made a change in the shading calculation that they feel will make the program more accurate and it resulted in a 3% decrease in production for some projects. Do you want us to recommend other programs that we feel might be better?" I think that would be an amusing conversation. Oh and tell them shading is a bear. :shock:


Top
   
PostPosted: Thu Aug 08, 2013 10:19 am 
User avatar

Joined: Mon Apr 16, 2012 7:29 pm
Posts: 1658
You are right.
Diffuse losses over 1.5 to 2% seem quite unrealistic (I did only observe 1% to 1.5% during my development).
And getting higher losses with backtracking is not logical.
I will carefully check my calculation.


Top
   
PostPosted: Thu Aug 08, 2013 7:03 pm 

Joined: Thu Aug 02, 2012 3:38 am
Posts: 105
Sounds like you need better QA on your changes André. People do depend on the developers of PVSyst putting their best effort into making sure changes are both sound in theory and in implementation.

This just reinforces what I said about overstating to anyone the level of accuracy of any energy production estimation program including PVSyst. They are just estimates with wide error bands, not something anyone can take as fact.


Top
   
PostPosted: Thu Aug 08, 2013 11:46 pm 
User avatar

Joined: Mon Apr 16, 2012 7:29 pm
Posts: 1658
I have carefully checked the program and everything seems to be OK.

For a tilted single axis system, I have analysed a clear day and a cloudy one, and everything is quite coherent. It is interesting to observe that the albedo loss is of the same order of magnitude than the loss on diffuse, either on the clear day, and in the monthly results.
Attachment:
File comment: Clear day diffuse and albedo losses for Normal tracking and backtracking systems
Tracking_Diffuse_ClearDay.png
Tracking_Diffuse_ClearDay.png [ 59.86 KiB | Viewed 21071 times ]
I also have analysed the monthly and yearly values, and I observe that as expected, the Shading losses on diffuse and albedo, for normal system, are higher than the ones with backrracking.
I don't see a higher loss with backtracking, as reported by other users. If you have such data with single-axis projects, please send me the full project.

NB: for dual-axis systems, there are several backtracking strategies possible. Depending namely on the nechanical tracking system.
In PVsyst only one has been developed in the present time: keeping the tilt according to the sun's height, the backtracking is applied to the azimuth. In this mode (probably far from being optimal) the diffuse and albedo losses may be very high, and there is no reason that they are lower than with not-backtracking systems.


Attachment:
File comment: Global results for both tracking options. Detailed contributions of all components: Beam, Diffuse and Albedo.
Tracking_Diffuse_Monthly.png
Tracking_Diffuse_Monthly.png [ 125.57 KiB | Viewed 21071 times ]


Top
   
Display posts from previous:  Sort by  
Post new topic  Reply to topic  [ 15 posts ]  Go to page 1 2 Next

All times are UTC+02:00


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB® Forum Software © phpBB Limited