Correct monitoring can help cut PV plant losses by more than 99.5%, according to a speaker appearing at Solar Asset Management Europe in Milan, Italy. Gerhard Mütter, technical director of Alternative Energy Solutions (AES) in Vienna, Austria, and Rising Alternative Energy in Delhi, India, told Solarplaza that AES had achieved this level of reduction on a 250 MW plant portfolio in Eastern Europe.
“We reduced in two years, from 2013 to 2015, from 4.5% of the possible yield that the plants lost, down to 0.02%,” he said. “The approach that we used was just to track every string using our self-developed analysis and improvement tool, called AES PIT.”
The case in question largely involved the advanced analysis of production data from PV strings to crack down on sloppy operations and maintenance (O&M) practices. However, said Mütter, correct monitoring was increasingly important for all PV asset managers as their plants get larger and older. “It is underestimated what impact on your work the size of the plant has,” he said.
“For example, if you have a 10 MW plant, which is not a large one, you need 20 hectares, or 400 metres by 500 metres.”
The need to walk around about a kilometre just to get from one end of the plant to the other means O&M processes can take much longer than expected, he said. It can also lead to unexpected local weather effects. “Even a cloud that is coming at 10 kilometres an hour will take three minutes to pass this park,” said Mütter.
This potential difference in local weather conditions from one end of a plant to another can create nonsensical results if you have limited sensor information. Another challenge with large plants is the ability to carry out visual inspections. A 10 MW plant using 250 W modules will have a total of around 40,000 panels. In a visual inspection, “you will be blind after the first five rows,” Mütter said.
Hence it makes sense to combine visual inspections with AES PIT results that can guide O&M teams to areas of the plant where production seems to be failing. Another potential problem for large plants is wear-out effects, where components can become damaged through movement or the action of animals.
Data analysis: tips for O&M best practice
- 1. Do not bother with module-level sensors on plants of more than 10 MW.
- 2. Trim your data sets according to the advanced analysis you want to carry out.
- 3. Aim for one-minute data sampling at string level.
This damage may not be easily visible to the eye and can be exacerbated with rapidly changing temperature levels that put extra stress on the components. “Even if it’s on 0.1% of your locations, you have four or five broken modules which you will not find with normal methodologies,” said Mütter.
To deal with such problems, AES recommends optimising data collection from the PV plant. On very large plants, however, it is easy to generate so much data that analysis becomes impractical. “On a 100 MW plant with 400,000 modules it doesn’t make sense to have records of every module,” Mütter said. “The optimum density is to have records at string level, or on input channels with two or three strings connected.”
Input channel sensors provide consolidated data for between 30 and 50 modules, which is still sufficient for relatively rapid trouble-shooting. Nevertheless many large-scale plants, above around 5 MW, are now built with string-level sensors, he said.
As well as how many modules to cover with each sensor, asset owners should consider how often they should sample data. Mütter said sampling sensor information every second would yield too much data for rapid analysis. Meanwhile looking at it on an hourly basis would not provide an accurate measure of how rapidly O&M teams were responding to faults.
The optimum sample time to see if a string is not working is once a minute, he said, but when looking out for particular effects over long-term data sets then even that level of granularity might be too much for effective analysis. Thus, for example, if an asset manager is concerned about the effects of rapid temperature changes on production then it does not make sense to analyse periods where the weather did not change.
Instead, “you have to find out when there was a fast change in temperature and during this time slot you take all minute records and do your calculations. That is one of the tricks to survive the calculation time,” said Mütter.
In a similar way, when checking actual production against a simulation it is usually sufficient to work with daily or even monthly data. Mütter said most large asset owners already had the technology in place to optimise their data analysis, but there “huge differences” in the way these analyses were being carried out.
“There are companies that just read the monthly report and compare it against budget, which is totally nonsense,” he claimed. “You have deviation against budget in both directions because a lot of it is down to the weather.”
The exact level of benefit that can be achieved with better data analysis is highly site specific but most asset owners will gain from an improved understanding of their plants. In the example of the 250 MW plant portfolio in Eastern Europe, for example, much of the reduction in losses came from data that revealed poor O&M practices.
In one instance, a growing loss in production at one end of the park, furthest from the entrance, turned out to be due to vegetation that was not being cut back by the O&M team. Such impacts are not unusual in newly commissioned plants everywhere, Mütter said, because vegetation often grows more quickly than usual on disturbed earth.
Being able to uncover such problems is easy with good data, said Mütter, and advanced data analysis using AES PIT on top of monitoring is not expensive. Once an advanced analysis system such as AES PIT is set up, it can cost just a few hundred euros a month for a 10 MW plant. As a result, asset owners can typically expect between a five-fold and 20-fold return on their investment in analytics, he said.