Solar Availability Series Part 4: Camelot’s Balanced Approach

elcome back for Part 4 of Camelot’s series on solar availability. If you’re just joining us for the series, here are some links to parts 1, 2, and 3. We’ve set the groundwork with a summary of the ongoing validation efforts from IEs, and the resulting changes the industry is making to their assumptions. We’ll revisit their reasoning here. We’ve also described how availabilities are calculated and reported, and touched on ways of maximizing availability by minimizing downtime. If you’ve followed along with the last few parts and you’ve been waiting for our own stance as an Independent Engineer (IE), look no further! Thank you for joining us.

Re-Setting the Scene

Until somewhat recently, the utility-scale solar industry didn’t have the kind of established history needed to accurately predict or validate what long-term average availabilities will be at newly-proposed projects. Engineering judgement said that a relatively simple solar project would see the equivalent of about 3-5 days of total site outages per year, leading to expected availabilities of about 98.5% to 99.2%. For modeling simplicity, most everyone assumed a relatively consistent availability throughout a project’s lifetime. However, as projects became operational, the industry started to question itself. Especially early in new projects’ operational lives, downtime was high and availabilities were lower than expected due to teething issues. Even after the initial startup period, many folks started seeing trends with their average availability levels below what they had hoped.

Over the last year we have started to see the beginnings of some robust data-backed approaches to redefining availability assumptions, aided by all the new operating data which is available to us. There have been three IEs who have recently updated their assumptions based on aggregated data from the projects they supported. ICF led the charge with its performance paper published by kWh Analytics in 2023. DNV and Natural Power followed suit with their own methodology updates in early 2024. Others with access to the data have weighed in as well, from NREL to kWh Analytics. Here, we focus in on the results of the IE validations, each of which took slightly different approaches and used different data sets. The table below summarizes the projects which went into the IEs’ comparisons, and some key comments from their results.

We’d like to highlight a few key findings from this comparison:

• Every IE relied on data from monthly operating reports produced by the operators, which are rarely independently calculated or verified. As described in part 2 of this series, there is no single, standard way that availabilities are defined or reported across the industry. The conclusions from these studies should be interpreted carefully, especially because the data QC processes have not been explicitly described.
• DNV’s analysis used more data and resulted in recommendations which are more clearly tailored to the sites.
• ICF found that fixed tilt systems showed lower availabilities than tracker systems while DNV found the opposite.
• Despite every IE noting lower availabilities early in a project’s life, only DNV adjusted their recommendation to treat the first year differently from other years.
• No IE has taken a stance on availability changes later in a project’s life yet.?

Here is a summary of the IE’s post-validation default availability recommendations. As you can see, only DNV makes a distinction between different kinds of projects at this time, though every IE noted that they are open to changing their assumptions based on project-specific data such as operator or technology history. In practice, however, IEs are often reluctant to deviate from their standard assumptions, as this requires going out on a proverbial limb. While that conservatism is understandable, it may be producing unintended consequences. For instance, if an IE will not give “credit” for more robust technology choices or operating strategies, then owners have little incentive to consider any options but those that can be considered “bankable” at the lowest possible cost. This approach penalizes owners for considering better than baseline equipment, spending more on O&M, or otherwise looking for creative solutions to improve availability.

The need for more data was a theme repeated by each company, and this will likely ring true for as long as we do this kind of work. Our availability assumptions will need to be updated regularly, just like we update our approaches to Energy Yield Analyses.

Camelot’s Recommendations

The Camelot team is compiling the data needed to supplement these studies and validate our conclusions, and we welcome the opportunity to work with industry partners on this effort. In the meantime, we base our own recommendations off the meta-study described above and in Part 1. Without further ado, here is our own take on availability projections:

1. Until we have more information, we should not be differentiating between different mounting types. ICF’s and DNV’s observations contradicted each other. It’s likely other factors influenced the analyses, especially the sample sizes and quality of the input data.
2. The factors which can impact downtime should be studied further, which means collecting more data, ensuring its accuracy, and capturing all potentially-relevant project details. In addition to mounting types, the difference between inverter technologies must be studied further as one of the primary sources of downtime observed at operating sites. For instance, the higher availability noted by DNV on smaller fixed-tilt sites than larger fixed-tilt sites may indicate a reliability advantage for string inverters over relatively small sites with central inverters. This would align with our general experience with operating sites but the data to positively confirm this is not yet available in sufficient quantity.
3. The major sources of downtime should be studied and modeled separately. Using an overall system availability as a metric can muddy the waters significantly, especially when trying to tease out the impact of different design decisions on future performance. When performing energy yield analyses for wind energy projects, some IEs will include assumptions for balance of plant availability, grid availability, and turbine availability separately. Not only can this improve our validations (data allowing), but it will improve the way we assess technology tradeoffs at the design stage. Swapping out a more robust system for a less-robust one should impact only the downtime assumption for that system. Camelot recommends the industry work towards a bottom-up availability model based on historical failure/downtime data at the module, tracker, inverter, MV, HV, and BOS levels. These levels correspond with likely failure points within the system and provide a lowest common denominator that can be adjusted during project design to optimize expected availability. Ensuring this approach has buy-in from IEs will provide a financial incentive to specify better equipment and design better sites.
4. Year-1 availability should be modeled separately from later years due to initial startup issues observed in each validation. Nearly all financial models are already set up to account for annually-varying losses, so adjusting our assumptions based on the clear signals we see from the data appears to be a no brainer.
5. The industry should start modeling a ramp-down in availability later in projects’ life, as DNV may have alluded to, because component failure rates impact availability trends. Without more data, it is difficult to say the magnitude of the decreases because of the other factors at play. However, based on our experience modeling availability at other infrastructure projects, Camelot considers it reasonable to model availability as a ramp-down as a project nears the end of its design life. The “bathtub curve” shown below is an Engineering concept which supports this idea. It shows how infant mortality failures likely contributed to the observed availabilities in the first 6-12 months of operation, and highlights the further need for more operational data as projects age.

?This is applicable to individual components in many physical systems. Aggregated across an entire system and accounting for typical replacements and maintenance, one might expect to see a flatter availability curve, but with some consideration for early- and late-stage failures. We have seen this already with 10-15 year old PV sites, where owners struggle to obtain compatible replacement equipment that can be “dropped in” to replace original equipment onsite. As technology continues evolving quickly, we can expect new module types, inverter technologies, sensing devices, and code requirements to all play a role in the maintainability of PV sites in the late stages of their useful life.

Camelot’s Balanced Approach

The summary below provides a graphical representation of each IE’s default availability recommendations over time, and includes Camelot’s own recommended defaults (when no other project-specific information is available). We note the following:

• Camelot’s approach accounts for the size impacts observed by DNV, which appears to be a strong signal in the data, but does not differentiate between technologies until more information is made available supporting the distinction.
• Much like DNV, Camelot’s recommended availability starts slightly lower in year 1 before reaching steady operations, as is supported by all studies.
• We recommend modeling availability declines after year 20 based on several factors, including the bathtub curve concept described above, the typical useful life for major components, and our expectation that the impacts of mid-life failures will likely offset by the efficiencies gained from experience during operations. While we see this assumption as a necessary recognition of late-stage wear-out failures, it’s worth noting that its impacts on a financial model are muted by the time value of money. On average, Camelot’s assumptions are less pessimistic than ICF, and strike a balance between the assumptions reported by Natural Power and DNV.
• Camelot will consider quantitative adjustment to our base availability assumptions for sponsor efforts that materially result in increased reliability, such as:

o?? Demonstrating better than average historical availability for project- specific equipment (e.g., inverters) through operational data (as described in item 3 above).

o?? Adding incentives to O&M Agreements for increased availability, beyond simply guaranteed levels

o?? Purchasing extra spare parts for more vulnerable system components likely to need frequent replacing

o?? Investing in predictive analytics and above-market O&M services to reduce the frequency and severity of unplanned maintenance events?

While these recommendations may be Camelot’s “default” values, as an IE which cares heavily about the accuracy of our projections, we will always consider factors such as operator experience or the relative track record of the technologies deployed at each site. As the saying goes, “show us the data.” Before we close, it is important to underscore an important point. Recent reporting that indicates PV projects are falling short of expected availability is a call to action for all of us. It is a call to action for more data, better analysis, and a deeper understanding of what causes PV systems to underperform. It is, notably, not a call to action for unnuanced conservatism. Simply whacking a few points off availability is, in our view, insufficient to the task of ensuring a better-performing PV fleet and it creates blind spots. We hope our fellow IEs will join us in not simply erring on the side of conservatism but, rather, will continue to advance our knowledge of these issues and build better, and more nuanced models that reward innovation, investment, and effort.

We hope you’ve found this series to be helpful, and we welcome the opportunity to partner with any of our readers who would be able to support with future efforts. Although this is the last of our solar availability series for now, we fully intend to revisit the topic in the future. For our storage-oriented audience, you can expect a similar discussion on availability assumptions for BESS technologies in upcoming articles.

About

Camelot Energy Group is a technical and strategic advisor to owners and investors in clean energy and energy storage projects, programs, and infrastructure. Guided by our core values of courage, empathy, integrity, and service we seek to support the energy needs of a just, sustainable, and equitable future. Our team has experience in supporting 7+GW of solar PV and 10+ GWh of energy storage and offers expertise in technology, codes and standards, engineering, public programs, project finance, installation methods, quality assurance, safety, contract negotiation, and related topics. Our services are tailored to a providing a different kind of consulting experience that emphasizes the humanity of our clients and team members, resulting in a high quality bespoke service, delivered with focus, attention, and purpose. Key services include: -Technical due diligence of projects and technologies -Owner’s representative and engineer support -Strategic planning -Training and coaching -Codes and standards consulting -Contract negotiation and support.