PV reliability lessons from 100,000 systems

By Dirk C. Jordan, Chris Deline, Bill Marion and Teresa Barnes, NREL. Mark Bolinger, Lawrence Berkeley National Laboratory.
The interrogation of data from 100,000 PV systems in the US provides fruitful insights into performance and reliability. Credit: First Solar.

Reliability plays a critical role in PV’s cost competitiveness with traditional energy sources. Many research groups and institutions around the world pursue to quantify PV field performance, degradation and failures.

However, data sets studying a large number of systems that provide a high-level overview of issues occurring in the field are still difficult to find []. In response to the global financial crisis of 2008, US Congress enacted the American Recovery and Reinvestment Act in 2009 (ARRA). Section 1603 of ARRA gave qualified renewable energy projects the option to elect a cash payment in lieu of the federal investment tax credit (ITC).

The award stipulated that annual PV production and comments relating to the performance needed to be reported. The data set comprised about 100,000 PV systems totaling to over 7 gigawatts (GW) direct current (DC) capacity or roughly 7% of the US fleet at the end of 2019. The insights gained from this data set provide valuable information of the performance and the state of reliability of the PV fleet in the USA. While the dataset is limited to systems in the USA the same lessons are more generic and may be applicable to other parts of the world.

Fleet performance

The data set consisted of annual production data for five years for each of the systems, the nameplate rating, an estimated production value and the location. The ratio of measured over predicted production could be calculated for all systems to assess system performance health. The data set is approximately divided into residential (1–25kW), commercial (25kW–1MW) and utility-scale systems (>1MW).

The division between groups is somewhat arbitrary but reflects the general trend between different types of systems, although individual systems at the respective limits may have been incorrectly classified. For systems over 5MW, in addition to the 1603 data, we generated our own production estimates using a separate data set acquired by the Lawrence Berkeley National Lab, which also included greater levels of detail on system specifics such as mounting configuration than was presented in the 1603 data. In general, we found good agreement between our own and the 1603 estimates, lending some credibility to the production numbers contained in the 1603 data set.

Figure 1. Cumulative distribution function of the five-year mean of the measured/predicted production ratio for normal systems not impacted by specific performance issues (top), and systems impacted by specific issues (bottom) discussed in the following sections. Different system sizes are indicated by different colour; the median (P50), the P90, and the unity ratio are indicated by dashed horizontal and vertical lines, respectively. A 10% band around the unity values is indicated in grey

The five-year mean of the measured over predicted production ratio is displayed in Figure 1 as a cumulative distribution function (CDF). The advantage of a CDF compared to a histogram is that it more easily allows comparison of multiple large distributions.

The top graph shows all “normal” systems, i.e. systems that were not knowingly impacted by some issue. The data are colour-coded by the size of the systems and the median or the P50 is indicated by a black horizontal dashed line, as is the P90 that is often used in financial models.

The unity value, i.e. systems performing as expected, is indicated by a vertical dashed line together with a grey 10% band around it. At the median, the CDFs of the “normally” operating systems show slightly higher production than expected. In addition, the utility-scale category exhibits a tighter distribution, indicated by a steeper curve, is most likely aided by closer supervision in the planning and operation phase and/or more accurate predicted values. The P90 value falls between 0.8 to 0.9; in other words, 90% of all systems produce approximately within 10% of the expected production.

The general asymmetry of the CDFs indicates the limited upside of the production ratio, but the much greater risk for energy loss. A minority of systems greatly underperform and overperform, clearly indicating a problem with the system, production estimate, or reporting. However, because no comments regarding the performance were entered, these systems had to be treated as “normally performing” systems and are included. An additional source of uncertainty might be the difference in accuracy of revenue grade meters typically used in utility-scale systems compared to standard meters more commonly used in residential applications.

The bottom graph of Figure 1 shows similar CDFs of systems that were impacted by specific issues in any of the five-year reporting period. Similar to the “normally” operating systems, some systems greatly under- and overperform because of the different impact of certain issues on performance. However, some general observations can be made: utility systems show a reduced performance at the median compared to “normal” systems, but they perform substantially higher than residential systems. This is a difference that we will explore in more detail below. Commercial systems fall between the utility systems (similar performance at the median) and the residential systems (similar performance at the P90).

This is an extract of an article first published in Volume 24 of PV Tech Power. The full article can be read here, or in the full digital copy of PV Tech Power 24, which can be downloaded via the PV Tech Store here


[[1]]        J. Leloux, L. Narvarte, D. Trebosc, “Performance Analysis of 10,000 Residential PV Systems in France and Belgium”, 26th European Photovoltaic Solar Energy, Hamburg, Germany, 2011. 

11 October 2022
PV CellTech Extra will be held as a series of live webinars and on-demand sessions on 11-13 October 2022. We'll be taking a slightly further forward-looking view at the technologies and roadmaps for new cell architectures set to dominate mass production during 2023-2025 with special emphasis on the potential timelines for technologies beyond the single-junction cell design, including perovskite and hybrid concepts.
29 November 2022
PV ModuleTech EMEA in Madrid on 29-30 November 2022 will address the factors underpinning the changing PV module landscape, gathering together all the key stakeholders across the value-chain from module production to field testing. Join us for presentations from the leading players in the sector, clearly identifying the opportunities and challenges set to impact module supply to Europe and the Middle-East over the next few years.

Read Next

September 2, 2022
Researchers at the US Department of Energy’s National Renewable Energy Laboratory (NREL) have constructed a perovskite solar cell that they say is both highly efficient and stable.
August 30, 2022
Thin-film solar manufacturer First Solar will invest up to US$1.2 billion to expand its manufacturing operations in the US, including setting up a vertically integrated factory in the country’s Southeast with an annual capacity of 3.5GWdc.
August 16, 2022
Chinese module manufacturer Risen Energy’s Hyper-ion module series has been tested and certified by German testing company TÜV SÜD.
August 10, 2022
A solar PV module frame made of recycled roll-formed steel has outperformed conventional aluminium frames in testing by an independent third party.
August 9, 2022
Renewables developer and operator Leeward Renewable Energy (LRE) has started the construction of a 196MW solar PV plant in London, Ohio.
August 4, 2022
The Biden administration has launched a US$26 million fund project to demonstrate the US electricity grid can reliably run on renewables.

Subscribe to Newsletter

Upcoming Events

Upcoming Webinars
September 27, 2022
15:30 AEST (UTC +10)
Solar Media Events
October 4, 2022
New York, USA
On-Demand Webinars, Solar Media Events
October 11, 2022
Virtual event
Upcoming Webinars
October 18, 2022
10am (EDT) / 4pm (CEST)