You are here : Home > Low carbon energy > Solar Energy/Solar PV > PV Forecasting

Solar photovoltaic energy: forecasting production and predicting system faults

PV Forecasting

Published on 7 May 2019

Forecasting solar PV energy production is a strategic challenge for government agencies, electric utilities, and managers of facilities like parking garages with EV charging stations. Knowing in advance how much solar photovoltaic energy will be available is crucial to planning ahead for backup energy from other sources, using stored energy to offset dips in production due to fluctuations in the weather (for example in island and other highly-variable climates), and, generally, to more effectively manage the different sources of available energy to achieve greater savings and reduce CO2 emissions. Just as important to PV plant operators is the capacity to predict potential system faults and estimate their impact on production. This requires measurable performance indicators (non-disruptive to production) that can be used to determine the relevant predictive and curative maintenance.

We have been researching PV system troubleshooting since 2006 and forecasting since 2009. Our researchers developed a forecasting solution under the ReactivHome project (financed by the French National Research Agency). The solution was able to predict how much solar PV energy would be available to a building the following day. A platform with forecasting tools leveraging the physical operating principles of a PV plant was developed—a break with the statistical and mathematical models that were the state of the art at the time. Using physical modelling on equipment data and actual measurement data results in a highly-effective learning model that naturally corrects over time the gaps between forecast data and actual measurements.
The technology was tested successfully during other research projects where accurate forecasting was important, in fields like home energy systems, mobility (electric vehicles), and energy management for commercial properties. Rolling out the technology in such a broad range of situations led to substantial improvements. "For example, a 24-hour forecast does not meet all needs. Therefore, 3-hour, 6-hour, and 15-minute forecasting capabilities were developed," said the head of the program. While the 24-hour forecast mainly uses meteorological data to determine the solar irradiance on the panels, the 3- and 6-hour forecasts use satellite data, a paid resource that provides irradiance data at a given point in time as well as cloud movement speeds. The 15-minute forecast uses 360-degree images captured by a camera at the site; the camera observes cloud formation and provides information that helps generate forecasts. "Integrating the three types of data into a learning model ensures that the solution becomes more and more robust over 10, 20, 30, and 40-day time horizons. The error rate rapidly drops to just 5% to 10% in European climates."
The current challenge is to improve the solution. Our researchers are leaving no stone unturned when it comes to reducing the already-low error rates even further. We are focusing on the 24-hour forecast, first by separating the direct and diffuse radiation in the spectrum and by integrating temperature data. So, rather than just using the primary data, the model also takes into account what the panel is actually receiving and capturing. This makes the meteorological data even more powerful. With each improvement achieved by our photovoltaics experts, the solution gains a few tenths of a point in terms of performance.

The CEA is also investing in troubleshooting research, and funded its own program on electric arc detection and termination in 2006. In 2009, under the DLD PV program funded by the French National Research Agency, we investigated fault signatures and identification and classification methods. Currently, we are focusing on two promising alternatives, one that leverages thermal images captured on-site, and another that entails plotting and analyzing the IV curve for each panel.


Solutions already available on the energy market

  • The market for solar PV production forecasting is marked by fierce competition; a number of companies offer very reliable 24-hour forecasts. Liten's solution is different in that it combines three time horizons (24-hour, 6-hour, 15-minute). Only two providers in France can deliver this type of data.

  • SteadySun, a startup founded in 2012 by a Liten researcher, is carving out a position on a variety of markets that could benefit from this technology. The company has already signed contracts with a number of energy providers in Europe and the United States seeking technologies to better predict demand and provide the backup energy needed to keep the grid operating as intended. Two additional solar-energy trading solutions are also available, providing users with data crucial to their trading every fifteen minutes.

  • Liten's solar PV forecasting technology is being used in a number of research projects at the national and EU levels, providing data that will drive further advances in this field.

  • The ReactivHome project (2009), funded by the French National Research Agency, laid the foundations for today's forecasting technology. The goal of the project was to enhance energy production and consumption at the individual-building level according to cost, environmental impact, and peak demand. The project partners were Liten, G2Elab, G-SCOP, Schneider Electric, and Orange Labs.

  • From 2009 to 2012, several projects funded by French Energy Agency ADEME leveraged technologies developed at Liten. One such project was Opera, which focused on securing an island grid (Mayotte). Another was DHRT2, a technology-development partnership between Toyota, the French National Solar Energy Institute INES, and the CEA to improve the overall building/vehicle energy system. Finally, the pioneering smart-grid project Reflexe, also funded by ADEME under the French government's economic stimulus package, was led by Veolia Environnement and brought together four premier partners: Alstom Grid, Sagemcom, CEA-Liten, and electrical engineering school Supélec.

  • The EU Horizon 2020 project Tilos, which kicked off in 2015, includes a forecasting-model work package where Liten is contributing to solar irradiance forecasting. This project brings together fifteen research institutes and businesses from seven European countries.

  • In 2012 Liten formed a partnership with Socomec to research electric arc detection. The goal is to release a product on the US market in late 2015-early 2016.

  • In 2012, Liten also formed a partnership with Urbasolar to enhance troubleshooting and management tools for PV plants.

  • ​Around ten researchers


Contact an expert to find out more