Solar forecasting with immediate scattering solar irradiance algorithms

An international team of scientists from Greece, Spain, Switzerland, France, the United States and the Netherlands have collaborated on a new paper investigating the application of solar irradiance nowcasting algorithms to the field of solar forecast. Their research was published in the journal Energies.

Study: Solar irradiance ramp prediction based on all-sky imagers. Image Credit: Simakova Mariia/Shutterstock.com

Solar forecasts and ramp events

The solar forecast is an indispensable tool for the solar energy sector. The efficiency of photovoltaic technologies can be significantly affected by environmental factors such as cloud cover, which means that an accurate prediction of solar irradiance is crucial to ensuring that solar energy technologies perform optimally. Therefore, accurate and efficient solar forecasts can significantly improve grid stability.

Forecasting is a complex undertaking that can be affected by sudden changes in solar irradiance over a short period. These sudden changes are known as ramp events. Currently, there is no formal consensus on the definition of a ramp event in the literature. Precise definitions are largely user and application dependent.

Applied Methodology Flowchart

Flowchart of the applied methodology. Image credit: Logothetis, SA et al., Energies

A ramp event can be identified by calculating the absolute power differences between the start point and the end point over a period of time. However, because solar irradiance is a dynamic and variable factor, it is not always possible to identify and accurately predict a ramp event. Strategies can be used to overcome these issues, but rely on user-defined thresholds to distinguish between ramping and non-ramp events.

Two broad categories of ramp events have been defined in the literature: solar irradiance events and solar energy ramps. Solar power installations can handle small amounts of uncertainty and variability, but there is a pressing need for accurate detection of extreme events that installations may not be able to cope with.

Several ramp event detection strategies have been introduced in recent studies, with algorithms prominent in these methods. Swing gate algorithms have been widely investigated by more than one study, and another strategy is to use all-sky imagers (ASIs) to provide measurements for deep learning methods.

One study attempted to correlate cloud types extracted from satellite images with variance in solar radiation. Another study combined cloud information with neural networks, using nowcasting algorithms. While there is a pressing need to provide accurate solar forecasting capabilities for the solar energy industry, this area has been largely unexplored.

The paper

This study is a benchmarking exercise under the IEA Photovoltaic Power Systems Program. The objective of this international framework is to strengthen collaboration in the field of solar energy to study the key role of photovoltaic technologies in the green economy.

The first benchmarking exercise of the study focuses on nowcasting irradiance across the sky. The exercise used data from a 2019 campaign in Spain that incorporated a bunch of whole-sky irradiance-based methodologies. Specifically, four systems were used in the project. Validation data revealed the effectiveness of this approach regardless of cloud cover conditions.

The second objective of the benchmarking exercise is to investigate the feasibility of using all-sky irradiance nowcasts for the prediction of ramp events. Brief descriptions of the applied methodologies and data are provided in the study, along with the methodology for detecting solar irradiance ramp events. The main objective of the study is to evaluate the effectiveness of all-sky irradiance nowcasts for the prediction of solar irradiance ramp events.

(a) Scatter plot and (b) CDF function of ramp rates (RR) for clear sky conditions at the time horizon

(a) Scatter plot and (b) CDF of ramp rates (RR) for clear sky conditions at time horizon D = 10 min. Solid black lines indicate the threshold boundary (99th quantile) that classifies non-ramp (red area) and ramped (green area) events. The bluish colored bar refers to the angle of the solar zenith. Image credit: Logothetis, SA et al., Energies

Conclusions of the study

The study presents a novel approach to detect and predict solar irradiance ramp events using whole-sky irradiance nowcasting algorithms. Some important conclusions were drawn in the document.

The connection between detected ramp events and nowcasting algorithms based on sky irradiance was revealed in the article. The algorithms can detect 55% to nearly 100% of ramp events. However, it was noted in the study that ASI-2 systems, which predict 65% of ramp events, produce spatial GHI predictions and stick to narrow areas in the benchmark, penalizing physical approaches, especially for the geolocation of clouds.

To overcome this problem, it is recommended to incorporate wider GHI forecast areas, which consequently improves the detection of ramp events. Additionally, UPS 3 through 5 suffer from false predictions that are not encountered with UPS systems 1 and 2. Increasing the forecast horizon decreases the accuracy of UPS-based systems. On average, the forecast accuracy for all systems is around 80%.

(a) Total number of ramp events measured and (b) the change in this number caused by the variation of the threshold limit (Thr) at each time horizon (D).  The vertical axis shows the change in the threshold limit, while the horizontal axis shows D. The color bar shows the number of ramp events (a) and the change in the number of ramp events (b) .

(a) Total number of ramp events measured and (b) the change in this number caused by the variation of the threshold limit (Thr) at each time horizon (D). The vertical axis shows the change in the threshold limit, while the horizontal axis shows D. The color bar shows the number of ramp events (a) and the change in the number of ramp events (b) . Image credit: Logothetis, SA et al., Energies

The relationship between forecast accuracy and underlying cloud conditions was also revealed in the study. Different cloud conditions were analyzed, with ASI 3-5, which are deep learning algorithms, outperforming other algorithms on days with overcast or scattered cloud conditions. Algorithms that incorporate physical steps work best on days with scattered clouds for half the day and no clouds for the other half.

The ASI-based nowcasting algorithms were also compared against other conventional forecasting strategies such as MAD and RMSD, demonstrating the improved performance of the forecasting strategy investigated in the study. The selection of the ASI-based nowcasting algorithm depends on the application. Additionally, the authors recommended that a combined approach would be beneficial for the accuracy of ramping event prediction.

Overall, the paper demonstrated the suitability of all-sky irradiance-based nowcasting algorithms as an integrated tool for predicting solar irradiance ramp events. This will improve the performance of solar power installations during extreme power fluctuation events, thereby improving efficiency and reducing costs.

Further reading

Logothetis, SA et al. (2022) Solar irradiance ramp prediction based on all-sky imagers Energies 15(17) 6191 [online] mdpi.com. Available at: https://www.mdpi.com/1996-1073/15/17/6191

Disclaimer: The views expressed here are those of the author expressed privately and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork, the owner and operator of this website. This disclaimer forms part of the terms of use of this website.

Sharon D. Cole