Improving Model Chain Approaches for Probabilistic Solar Energy Forecasting through Post-processing and Machine Learning

TL;DR

This paper advances probabilistic solar forecasting by evaluating how post-processing can be embedded in a GHI-to-PV model chain and by testing neural-network-based post-processing alongside a direct PV forecasting model. It systematically compares four post-processing strategies, demonstrates that post-processing PV outputs yields the largest gains, and shows that hour-of-day information via hourly models or embeddings improves calibration. Neural-network post-processing provides incremental benefits over EMOS, while a direct NN PV forecast offers a competitive alternative that bypasses the model chain. The findings highlight practical paths to improve probabilistic solar forecasts and suggest directions for future multi-site studies and integration with emerging AI-based weather models.

Abstract

Weather forecasts from numerical weather prediction models play a central role in solar energy forecasting, where a cascade of physics-based models is used in a model chain approach to convert forecasts of solar irradiance to solar power production, using additional weather variables as auxiliary information. Ensemble weather forecasts aim to quantify uncertainty in the future development of the weather, and can be used to propagate this uncertainty through the model chain to generate probabilistic solar energy predictions. However, ensemble prediction systems are known to exhibit systematic errors, and thus require post-processing to obtain accurate and reliable probabilistic forecasts. The overarching aim of our study is to systematically evaluate different strategies to apply post-processing methods in model chain approaches: Not applying any post-processing at all; post-processing only the irradiance predictions before the conversion; post-processing only the solar power predictions obtained from the model chain; or applying post-processing in both steps. In a case study based on a benchmark dataset for the Jacumba solar plant in the U.S., we develop statistical and machine learning methods for post-processing ensemble predictions of global horizontal irradiance and solar power generation. Further, we propose a neural network-based model for direct solar power forecasting that bypasses the model chain. Our results indicate that post-processing substantially improves the solar power generation forecasts, in particular when post-processing is applied to the power predictions. The machine learning methods for post-processing yield slightly better probabilistic forecasts, and the direct forecasting approach performs comparable to the post-processing strategies.

Figures (8)

• Figure 1: Schematic illustration of the different strategies for applying post-processing methods within a model chain approach for PV power prediction.
• Figure 2: Visualization of exemplary GHI ensemble predictions, GHI observations, and PV power observations at the Jacumba solar plant on January 21, 2020. Note that the PV power estimates are scaled by a factor of 50 to allow for a straightforward visual comparison with the GHI datasets, and that the clear-sky GHI values are included here for visualization purposes only.
• Figure 3: Exemplary probabilistic GHI forecasts based on the ECMWF ensemble and the EMOS hourly post-processing approach for dates in January 2020. The colored areas indicate central prediction intervals. The lines show the ensemble median.
• Figure 4: Hourly values of the CRPS (a), the bias (b) of the mean forecast, as well as the coverage (c) and width (d) of 96.1% prediction intervals for the GHI forecasts. All values are averaged over the test dataset.
• Figure 5: Verification rank histogram of the ECMWF ensemble forecasts and PIT histograms for all considered post-processing methods for GHI for hours 10, 14, and 18.