Machine learning-based probabilistic forecasting of solar irradiance in Chile (2411.11073v2)

Abstract: By the end of 2023, renewable sources cover 63.4% of the total electric power demand of Chile, and in line with the global trend, photovoltaic (PV) power shows the most dynamic increase. Although Chile's Atacama Desert is considered the sunniest place on Earth, PV power production, even in this area, can be highly volatile. Successful integration of PV energy into the country's power grid requires accurate short-term PV power forecasts, which can be obtained from predictions of solar irradiance and related weather quantities. Nowadays, in weather forecasting, the state-of-the-art approach is the use of ensemble forecasts based on multiple runs of numerical weather prediction models. However, ensemble forecasts still tend to be uncalibrated or biased, thus requiring some form of post-processing. The present work investigates probabilistic forecasts of solar irradiance for Regions III and IV in Chile. For this reason, 8-member short-term ensemble forecasts of solar irradiance for calendar year 2021 are generated using the Weather Research and Forecasting (WRF) model, which are then calibrated using the benchmark ensemble model output statistics (EMOS) method based on a censored Gaussian law, and its machine learning-based distributional regression network (DRN) counterpart. Furthermore, we also propose a neural network-based post-processing method resulting in improved 8-member ensemble predictions. All forecasts are evaluated against station observations for 30 locations, and the skill of post-processed predictions is compared to the raw WRF ensemble. Our case study confirms that all studied post-processing methods substantially improve both the calibration of probabilistic- and the accuracy of point forecasts. Among the methods tested, the corrected ensemble exhibits the best overall performance. Additionally, the DRN model generally outperforms the corresponding EMOS approach.

• The paper improves the calibration and accuracy of solar irradiance forecasts using ensemble post-processing techniques.
• The authors compare EMOS and DRN approaches, demonstrating that the machine learning-based DRN outperforms traditional methods.
• The findings offer actionable insights for optimizing photovoltaic output and grid stability in solar-rich regions of Chile.

Machine Learning-Based Probabilistic Forecasting of Solar Irradiance in Chile

The paper "Machine learning-based probabilistic forecasting of solar irradiance in Chile" provides a detailed examination of methods for enhancing the predictability of solar irradiance, crucial for the integration of photovoltaic (PV) energy into Chile's power grid. This paper is particularly significant given Chile's reliance on solar power, with 39.7% of its energy demand met by PV, notably in sun-rich regions like the Atacama Desert. Despite the abundant solar resource, the variability of solar irradiance poses challenges for accurate energy production forecasts, which is where this research contributes.

The authors use ensemble forecasts generated by the Weather Research and Forecasting (WRF) model, which, while providing probabilistic predictions through multiple runs with varying initial conditions, still present issues like uncalibration and bias. To address these issues, the paper applies advanced statistical post-processing techniques, specifically ensemble model output statistics (EMOS) and distributional regression networks (DRN). These methods are evaluated using data from 30 locations across Regions III and IV of Chile, offering insights into optimizing short-term solar predictions for such regions.

In detail, the paper compares the traditional EMOS approach utilizing a censored Gaussian distribution with a machine learning-based distributional regression network (DRN) method. Additionally, the authors develop a neural network-based post-processing technique aimed at further improving ensemble forecasts. Evaluating these post-processing approaches involves assessing ensemble predictions against station observations for calibration and accuracy improvements.

Key Findings:

1. All post-processing methods tested significantly enhance both the calibration of probabilistic forecasts and the accuracy of point forecasts compared to raw WRF ensemble outputs.
2. Among the various techniques, the corrected ensemble approach was identified as having the best overall performance. This method also exhibited the most substantial improvements in prediction reliability.
3. The DRN model generally outperformed the EMOS approach, suggesting that the incorporation of machine learning can provide more precise probabilistic forecasts.

Implications and Future Directions:

The paper exemplifies how robust statistical and machine learning methods can be applied to improve renewable energy forecasts, directly aiding the management of energy resources in solar-rich regions. From a theoretical perspective, the enhancement over traditional models via neural networks and advanced statistical techniques illustrates the evolving landscape of forecast modeling in meteorology and energy applications.

Practically, the improvements in forecast accuracy highlight potential operational benefits for energy grid managers in Chile, potentially reducing the impacts of solar power variability on grid stability and efficiency.

The paper paves the way for future research into more sophisticated multivariate and temporally-consistent forecasting methods, including exploration into other machine learning architectures or hybrid models that combine data-driven approaches with physical models. As machine learning continues to mature, its integration with traditional forecasting systems will likely become more prevalent, offering the potential for increasingly accurate and reliable renewable energy forecasts.