The increasing integration of renewable energy into the power system brings with it a new challenge: variability. Unlike conventional energy sources such as gas, coal, or hydropower, the production of solar and wind energy depends on ever-changing and uncontrollable weather conditions.
As a result, anticipating how much energy will be generated in the coming hours has become a complex yet essential task to ensure the stability and efficiency of the power system.
Why is short-term renewable energy forecasting so difficult?
The main challenge lies in the very nature of renewable resources. Wind is inherently variable. It can change direction and speed in short periods, though not always abruptly or unpredictably.
Moreover, these variations are not usually uniform across a region. Wind farms located near each other may experience significantly different wind intensities due to factors such as topography, natural or man-made obstacles, or the influence of local microclimates. This spatial variability makes accurate wind power forecasting difficult.
Unlike wind, the position of the sun can be determined with complete accuracy at any time of year and anywhere on the planet. This allows us to calculate exactly how much solar radiation a facility would receive under ideal conditions. However, in practice, the actual amount of solar energy reaching the panels depends on many other factors, one of the most important being cloud cover.
Clouds can drastically alter solar radiation within minutes, especially under partly cloudy or variable sky conditions. Knowing whether the sky will be clear or overcast in the coming minutes or hours is critical to estimating how much energy a photovoltaic plant will be able to generate. Yet anticipating such rapid changes remains a challenge.
The consequences of poor forecasting
When renewable energy generation forecasts are inaccurate, the consequences ripple across the entire electricity system.
To begin with, the system operator must compensate in real time for the difference between the forecasted and actual generation. This is done using balancing mechanisms, which often involve activating more expensive and more polluting backup power plants, such as gas-fired units.
In parallel, renewable energy producers who fail to deliver the committed amount of energy to the electricity market must cover the deviation. This is typically done by purchasing energy at the last minute on the balancing market, usually at higher prices.
As the share of renewables grows, forecast uncertainty makes grid operation more complex. Real-time decision-making becomes harder, especially during rapidly changing weather conditions.
All these additional costs can ultimately affect the price paid by end consumers. More accurate forecasting not only improves system efficiency but also helps contain electricity prices.
New solutions and tools for better forecasting
Traditionally, the industry has relied on classical statistical models, such as regressions or autoregressive models, that combine weather forecasts with historical production data.
In recent years, with the rise of machine learning, many companies have adopted deep learning techniques that enhance forecast accuracy. These models learn directly from large volumes of meteorological and production data and adapt better to the complexity and variability of renewable resources.
In this context, Ravenwits has developed and validated proprietary models based on convolutional neural networks (CNNs), a technique especially useful for analyzing weather maps, such as those generated by numerical weather prediction (NWP) models, by treating the data as images. These models have been successfully validated and are now used commercially by several leading players in the energy sector.
However, the potential for improvement doesn’t end there. As part of a project supported by the Community of Madrid through its Grant Program for the Application of AI in Industry (2024), Ravenwits is exploring the use of an even more advanced architecture: graph neural networks (GNNs). These models are particularly powerful for understanding how weather conditions propagate and affect different plants in an interdependent manner.
In a context where precision in renewable forecasting is more critical than ever, Ravenwits aims to improve the accuracy of commercial forecasting systems by incorporating cutting-edge artificial intelligence techniques.