The transition to a decarbonized electricity system has turned renewable energy generation forecasting into a strategic asset. As solar and wind penetration increases, balancing supply and demand depends more than ever on accurately predicting how much energy will be produced and when. This is no easy task, given the intermittent and variable nature of solar and wind resources, both subject to short-term fluctuations that are difficult to predict.

Transmission system operators (TSOs), utilities, developers, and asset managers all require accurate forecasts to make informed operational, financial, and regulatory decisions.

In this context, Ravenwits has positioned itself as a leading innovator in applied renewable forecasting. Their work has shown that powerful models are not enough; effective forecasting requires solutions tailored to real-world plant operations, dynamic atmospheric behavior, and the limitations of available data.

Thanks to this approach, Ravenwits has developed tools that help clients anticipate production more accurately, reduce imbalance penalties, improve renewable integration, and optimize the management of assets such as batteries and hybrid plants.

New challenges, new architectures

Historically, renewable forecasting relied on traditional statistical methods such as linear regression and autoregressive models. While these approaches provided reliable service for years, especially in systems with lower renewable penetration, they have begun to show limitations as systems become more complex and dynamic.

The advent of artificial intelligence has brought a step change in forecasting capabilities. Ravenwits has successfully implemented Convolutional Neural Networks (CNNs) to interpret weather maps as images, significantly improving forecast accuracy. These networks can detect complex spatial patterns in the evolution of clouds, temperature, or irradiance, phenomena that classical models struggle to capture.

The next big challenge is to model the complex spatial relationships between different plants and geographical areas. For this, Ravenwits is exploring more advanced architectures such as Graph Neural Networks (GNNs), which represent each plant as a node and learn how they influence one another over time and under different weather conditions. For instance, if a dense cloud affects a plant in the west, a GNN can anticipate its impact on a neighboring downwind plant a few minutes later, something conventional methods cannot easily model.

One of the most demanding tasks at present is photovoltaic nowcasting, i.e., forecasting solar generation over the next few minutes up to two hours. This is essential for managing the effects of cloud movements and localized phenomena poorly captured by traditional weather models. Ravenwits is developing dedicated models using satellite imagery and CNNs to tackle this.

The challenge of scalability is also being addressed. With thousands of renewable installations worldwide, each with its own data and local conditions, the goal is not only to forecast accurately but also to do so automatically and at scale. To achieve this, cloud-based platforms are being developed to validate data, train models, and monitor their performance with minimal human intervention, ensuring reliable and efficient service even with limited personnel.

These initiatives are supported by the Community of Madrid through its 2024 Grant Program for the Application of Artificial Intelligence in Industry.

Tangible impact: from research to real-world operations

The technologies developed by Ravenwits are not just theoretical proposals or academic papers. These services are commercially offered to major energy stakeholders, including transmission system operator Red Eléctrica de España (REE), and large asset owners and managers such as EDP, Nexus Energía, and Repsol.

For a TSO, having a more accurate forecast means less reliance on balancing reserves, fewer energy curtailments, and better congestion management. For a plant manager, it translates into smarter decisions about when to dispatch energy, store it, or trade it in intraday markets.

Ravenwits’ innovations also enable these benefits to be scaled up. The ability to adapt models across different technologies, regions, and forecasting horizons makes it possible for a wide range of installations, from large-scale wind farms to industrial self-consumption systems, to benefit from advanced forecasting tools.

At a time when the global grid is undergoing unprecedented transformation, renewable energy forecasting has become a key tool for ensuring system stability and efficiency. In this context, Ravenwits stands out with solutions that are innovative, scalable, and aligned with the demands of today’s energy market.