Spain’s energy transition is being driven by the National Integrated Energy and Climate Plan (PNIEC), which establishes a sustainable strategic framework aiming for 81% renewable electricity generation by 2030. The PNIEC foresees the installation of an additional 160 GW of renewable capacity, broken down as follows:

Photovoltaic solar energy: 76 GW, including 19 GW for self-consumption.

Wind energy: 62 GW, of which 3 GW will be offshore wind.

The PNIEC sets targets that will establish Spain as a leader in photovoltaic energy, but it also brings challenges related to profitability and integration of various renewable technologies into distribution grids—key aspects for ensuring a balanced and sustainable transition.

Among the strategies adopted by the renewable sector are self-consumption, battery installations (stand-alone or hybrid models), and Balancing Services.

Batteries allow energy to be stored during low-demand periods and released during consumption peaks, optimizing operational flexibility and profitability. Meanwhile, System Balancing Services (SSAA) generate additional revenue to help stabilize the grid and ensure system reliability.

However, this transformation also requires an evolution in current energy forecasting models.

Next-generation challenges in renewable energy forecasting

In the field of renewable energy forecasting, machine learning algorithms for time series have established themselves as the paradigm that delivers the most accurate predictions for energy generation in wind and photovoltaic plants.

These complex systems are fed with both structural and geolocation data of the plant, as well as historical production data and historical and forecasted data of the main meteorological variables impacting each technology. This information allows the models to capture the specific dynamics of each installation while considering the topographical characteristics of its location.

Forecasting models are usually calibrated using energy export data published by the Grid Operator, which are ultimately used to calculate settlements for deviations between forecasted and actual energy generation.

However, the calibration process based on this information is far from trivial. Models start from the assumption that historical data reflects a scenario of 100% plant availability.

In practice, factors such as grid constraints (curtailments) or periods of unavailability can distort the representativeness of the data. Therefore, it is essential to accurately identify and label these events, ensuring that the algorithm can distinguish between normal operating conditions and atypical situations. This process is key to avoiding biases during model training and ensuring that forecasts are robust and reliable.

When a long historical dataset is available, sometimes spanning several years, algorithms can be calibrated more accurately to the plant’s operational behavior. This allows them to capture seasonal patterns, long-term trends, and variations inherent to the installation itself.

However, when historical data is insufficient, the calibration process becomes an iterative and progressive exercise. As new data accumulates, algorithms refine their parameters, gradually improving their predictive capabilities.

This approach, however, faces challenges in scenarios such as battery integration, self-consumption systems, or participation in Balancing Services. In these cases, historical data based on energy export may lose validity, as new operational conditions significantly alter the generation dynamics of each asset.

Even in well-trained plants, incorporating this new historical data can have a disruptive effect, forcing interruptions in the usual adjustment process.

Incorporating SCADA and additional operational data

To address these challenges, it is necessary to adapt forecasting models to the new realities of the power system using an asset-oriented approach.

The key lies in obtaining data that reflects the plant’s operational reality beyond just export data. An effective solution is the integration of advanced monitoring systems such as SCADA (Supervisory Control and Data Acquisition).

SCADA provides real-time data on the status of the plant and its various components, including:

Meteorological variables: Accurate data from sensors installed at the plant, such as solar radiation, wind speed, or temperature.

Producible energy: Information on the energy generated by the plant at any given moment based on the available resource, regardless of grid or storage limitations.

Battery status: Charge level, charge and discharge cycles, and other critical parameters for plants with storage systems.

Internal consumption: Data on the energy consumed by the plant itself or by self-consumption systems, allowing differentiation between actual generation and consumption.

La integración de datos proporcionados por sistemas de monitorización avanzados como SCADA (Supervisory Control and Data Acquisition) en los modelos de predicción, no solo mejora su precisión, sino que también los hace más robustos a los cambios en las condiciones operativas. Al basarse en información que refleja la realidad física de la planta, los algoritmos pueden adaptarse mejor a escenarios complejos, como la gestión de baterías o la respuesta a señales de control en tiempo real.

Advanced forecasting for efficient renewable energy management

In summary, to achieve a reliable forecasting system adapted to the requirements of the PNIEC, it is essential to integrate detailed operational data, implement advanced SCADA systems, and adopt emerging technologies that enable optimal management of renewable generation.

At Gnarum, we provide our clients with personalized support to implement forecasting systems that take these variables into account, ensuring better results and minimizing deviations in energy production.