From 18 March 2025, the Spanish electricity market will take a decisive step towards greater operational granularity with the implementation of the intraday market in 15-minute intervals (QH) and the settlement of deviations at the same resolution (ISP15).

From 1 May onwards, deviations will be settled on the basis of fifteen-minute granularity, in accordance with the Resolution of 28 March 2025 of the Secretary of State for Energy.

This marks the end of the transitional period during which REE has settled deviations by applying the same measurement to all fifteen-minute periods. Starting in May, settlements will be made on an hourly basis, either using actual meter data or interpolations, at all production facilities and consumption points in the electricity system.

This new update to the electricity system means that forecasting systems will also require hourly interpolation to quarter-hourly, so that the different market agents can adjust their positions in the intraday market, define fortnightly programmes and reduce the impact of deviations.

Below, we provide you with the keys to ensuring that these forecasts meet the new market requirements. 

  

The implementation of the quarter-hourly market (QH) in the Spanish electricity sector is a significant change that allows for more accurate allocation of deviations and more optimised operation of the electricity system. However, it requires agents to adapt their forecasting, operation and settlement systems to handle these more granular intervals.

From 1 May onwards, deviations will be settled on the basis of fifteen-minute granularity, in accordance with the Resolution of 28 March 2025 of the Secretary of State for Energy.

This marks the end of the transitional period during which REE has settled deviations by applying the same measurement to all fifteen-minute periods. Starting in May, settlements will be made on an hourly basis, either using actual meter data or interpolations, at all production facilities and consumption points in the electricity system.

This new update to the electricity system means that forecasting systems will also require hourly interpolation to quarter-hourly, so that the different market agents can adjust their positions in the intraday market, define fortnightly programmes and reduce the impact of deviations.

Below, we provide you with the keys to ensuring that these forecasts meet the new market requirements. 

  

The FORECAST Service enabled by Gnarum, allows producers to reduce their diversion costs from the early stages of implementation of each installation, through the various management platforms of the Last Resort Suppliers (CUR).

As a general rule, renewable energy producers use a Representative Agent in the free market to intermediate with OMIE and REE in the management and optimisation of the sale of their energy. 

However, it can take several weeks, or even months, from the time a plant starts feeding energy into the grid to the start of its representation contract. During this period of time, the various CURs set up a web platform so that the producer can manage its generation forecasts by means of manual uploads, or through a file in a specific format. 

If during this period generation forecasts are not available in accordance with the technology and geographical position of the plant, and therefore if these forecasts are not uploaded to the CUR platform, all the production generated will be diverted, with the economic loss that this represents. 

Point forecasting methodologies entail estimating a single scalar value as the anticipated future value of a target variable.

This approach remains ubiquitous across numerous industrial and research applications due to its provision of a definitive, easily interpretable output. 

In the context of the energy sector, for instance, a wind power forecasting model may output a point estimate of 35 MWh for generation during a specific hourly interval. The primary strength of point forecasting lies in its operational simplicity, which seamlessly integrates into decision-making workflows and resource-allocation processes. 

Nonetheless, this technique intrinsically omits any formal representation of uncertainty surrounding the forecast. As a result, reliance on a solitary estimate can precipitate substantial deviations if exogenous factors—such as abrupt meteorological shifts or turbine outages—differ from those assumed during model training. In contrast, probabilistic forecasting approaches yield a full probability distribution over potential outcomes, thereby enabling explicit quantification of predictive uncertainty. 

Rather than asserting that wind generation will be exactly 35 MWh, a probabilistic model might, for example, specify that there is an 80 percent likelihood that generation will fall within the 30–40 MWh range. This distributional insight permits risk-aware decision-making and more robust integration of forecasts into stochastic optimization routines.

Probabilistic forecasting model as a key approach to risk management

Conformal Prediction (CP) constitutes a robust framework for generating prediction intervals with formal validity guarantees, without imposing any assumptions regarding the underlying data distribution. Its principal advantage lies in its ability to deliver coverage guarantees at any prescribed confidence level, making it especially well‐suited for probabilistic forecasting in the energy sector.

In our experience at Gnarum, CP has proven to be an indispensable calibration paradigm, conferring enhanced reliability to forecasting systems.En la experiencia de Gnarum, CP se ha consolidado como un paradigma de calibración muy valiosa para aportar confiabilidad a los sistemas de forecasting.

Key attributes of Conformal Prediction include:

• Distributional Flexibility: CP requires no prior assumptions about the data distribution, allowing interval construction to adapt to arbitrary, real‐world data patterns.
• Model Agnosticism: The framework can be overlaid on any predictive model—from classical linear regression to complex deep neural networks—without modification to the underlying algorithm.
• Dynamic Adaptability: By tailoring calibration strategies to specific operational contexts, one can adjust the width of prediction intervals in real-time, based on the evolving performance of the base point‐prediction models.