The real-time operation of the electricity system, carried out by Red Eléctrica as the Spanish TSO, must ensure its correct operation and guarantee the continuity and security of the electricity supply by monitoring the system at all times.

Electricity has the peculiar characteristic of having to be generated at the same time that it is consumed, given that it cannot be stored in large quantities, therefore, it is essential to maintain a consistent balance between the amount of electricity required and the amount produced. Therefore, in today's climate, in which electricity sources are increasingly diversifying, and particularly within the energy transition process where the cornerstone is renewable energy production, there is a need for continuous development and improvement of prediction models for system operation. Furthermore, given the uncertainty associated with renewable energy sources, which depend mainly on the wind and sun, a development towards probabilistic forecasting models capable of limiting this uncertainty is essential for ensuring the associated risks can be managed.

These statistical methods are not based on a single model, but rather on the application of a compendium of models that simultaneously combine all the necessary models in order to predict the demand and generation variables that affect the electricity system's operation as accurately as possible. Therefore, these prediction models (for both demand and generation) provide important benefits for electricity system operation:

1. They combine different statistical models depending on how accurate they are for each situation, instead of relying on a single model for the whole forecast. This prevents misspecification errors and biases.
2. The information from different models is made available to avoid erroneous predictions in situations that generate anomalous data.
3. They allow strategies to be developed for optimising the combination of different forecasters, according to previously tested empirical and theoretical results.

In order to create this optimal combination of forecasters, Elewit, in collaboration with Red Eléctrica and the College of Industrial Engineering (Escuela Técnica Superior de Ingenieros Industriales - ETSII) of the Polytechnic University of Madrid, has launched a research project to improve and develop the methodology used in these forecast combination models.

This project's objective is to optimise management of the electricity system, providing operators with probability ranges for the combined forecast that are adapted specifically for them. The more appropriate the combination of forecasters that make up the statistical model resulting from the project, the better renewable energy generation can be managed, and the more useful the predictions obtained through techniques based on analytics and big data will be.

Solution provided by the project

The purpose of this project, named CONPP, is to develop a methodology for constructing forecast probability intervals, combining the intervals of individual forecasters that best approximate the required estimates, given a series of initial forecasting models that reflect the daily demand and production curves in the Spanish peninsula.

The main objective of this research is to develop a procedure for the construction of forecasting intervals linked with forecasters obtained as a combination of independent forecasters. The method requires the calculation of a variance matrix for the prediction accuracy of the individual forecasters, the EWMA (Exponentially Weighted Moving Average) method is proposed for these estimations, in order to achieve a dynamic adaptation of the model that suits the current situation.

Project results

CONPP has successfully developed a methodology to calculate the probability ranges of the combined wind forecast based on the probability ranges of the individual forecasts used in the combination.

The project has also designed a methodology to evaluate the probability ranges provided by the individual forecasts.

As a continuation of the project, it is expected to explore and evaluate its implementation in photovoltaic demand and generation, as well as to expand its use to the rest of the non-peninsular systems.