Next Generation Forecasting Solutions for Wind Energy – Results from the Smart4RES Project
Résumé
1. Abstract
Forecasting variable renewable energy (RES) production, and namely wind power, is a quite mature technology. The majority of actors (grid operators, aggregators, traders...) use today operational forecasting tools and services to optimize their decisions related to wind power integration in power systems and electricity markets The horizons of interest are a few minutes to a few days ahead. Despite that maturity, RES forecasting is a very active field of research internationally due to the high level of uncertainty (mainly from weather conditions), the increasing number of emerging use cases in power systems that involve development of new forecasting products and the need to continously improve accuracy especially in systems with high RES penetration. In fact, wind power prediction errors may reach very high levels (i.e. above 50%) especially in situations of weather fronts, ramps etc, and these may have a potentially high impact to the grid and result in financial losses in the markets. The H2020 Smart4RES project was initiated in 2019 to develop next generation forecasting solutions with increased forecasting performance and also innovative approaches to leverage the value (technical or economic) of forecasts by considering the whole model chain from weather forecasting to end-use applications. The project which will be completed in 2023 has followed some disruptive directions of research. In this Conference we will present the final results both regarding forecasting and also the optimal use of forecasts in a broad range of applications that range from grid management to storage/RES hybrid plant management and trading to multiple markets.
2. Method
The presentation will focus on highlight results from the project like:
* Ultra-high spatial/temporal resolution (i.e. 50m) forecasting of wind fields using LES (Large Eddy
Simulations).
* Seamless weather forecasting where ensemble forecasts from different models are coupled to cover multiple horizons.
* Seamless probabilistic wind power forecasting where a unique model was developped able to cover all prediction horizons and available inputs.
* Resilient forecasting where an approach based on robust optimisation is proposed to handle missing or corrupted data (i.e. from cyber attacks).
* Data sharing solutions that permit to share data while respecting confidentiality and privacy constraints and thus enable among others spatiotemporal forecasting.
* Solutions based on two completely new paradigms: value-oriented forecasting and prescriptive analytics. The later jointly considers forecasting and optimisation steps in applications.
3. Results
Although the project deals with RES forecasting in general this presentation will focus on the use cases that refer to wind power forecasting and will present the final results of the research project. The developed methods were validated with real-world data from various wind farms in Europe. The results on forecasting using LES are based on data from the island of Rhodes. The proposed methods like seamless RES forecasting are compared to state-of-the-art wind power forecasting methods benefits are quantified both in terms of accuracy and regarding simplification of the model chain. The value-oriented forecasting and the prescriptive analytics approaches are illustrated for the case of trading to energy markets. These later approaches are based on artificial intelligence enhanced with interpretability.
4. Conclusions
Smart4RES is the only ongoing European project on short-term forecasting and applications. The project has developed disruptive research supported by high-quality publications. New concepts have been introduced and validated. Seamless forecasting (a single model for all data, horizons and all RES technologies – replicable to PV and aggregations of PV/Wind) marks a corner stone in RES forecasting technology. Data sharing solutions bring benefits to wind power forecasting and are replicable to other sectors. The prescriptive analytics approach permits to link data to decisions through interpretable AIbased models. Finally, the forecasting solution proposed based on robust optimisation makes wind power forecasting resilient and becomes a paradigm on how resilience can be introduced in the intelligence layer of future power systems.
5. Learning Objectives
Aggregators and wind power producers will learn innovative approaches to trade on multiple electricity markets in order to increase revenue and mitigate technical and economic risks, including the case of a joint operation of wind and storage. Regulators and system operators will see the potential benefit of new forecasting products of weather and wind production in markets and systems with high renewable penetration. This can help them when selecting or monitoring forecasting services in the near future. Aggregators, wind power producers and system operators will be able to grasp the potential value of their data streams in a collaborative framework and understand the constraints related to privacy-preservation for a reliable application of data sharing in their daily process.
Mots clés
Renewable Energy
Forecasting
Wind farms
Wind energy
Electricity markets
Energy trading
Day-Ahead market
Artificial intelligence
AI
XAI
Explainable AI
Interpretable IA
Data science
Grid Management
Energy Meteorology
Robust forecasting
Solar forecasting
Privacy-preserving Data sharing
Data markets
Satellite images
Sky camera
Photovoltaics
Virtual power plants
Prescriptive analytics
Large Eddy Simulation
Ensembles
Numerical Weather Predictions
European project
Origine : Fichiers produits par l'(les) auteur(s)