The “Weather Intelligence for Renewable Energies” Benchmarking Exercise on Short-Term Forecasting of Wind and Solar Power Generation

Simone Sperati,Pierre Pinson,George Kariniotakis,Stefano Alessandrini

doi:10.3390/en8099594

Abstract

A benchmarking exercise was organized within the framework of the European Action Weather Intelligence for Renewable Energies (“WIRE”) with the purpose of evaluating the performance of state of the art models for short-term renewable energy forecasting. The exercise consisted in forecasting the power output of two wind farms and two photovoltaic power plants, in order to compare the merits of forecasts based on different modeling approaches and input data. It was thus possible to obtain a better knowledge of the state of the art in both wind and solar power forecasting, with an overview and comparison of the principal and the novel approaches that are used today in the field, and to assess the evolution of forecast performance with respect to previous benchmarking exercises. The outcome of this exercise consisted then in proposing new challenges in the renewable power forecasting field and identifying the main areas for improving accuracy in the future.

Highlights

The renewable energy production share is constantly increasing due to both the scarcity of fossil fuel resources and to some strategic incentives made at global level to reduce carbon emissions in the atmosphere
A probabilistic approach focuses on informing about the full range of potential events in terms of power generation, for example through a set of conditional probability density functions (PDF) or a set alternative scenarios
The results show a best score of 9.5% mean absolute error (MAE)/nominal power (NP) and 43.7% MAE/mean power (MP), achieved again by id06

Summary

Introduction

The renewable energy production share is constantly increasing due to both the scarcity of fossil fuel resources and to some strategic incentives made at global level to reduce carbon emissions in the atmosphere. Probabilistic predictions can be based either on ensemble models, in which a model is run different times from slightly perturbed initial conditions [2], or otherwise produced using statistical methods (e.g., quantile regression [3]). This provides both a prediction about the probabilities of occurrence of a given event (i.e., produced power greater than a given threshold) and information about the expected uncertainty affecting any issued single value forecast. It is becoming widespread, especially regarding wind energy

Results

Discussion

Conclusion