Demystifying the use of ERA5‐land and machine learning for wind power forecasting
Open Access
- 1 December 2020
- journal article
- conference paper
- Published by Institution of Engineering and Technology (IET) in IET Renewable Power Generation
- Vol. 14 (19) , 4159-4168
Abstract
Wind is a highly unstable renewable energy source. Accurate forecasting can mitigate the effects of wind inconsistency on the electric grid and help avoid investments in costly energy storage infrastructure. Basing the predictions on open‐source forecast models and climate data also makes them entirely free of charge. The present work studies the feasibility of using two machine learning (ML) models and one deep learning (DL) model, random forest (RF) regression, support vector regression (SVR), and long short‐term memory (LSTM) for short‐term wind power forecasting based on the publicly accessible ERA5‐Land dataset. For each forecast model, a selection of hyperparameters is first tuned, followed by determining the best performing input data structure using surrounding data grid points and increasing the time interval of data affecting a single prediction. Both the ML models and the DL model perform better than the baseline (BL) model when forecasting wind speed up to 24 hours ahead. However, a reduced forecast duration is needed to achieve satisfactory wind turbine (WT) power output forecast accuracy. Most notably, the RF is able to produce 3‐hour forecasts with the combined WT power output prediction error amounting to less than 10 % of the WT's nominal power.Keywords
This publication has 40 references indexed in Scilit:
- Forecasting Short-Term Wind Speed Using Support Vector Machine with Particle Swarm Optimization2017
- Short‐term wind power forecasting and uncertainty analysis using a hybrid intelligent methodIET Renewable Power Generation, 2017
- An Empirical Study on Machine Learning Models for Wind Power Predictions2016
- Wind energy in Romania: A review from 2009 to 2016Renewable and Sustainable Energy Reviews, 2016
- Wind speed and solar irradiance forecasting techniques for enhanced renewable energy integration with the grid: a reviewIET Renewable Power Generation, 2016
- Joint optimisation of generation and storage in the presence of windIET Renewable Power Generation, 2016
- Isolated power system in Russia: A chance for renewable energies?Renewable Energy, 2016
- Two Machine Learning Approaches for Short-Term Wind Speed Time-Series PredictionIEEE Transactions on Neural Networks and Learning Systems, 2015
- Review on Methods to Fix Number of Hidden Neurons in Neural NetworksMathematical Problems in Engineering, 2013
- Long Short-Term MemoryNeural Computation, 1997