26–27 Feb 2020
RMIB
Europe/Brussels timezone

Deep learning methodologies for seamless short-range forecast for renewables

Not scheduled
20m
Conference Room (RMIB)

Conference Room

RMIB

Ringlaan 3 B-1180 Brussels Belgium

Speaker

Ms Petrina Papazek (ZAMG)

Description

Seamless forecasting from nowcasting to days-ahead for the energy sector allows for targeted trading, maintenance, and providing active power grid safety measures such as re-scheduling of solar and wind power production farms. As direct NWP model output often is too coarse for targeted forecasts, post-processing is needed. In the past decades, machine learning tools are being used more frequently in the energy community to provide such forecasts. Here, a deep learning-based methodology for seamless forecasting for up to two days ahead is presented.
Data of NWP models, gridded observations, and point data of both meteorological observation sites and renewable energy sites (wind turbines and solar power sites). The proposed method combines a convolutional neural network (CNN), a long short-term memory (LSTM), and fully connected layers into a common neural network. As CNNs efficiently solve image processing tasks and gridded meteorological data can be considered as images, they are applicable to gridded data sources. LSTMs are used in speech recognition and, thus, suitable for time series tasks. In addition to the deep learning forecasting methodology, different spatial and temporal resolutions, topographic information, and data mining-pre-processing of the input data are applied.

The proposed methodology is in a first step applied to meteorological observation sites. In a second step, it is applied to selected wind power and solar power sites in Austria.

Primary author

Co-authors

Presentation materials

There are no materials yet.