Forecast wind power and electricity production of wind turbines
This repository contains several models and allows to easily modify some hyper parameters.
Classic batch generator to configure, which provide inputs and teaching signals
Extract any sequence randomly in the training set and feeds it to the network
Use this model for a many to many architecture such as on this figure. This model has to be fed time_steps (via the config) samples as input and will predict forecast_steps steps in the future.
Model should be trained with forecast_steps=1. The model uses the Stateful parameter of keras, and predictions of the model are used as input for next steps, such as in the figure below.
The oversimplyfied persistence model used as a benchmark algorithm.
File containing an ARIMA model to configure and test for benchmarking.