Machine learning in energy forecasts with an application to high frequency electricity consumption data
This file contains Python codes related to the research paper 'Machine learning in energy forecasts with an application to high frequency electricity consumption data'. (see https://www.uni-marburg.de/en/fb02/research-groups/economics/macroeconomics/research/magks-joint-discussion-papers-in-economics/papers/2021-papers/35-2021_heilmann.pdf ) The file 'Paper experiment code' contains the complete experiment of the paper, but without the used dataset 'HessianLoad'. The dataset can be requested at https://www.uni-kassel.de/eecs/en/sections/intelligent-embedded-systems/downloads . The codes of the approach Auto-LSTM can be found in the sub file 'rl_forecast'. Together with the dataset 'HesseanLoad' the main file 'experiment_file.py' (respectively 'util.py') is in principle fully functional. However, we strongly recommend NOT to run the experiment, because of very long calculation times. The numerical results of the experiment are can be found in the sub file 'results'. As addition to the full experiment, we provide an easy example in the file 'Easy Example'. The main code 'EASY EXAMPLE.py' is finctional with the provided exemplary data. The example contains the simplified application of the approaches ARIMAX and ANN on the basis of a shortened input dataset. It can be used to reproduce the machine learning rpcoess, introduced in the paper.
Steps to reproduce
You need a environment to run Python. A list of all Python packages installed during the develpoing process is appended (note, that probably not all the packages are relevant). The main files of the experiment code ,'experiment_file.py' (respectively 'util.py' in rl_forecast), are in principle fully functional. We recommend not to run the experiment files, because of very long calculation times. The code 'EASY EXAMPLE.py' is fully functional and can be used with the provided exemplary data. For further information, please contact the author.