Electric load forecast using Long-Short-Term-Memory (LSTM) recurrent neural network
Branch: master
Clone or download
Latest commit f656522 May 20, 2018
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data Added electric data .rar file May 20, 2018
.gitattributes Initial commit May 20, 2018
.gitignore Updated .gitignore May 21, 2018
LICENSE Initial commit May 20, 2018
README.md Update README.md May 20, 2018
__init__.py Initial commit May 20, 2018
data.py Initial commit May 20, 2018
lstm.py Refactored code for readability May 21, 2018
main.py Refactored code for readability May 21, 2018
plot.py Refactored code for readability May 21, 2018

README.md

lstm-electric-load-forecast

The task is to predict values for a timeseries of the history of over two million minutes of the power consumption of a household. The dataset can be found in the data folder as a .rar file which must be unzipped or alternatively in: https://archive.ics.uci.edu/ml/datasets/Individual+household+electric+power+consumption#

After downloading the data make sure the file is stored in the data folder.

I used a three layer multiple-input multiple-output LSTM recurrent neural network to predict future 5 minutes using previous 10 minutes. The code is modular so you can specify the number of minutes to consider in one step for prediction as well as the number of predictions. For e.g: Use the previous 30 minutes to predict the next 15 minutes.