search

demmojo / lstm-electric-load-forecast

Electric load forecast using Long-Short-Term-Memory (LSTM) recurrent neural network
MIT License
78 stars 31 forks source link
deep-learning keras lstm lstm-neural-networks rnn time-series-prediction

readme

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.