Power consumption forecast is a vital tool that enables Transmission System Operator (TSO) to optimize performance of the grid and balance energy production and demand. It is a crucial step in order to ensure security, stability, and profitability.
The aim of the project is to predict energy consumption for the next day (D+1) both on the national and regional level (13 districts) during public holidays in France.
We created an artificial intelligence network based on the state-of-the-art sequence learning algorithms (GRU and LSTM) which is meant to simulate behavior of the energy consumers. The data was provided by RTE Transmission System Operator (http://www.rte-france.com/) through a challenge launched on platform DataScience.net (https://www.datascience.net/fr/challenge/32/details).
- Python 3.5
- Pandas 0.20
- Matplotlib
- Tensorflow 1.0 +, GPU support with CUDA 8.0 and cuDNN 6.0
- Keras
- numpy
- scipy
- yaml
- HDF5 and h5py
- Elephas (http://maxpumperla.github.io/elephas/) – library for distributed deep learning with Keras and Spark
- liblapack-dev
- libblas-dev
- gfortran
- Spark
- PySpark
- Hyperas (http://maxpumperla.github.io/hyperas/) – hyperparameter optimization library for Keras