Case study

An Italian Energy Trader approached us to model the oscillations of the energy price.


We implemented and trained a Recurrent Convolutional Neural Network for a notable Italian Energy Trading company to forecast the sign of the oscillation of the price of energy on the market with 70% accuracy. The training has been performed with few hundreds of data points.

Our models were able to extrapolate a +12 and +24 hours energy price forecast.


1) Dataset integration (energy price, weather forecast, calendar)
2) Model selection: we implemented and tested several models in order to identify the one realizing the ideal trade-off between accuracy, computational complexity, and interpretability.
3) We fine-tuned the initial implementation to be modular and easily adaptable to different applications by a reasonably skilled programmer. The code is hosted on a repository along with its documentation, tests, and examples.