Success case

Electricity consumption forecast

The challenge

For electricity providers, having an accurate prediction of energy required to satisfy consumer demand is crucial, as anticipatory purchasing is less expensive. Currently, they are predicting energy demand 8 days in advance with an hourly resolution.

Objectives

Minimize the costs of anticipatory energy purchasing for excess or insufficiency, through accurate energy demand estimation. Excess energy means selling the surplus at the last minute or, conversely, buying at market price at the last hour.

How have we done it?

  • Temporal fusion Transformer
  • Feature engineering of weather information
  • Temporal derivatives
  • Customer profiling
  • Amazon Deep AR
  • Gluons & MX Net (Python)
  • AWS Sagemaker

The result

The new model reduces error by 52% when compared to the current model. This implies a cost reduction of approximately €100k every month and a half for the relevant group of clients on the most significant tariff, which is the 3.0 tariff.