Predict 48 hours ahead short-term electricity consumption/generation for an energy supplier in the Netherlands.
The dataset aggregates household and office consumption into a single time series. Competitors may use Open-Meteo historical weather data and no lag features of the otuput as in reality data comes with three days lag . Train on 80% of the data; the remaining 20% fore testing. Evaluate the models first on that testing data with nrmse, and nmae. Final rankings will be verified on an additional private holdout not published on Kaggle.

Goal: Build a machine learning model in python with NRMSE < 5% and NMAE < 5% on the private holdout.
The public leaderboard uses NRMSE for ranking.