nicknobad / forecastpowerdemand

Half-hourly UK Power Demand Forecast for the next 2 weeks using scikit-lean ML models

Work in progress in terms of splitting the big script into modules and making it more functional

Prerequisites:

1. API keys for elexon and weatherbit
2. python packages

At the moment the script works with Initial Demand Outturn. To train on the Final Demand Outturn data you have to download the historical datasets

Steps:

0. Inputs

1. FUN for holidays, doy, moy, dow

2. Configs

3. Download and prepare (clean, process and format) the datasets

   2A. Demand

   2B. Weather

4. Train, CV and Test

   4A. Simple CV

   4B. Predictions on train, val and test

   4C. ToDo: Plot best model preds in the test dataset vs real

   4D. ToDo: Tune several models and benchmark their performance

5. Download weather forecast

6. Download demand forecast from NG; Combine 6A with 6B

   6A. DA HHly demand Forecast from NG

   6B. OC2-14 Peak demand. Explode it to HHly.

7. Merge Fct Demand and Weather (from Steps 5 and 6) for Step 8. Get all time vars and Holidays Forecast HHly demand for the next 2 weeks and plot it