Compare improvements by T-forecasts customers and/or perfect T-forecasts customers

[JanMaartenvanDoorn]

Definition of ready:

• https://github.com/alliander-opensource/AIFES/issues/48
• https://github.com/alliander-opensource/AIFES/issues/49

WHAT: To assess the effect of improved customer transmission-forecasts we investigate how the overall forecast quality increases when t-forecasts are perfect.

HOW:

• Run three backtests:
  • Subtract the measurements of the large customers from the total load and run a backtest on the residual load.
    • Add "the perfect T-forecasts" (measurements) of the large customers
    • Add the actual T-forecasts of the large customers
  • Run a backtest of the total load
  • evaluate the quality of the three different backtest forecast timeseries
• [Optional] repeat the first back test with different degrees of artificial error (noise) added to the t-Forecasts and make a graph of degree of noise vs resulting forecast-quality
  • Compare all backtest forecasts with tooling in the test bench
• Draw conclusion

TEST:

• Notebook with analysis
• Conclusion about how strongly improved customer t-forecasts would improve overall forecast quality

[MartijnCa]

There is a problem with the data quality. The sum of the C-ARM measurements of all clients behind the substation Middenmeer does not add up to the total of the measured load on the substation.

The analysis of this issue cannot be performed as planned, but I copied the results of a relevant previously performed analysis to the SharePoint: https://alliander.sharepoint.com/:f:/r/teams/PortfolioSO/Gedeelde%20documenten/Collaboration%20Allianer%20-%20TenneT%20T-Prognose/Data%20and%20results/Results/%2312_improvements_current_and_perfect_customer_power_forecasts?csf=1&web=1&e=vfXOly