FROM FORECAST TO FLEXIBILITY
Tomorrow’s prices.
Today’s decisions.
Compare a linear benchmark with quantile boosting.
Then put the forecasts to work
in a constrained battery.
Choose a run or start a synthetic demo. Demonstration results are not market evidence.
Research readiness: exploratory. Original forecast publication evidence and a final
untouched holdout remain unresolved.
Ready
01 / PRICE DISTRIBUTION
A range, not a promise.
ActualLightGBM medianLEAR80% interval
Run an experiment to inspect its out-of-sample forecasts.
Axis timestamps end in Z = UTC. Delivery-day selections use Europe/Berlin; a local day can
start on the previous UTC date.
02 / FORECAST QUALITY
Skill, measured.
No results yet.
The DM comparison uses matched daily absolute losses.
03 / STORAGE VALUE
Forecasts into euros.
Totals for the selected experiment
Battery assumptions load with the selected experiment.
04 / PHYSICAL DECISIONS
Charge. Hold. Discharge.
Choose a completed experiment.
Charging (+)Discharging (−)
Choose a completed run.
Stored energy appears here.
Separate axes: power in MW above; stored energy in MWh below. Z means UTC; the selected
delivery day uses Europe/Berlin.
05 / COST & PHYSICAL SENSITIVITY
Test the assumptions behind the value.
One assumption changes at a time. Every policy is reoptimised using the same frozen forecasts and settled at actual prices. Perfect foresight uses the matching settings. These comparisons do not select a winning strategy.
Sensitivity results load with the selected experiment when available.
Costs are in EUR/MWh; capacity in MWh; efficiency is a fraction. Values exclude market impact, capital costs and unconfigured grid/tax charges.
06 / REPRODUCIBILITY
Know what the result means.
Trial accounting
Archive inventory is generated by research_audit.py; earlier unarchived research trials
remain unknown.
Data available locally
Assumptions & limitations
- No forecasts or battery results have been generated yet.
Experiment log
No active job.