Reproducibility Capsule

The reproducibility capsule provides scripts to reproduce the experiments from the OpenDT paper.

Experiments

Experiment	Description	Configuration
Experiment 1	Power prediction without calibration	experiment_1.yaml
Experiment 2	Power prediction with active calibration	experiment_2.yaml

Running Experiments

1. Start OpenDT

Run with the experiment configuration:

# Experiment 1 (without calibration)
make up config=config/experiments/experiment_1.yaml

# Experiment 2 (with calibration)
make up config=config/experiments/experiment_2.yaml

2. Wait for Completion

With speed_factor: 300 and the SURF workload (~7 days), each experiment takes approximately 1 hour.

Monitor progress via the Grafana dashboard at http://localhost:3000.

3. Generate Plots

After the simulation completes:

python reproducibility-capsule/generate_plot.py

The interactive script will:

1. Prompt you to select an experiment (1 or 2)
2. Show available data sources with timestamps
3. Generate a plot comparing:
   • Ground Truth (actual power)
   • FootPrinter baseline
   • OpenDT prediction

Plots are saved to reproducibility-capsule/output/.

Output

Generated plots show:

• Left Y-axis: Power draw (kW)
• Right Y-axis: MAPE percentage
• X-axis: Time

Three time series are plotted:

• Ground Truth (actual measurements)
• FootPrinter (baseline predictor)
• OpenDT (digital twin prediction)

MAPE values for both FootPrinter and OpenDT are displayed in the legend.

Reference Data

The reproducibility-capsule/data/ directory contains reference data for the baseline comparisons:

• footprinter.parquet - FootPrinter predictions
• real_world.parquet - Ground truth measurements