Supplementary Materials — engproc-4521576
Reactive-Power-Based Feasibility Assessment of Distributed Generation Using Machine Learning Surrogates

CONTENTS
File S1_dataset_opf_revision_20000_all.csv
  Complete 20,000-scenario OPF output. Main fields include scenario ID, solver status/termination,
  DG connection bus, P_DG, loading factor alpha, signed optimal Q_DG*, |Q_DG*|, voltage extrema,
  0.95-1.05 p.u. compliance label, voltage violation measures, OPF solve time, and objective value.

File S2_dataset_opf_revision_20000_compliant.csv
  Subset of 12,300 scenarios that satisfy the 0.95-1.05 p.u. engineering voltage criterion.
  This subset is used for Stage-2 regression.

File S3_OPF_dataset_generation.ipynb
  Python/Pyomo notebook used to generate the revised scenario set.

File S4_Stage2_regression_training.m
  MATLAB script for DT, RF, SVR, and NN regression with 10-fold cross-validation and inference timing.

File S5_Stage1_feasibility_classifier.m
  MATLAB script for the Stage-1 DT/RF feasibility classifier using stratified 10-fold cross-validation.

File S6_Stage2_regression_metrics.csv
  Out-of-fold regression metrics reported in the manuscript.

File S7_Stage1_classifier_metrics.csv
  Out-of-fold classification metrics for DT and RF.

File S8_Stage1_OOF_predictions.csv
  Out-of-fold Stage-1 predictions and feasibility scores for the solver-successful scenarios.

File S9_Stage1_metrics_by_bus.csv
  Classification performance broken down by DG connection bus.

REPRODUCIBILITY SETTINGS
- Random seed: 20260911
- Candidate DG buses: 2-22
- P_DG sampling range: approximately 0-1.8 MW
- Loading factor alpha: 0.45-1.10 according to the mixture described in the manuscript
- Numerical voltage envelope: 0.70-1.15 p.u.
- Engineering compliance band: 0.95-1.05 p.u.
- DG reactive-power limit: +/-0.5 MVAr
- DG apparent-power rating: 2.0 MVA
- Soft voltage penalty: M = 1e6
- Slack regularization weight: w_s = 1e-3
- OPF: Pyomo + IPOPT, default solver tolerances
- Stage 1: stratified 10-fold cross-validation
- Stage 2: 10-fold cross-validation; only P_DG and alpha standardized for SVR/NN

EXECUTION ORDER
1. Run S3_OPF_dataset_generation.ipynb to regenerate the OPF datasets.
2. Run S5_Stage1_feasibility_classifier.m using S1 to reproduce Stage-1 classification results.
3. Run S4_Stage2_regression_training.m using S1/S2 to reproduce Stage-2 regression results.

NETWORK SOURCE
The feeder model is derived from the publicly available IEEE 34-Node Test Feeder:
https://site.ieee.org/pes-testfeeders/resources/
