Data for Residual-Ambiguity-Aware Sensor Retrofit for Distribution Network Topology Identification

Description

This dataset supports the article “Residual-Ambiguity-Aware Sensor Retrofit for Distribution Network Topology Identification.” It contains sensor-placement configurations, experimental settings, numerical evaluation results, and supporting data for the IEEE 33-bus and IEEE 123-bus studies. The files document the primary comparison, independent validation, classifier and noise robustness tests, and residual-screened sequential forward selection experiments.

Steps to reproduce

Download and extract SEGAN_REPRODUCIBILITY_PACKAGE_V1.zip.Annotation 1 Open a terminal in the extracted package root. The minimal reproduction workflow requires Python 3.11 and NumPy 2.3.5. python -m venv .venv .venv\Scripts\activate python -m pip install -r environment/requirements-minimal.txt On Linux or macOS, activate the environment using source .venv/bin/activate. Then execute the following commands in order: python code/validate_package.py python code/validate_n13r80_raw.py python code/placement_algorithm.py all python code/quick_reproduce.py python tests/test_quick_reproduce.py validate_package.py verifies every packaged file against the supplied SHA-256 manifest. validate_n13r80_raw.py checks the primary IEEE 123-node raw archive, comprising 128 data shards, 4096 operating scenarios, and 110592 scenario-topology AC power-flow solutions with no excluded solver failures. placement_algorithm.py all reconstructs the reported IEEE 123-node and IEEE 33-bus sensor layouts from the frozen placement inputs. quick_reproduce.py recomputes 21 reported performance metrics and 17 paired 95% confidence intervals using 10000 operating-scenario cluster-bootstrap resamples. Reproduced tables are written to the reproduced/ directory and automatically compared with the archived reference values using an absolute tolerance of \(10^{-12}\). The final command performs an end-to-end smoke test and should report PASS. This quick workflow reproduces the reported metrics and statistical comparisons from the frozen raw data and prediction artifacts; it does not retrain the classifiers or rerun the OpenDSS simulations. The complete software environment, experiment configurations, sensor sets, model receipts, raw-data manifests, and file hashes are included in the package for independent auditing.

Institutions

• East China Jiaotong University

  Jiangxi, Nanchang

Categories

Licence