KAHM-Based Power System State Classification with Differential Privacy

This repository contains the code, data, and results for the project on state classification of the IEEE 39-bus system using a Kernel Affine Hull Machine (KAHM) with differential privacy, federated learning, and false data injection attack (FDIA) detection.

Important: Execution Environment

The code is a single-file PowerFactory script (Code/KAHM_Based_Power_System_State_Classification_.py) designed to run inside DIgSILENT PowerFactory (with Python API) on a university virtual machine. It imports the powerfactory module at the top of the file.

Absolute file paths (e.g., R:\Computational_Intelligence\...) refer to that environment.

Because PowerFactory is a licensed software, and because powerfactory is imported at module level, the file cannot be imported or executed at all on a standard personal computer without a PowerFactory installation present — even though most of the analytical code (the KAHM classifier, differential privacy mechanism, federated aggregation, and FDIA attack/detection modules) is pure Python and does not itself call the PowerFactory API.

All outputs and logs from a successful run are provided in the Results/ and Simulation_log/ folders, enabling full inspection of the methodology and results without needing a PowerFactory license.

If you have access to a PowerFactory installation, you can adapt the paths in Code/KAHM_Based_Power_System_State_Classification_.py (see the __main__ block at the bottom of the file) and run the pipeline via PowerFactory's embedded Python environment.

Repository Structure

• Code/ — Single-file PowerFactory pipeline implementation (KAHM_Based_Power_System_State_Classification_.py)
• Power_Network_Data/ — IEEE 39-bus network input data (buses, lines, generators, loads, transformers, shunts, SVC, system-level CSVs)
• Report/ — Written report and code-explanation PDF
• Results/ — Generated outputs:
  • Classification_Outputs/ — KAHM/DP accuracy, confusion matrices, distances, gamma CV results (CSV)
  • Network_Outputs/ — Per-scenario/per-contingency bus, line, generator, and transformer results (CSV)
  • HTML_REPORT.html — Dashboard summary
  • convert_all.py — Output conversion script
• Simulation_log/ — Full run log (psa_kahm_20260705_194236.log) corresponding to the results reported in the paper

Methodology

A complete description of the methodological approach, parameter choices, and experimental design can be found in the accompanying written report.

Author

Chirag Narula Msc Student @ Uni Rostock chiragnarula1525@gmail.com /chirag.narula@uni-rostock.de Matriculation Number: 226100055