Replication Data for: Voltage-Based Faulty Section Identification in Rural Networks: Addressing IBR Integration and SWER Circuit Challenges

This repository contains the simulation files, converted data, final datasets, and Jupyter Notebooks necessary to fully replicate the findings of our research paper.

⚠️ **VERSION 2.0.0 UPDATE:** Following a rigorous peer-review process, this repository has undergone a massive methodological upgrade. The framework now features a optimized Multi-layer Perceptron (MLP) architecture, multiresolution feature extraction, Ablation Studies, and extreme stress-testing datasets for High Impedance Faults (HIF) and operational noise.

## Repository Structure

The dataset and scripts are organized into sequential stages representing the complete upgraded data pipeline:

* **`1_Simulations_ATPDraw_ATP/`**: Contains the base simulation models (`IEEE34NODE_DG.acp`/`.atp` and `IEEE34NODE_DG_SWER.acp`/`.atp`).

* **`2_Data_Generation_Code/`**: Contains the Jupyter Notebooks used to dynamically inject fault parameters and automate the ATP engine execution. It includes scripts for both standard fault scenarios and the newly added High Impedance Faults (HIF).

* **`3_Converted_Data_MAT/`**: Contains the foundational dataset comprising 2,800 `.mat` files generated from the raw `.pl4` files covering standard fault resistances (up to 50 Ω). *(Note: Due to file size limitations, this folder is provided in compressed .rar or .zip volumes).*

* **`3_Converted_Data_MAT_High_Impedance/`**: Contains the stress-test dataset comprising 2,240 `.mat` files specifically generated for High Impedance Faults (HIF) ranging from 100 Ω to 250 Ω. *(Note: Due to file size limitations, this folder is provided in compressed .rar or .zip volumes).*

* **`4_Machine_Learning_Dataset/`**: Contains the signal processing notebooks used for Clarke Transform, DWT, and noise injection (Data Augmentation). Crucially, this folder contains the extensive `.csv` datasets generated for our **Ablation Study**, covering different wavelet coefficient combinations (`cD1`, `cA1`, `cD1+cA1`, `cA3+cD3+cD2`) across varying Signal-to-Noise Ratios (Original, 40dB, 30dB, and 20dB).

* **`5_Machine_Learning_Code/`**: Contains the scripts for the fully upgraded diagnostic pipeline, including comprehensive Model Benchmarking, Hyperparameter Tuning via 10-fold Cross-Validation (`1_Model_Benchmarking_Tuning_and_Ablation.ipynb`), and the final proposed SMOTE-enhanced MLP architecture (`2_Proposed_MLP_Classifier_pipiline.ipynb`).

## Data Generation and Processing Methodology

### 1. Simulation Automation

Initially, the base simulation models were developed in ATPDraw and exported as base ATP files (Folder 1). The Python scripts in Folder 2 loaded these base `.atp` files, dynamically injected specific fault parameters (faulted bus, resistance, inception angle, and fault type), and executed the ATP engine in batch mode to generate over 5,000 fault scenarios.

> **Note:** The foundational automation algorithm is detailed in our preprint: *Automatic Generation of a Fault Database in Electrical Power Distribution Networks Using ATPDraw/ATP* (DOI: 10.21203/rs.3.rs-4661055/v1).

### 2. Data Conversion (PL4 to MAT)

The raw `.pl4` files generated by the automated ATP runs were converted into MATLAB-compatible `.mat` matrices (Folders 3) using the **PL42MAT** routine (August 2009 release). The Python extraction scripts in Folder 4 read these matrix structures to construct the comprehensive machine learning datasets.

## Prerequisites & Tools Used

* ATPDraw (Windows version 7.2) and Alternative Transients Program (ATP)

* PL42MAT (August 2009)

* Python 3.x (Jupyter, Pandas, NumPy, Scikit-learn, PyWavelets)

* 7-Zip, WinRAR, or any compatible archive extractor (required to decompress the multi-part `.rar` or `.zip` dataset volumes)

## Acknowledgments

The authors acknowledge the Federal Institute of Piauí (IFPI) for the academic leave granted to Francisco Carlos Moreira Abreu, providing the necessary support for this work.

## Contact

For any questions regarding the data, code, or methodology, please contact:

**Francisco Carlos Moreira Abreu**

Federal Institute of Piauí (IFPI)

Email: carlos.abreu [at] ifpi.edu.br

ORCID: [https://orcid.org/0009-0002-3495-9837](https://orcid.org/0009-0002-3495-9837)