Abstract

This dataset supports the research paper "Synchronized Waveform Monitoring Unit Application: Lissajous DQ Curve to Improve Situational Awareness in Power Distribution Networks". It contains electromagnetic transient (EMT) simulation results from the IEEE 33-bus distribution system, including:

36,000 training samples (6,000 per class across 6 event categories)
18,000 validation samples (3,000 per class)
Pre-trained LSTM models for event classification

The dataset enables reproduction of the novel synchronized Lissajous DQ curve method, which compresses three-phase power quality events into single plots using Park Transform. Data is provided in both CSV (individual samples) and NPY (compiled arrays) formats, with dimensions (samples, timesteps, features) for direct machine learning applications. The accompanying pre-trained models demonstrate 100% classification accuracy while reducing computational requirements by 43-44% compared to conventional methods.

This comprehensive dataset supports research in:

Power quality event detection
Waveform data compression
Time-domain analysis of synchronized measurements
Machine learning applications in power systems

Instructions:

Instructions for Use

Dataset Structure

Training Data (6000 samples/class)
- csv/: Individual EMT simulation results (36,000 CSV files)
- npy/: Compiled arrays for machine learning (4 NPY files, shape: 36000×503×Z)
Validation Data (3000 samples/class)
- Identical structure to training data (18,000 CSV files)
Pre-trained Models
- 20 optimized LSTM models per method (.pth files)
- Naming convention: {num_of_epochs}_{learning_rate}_{hidden_layers}_{accuracy}.pth

Important Note

Before running any scripts: Update all file paths in the Jupyter notebooks to match your local directory structure.

Processing Workflow

Data Generation: 10-cycle EMT simulations (0.2s at 50Hz) of 6 event types:
- High impedance faults
- Capacitor switching
- Incipient faults
- Voltage dips
- 1-phase faults
- 3-phase faults
Data Preparation: Use PrepareData.ipynb to:
- Convert CSV to NPY format
- Separate by analysis method (VI, D, Q, Z components)
Model Training: LSTM.ipynb includes:
- 32-unit hidden dimension
- Batch size 32
- Adam optimization
- Hyperparameter search (epochs, learning rate)
Model Validation:
- Use LoadnTestModel.ipynb to evaluate performance on validation data
- Loads pre-trained models from Trained_model/
- Tests against Folder 2 NPY files
- Generates classification metrics and timing results

Recommended Usage

For waveform analysis: Use CSV files with timestep metadata
For machine learning: Load NPY arrays directly
For classification: Load pre-trained models from LissajousD/ or LissajousQ/
For model validation:
- Select optimal .pth file from Trained_model/ subfolders
- Modify LoadnTestModel.ipynb paths to point to:
  - Validation data (Folder 2 NPY files)
  - Selected model file

Funding Agency:

Indonesia Endowment Fund for Education Agency (LPDP)

Datasets

Standard Dataset

EMT Simulation results of Synchronized Lissajous D, Q, and synchronized Lissajous dv vs di (Classification Purposes)