Weather Impacts and Outage Prediction Using Distribution Networks' Topology and Physical Features - Outage Model CLI

This project includes a Command Line Interface (CLI) for preprocessing weather and static data for model training, as well as training and validating a GATRNN model for predicting outages.

Authors

• Colin T. Le
• Kenneth McDonald
• Zhihua Qu

Setup

To get the CLI running, first install the package in editable mode:

pip install --editable .

Then, visit the PyTorch website and follow the instructions to install the correct PyTorch package for either CPU or CUDA.

CLI Commands

1. Preprocess Data

outage-model preprocess [OPTIONS]

Preprocess the weather and static data for model training.

Options:

• --node-static-features: One or more physical node features to be considered for modeling.
  Example: --node-static-feature elevation
• --edge-static-feature: One or more physical edge features to be considered for modeling.
  Example: --edge-static-feature length
• --data-folder PATH: Input relative path to the folder containing the model data.
• --weather-features: One or more weather features (see the list of possible weather events).
• --output PATH: Output path to save both the CSV file and the pickle file.

2. Train and Validate Model

outage-model train-validate [OPTIONS]

Train and validate the GATRNN model with the given parameters.

Options:

• --pkl-file PATH: Input path to the pickle file containing datasets.
• --epochs INT: Number of training epochs.
• --learning-rate FLOAT: Learning rate for the optimizer.
• --hidden-size INT: Hidden size for the model.
• --validation-scale FLOAT: Scale for validation data.
• --output-model-file PATH: Output path to save the trained model as a .pth file.

Example Commands

Below are example commands for two types of datasets: Real Weather and Synthetic Weather.

Before Running These Commands

1. Download the datasets from the links provided (not shown here).
2. Create a new folder called ./model_data in the root directory of this project.
3. Place the downloaded .hdf5 files in the ./model_data folder.

Real Weather Dataset

1. Preprocess Data

outage-model preprocess \
    --dataset-file model_data/datasets_real_weather.hdf5 \
    --edge-static-features vegetation \
    --edge-static-features length \
    --node-static-features elevation \
    --node-static-features vegetation \
    --node-static-features flood_zone_num \
    --node-static-features Depth \
    --weather-features wspd \
    --weather-features prcp \
    --output output_real

2. Train and Validate

outage-model train-validate \
    --pkl-file output_real/dataDict.pkl \
    --output-model-file output_real/gatrnn_model.pth \
    --epochs 1500 \
    --learning-rate 0.001 \
    --hidden-size 40 \
    --optimizer adam

3. Evaluate

outage-model evaluate-gatrnn \
    --pkl-file output_real/dataDict.pkl \
    --model-file output_real/gatrnn_model.pth

Synthetic Weather Dataset

1. Preprocess Dataset

outage-model preprocess \
    --dataset-file model_data/datasets_fake_weather.hdf5 \
    --edge-static-features vegetation \
    --edge-static-features length \
    --node-static-features elevation \
    --node-static-features vegetation \
    --node-static-features flood_zone_num \
    --node-static-features Depth \
    --weather-features wspd \
    --weather-features prcp \
    --output output_fake

2. Train and Validate

outage-model train-validate \
    --pkl-file output_fake/dataDict.pkl \
    --output-model-file output_fake/gatrnn_model.pth \
    --epochs 100 \
    --learning-rate 0.001 \
    --hidden-size 40 \
    --optimizer adam

3. Evaluate

outage-model evaluate-gatrnn \
    --pkl-file output_fake/dataDict.pkl \
    --model-file output_fake/gatrnn_model.pth