Short-Term Load Forecasting (STLF) with ADMS Integration

This repository contains an end-to-end pipeline for Short-Term Load Forecasting (STLF) using advanced deep learning techniques, combined with an Automated Distribution Management System (ADMS)-based optimization module. It is designed to forecast regional electrical load and optimize load shedding/restoration decisions to maintain grid balance.

Key features include:

• Advanced Forecasting: Leveraging neural networks (CNN/LSTM/attention architecture) for high-accuracy load prediction at granular (5-minute) resolution.

• Automated Grid Management: Simulates real-world utility operations by optimizing load shedding and restoration with priorities, cooldown logic, and demand/supply balancing using Integer Linear Programming (ILP).

• Transparent Workflow: Every step, from raw data preprocessing to ADMS optimization, is modular, clearly documented, and fully reproducible via provided Jupyter notebooks and open dataset structure.

• Real & Synthetic Data Friendly: Out-of-the-box compatibility with both real and simulated generation/demand traces, supporting rapid prototyping and operational benchmarking.

• Visualization & Logging: Generates intuitive visualizations and action logs for model diagnostics, operational intervention review, and reporting.

Project Structure

STLF_project/
├── data/
│   ├── SEM.csv                  # Raw electrical load data
│   ├── demand_5min.csv          # 5-minute resampled load data (output)
│   ├── features_5min.csv        # Engineered feature dataset
│   ├── demand_forecast.csv      # Forecasted load data output
│   ├── generation_forecast.csv  # Generation forecast input (optional)
│   └── ...                     # Other data supporting files
├── models/                      # Saved models and scalers
├── plots/                       # Visualizations output
├── logs/                        # ADMS log outputs
└── stlf_notebook.ipynb          # Main Jupyter notebook with full workflow

Setup Instructions

Environment

• Python 3.7 or higher recommended
• Install dependencies via pip:

pip install tensorflow scikit-learn pandas numpy matplotlib seaborn holidays requests pulp joblib

Alternatively, run this command in your notebook or terminal.
You can also install all dependencies at once by running:

pip install -r requirements.txt

(See section Requirements File below)

Data Preparation

• Place your raw load data as SEM.csv inside the data/ folder.
• Optionally provide generation forecasts as generation_forecast.csv in the same folder to run realistic ADMS simulations.
• The pipeline automatically processes and creates the required files (demand_5min.csv, features_5min.csv, etc.).

Usage

1. Mount your Google Drive or set your base_path accordingly in the notebook.
2. Run the notebook stlf_notebook.ipynb sequentially to:
   • Load and preprocess data
   • Engineer features
   • Define, train, and evaluate the forecasting model
   • Run ADMS optimization for load shedding/restoration decisions
   • Generate output plots and logs
3. Inspect results and logs inside the plots/ and logs/ directories.

Evaluation Metrics and Their Interpretation

Metric	Purpose	Ideal Value / Meaning
MAE	Mean Absolute Error: average absolute deviation (in MW)	Lower is better; indicates average forecasting error
MSE	Mean Squared Error: squares errors to penalize large errors	Lower is better; sensitive to large deviations
RMSE	Root MSE: square root of MSE in MW (same units as target)	Lower is better; easily interpretable as MW error scale
MAPE	Mean Absolute Percentage Error: percentage error relative to true load	Lower is better; 0.9 is high efficiency

Replace values above with those from your latest run's output.

Customization

• Adjust model architecture, training parameters, and sequence length inside the notebook.
• Customize load profiles, priorities, and the ADMS operational parameters like buffer and cooldown periods to match your system’s requirements.
• Replace synthetic generation forecasts with real/forecasted generation data for realistic operation.

Requirements File

The requirements.txt file included in the repository contains exact package dependencies to reproduce the environment easily.
To install all required packages, run:

pip install -r requirements.txt

Contents of requirements.txt:

tensorflow>=2.6.0
scikit-learn>=1.0
pandas>=1.3.0
numpy>=1.21
matplotlib>=3.4
seaborn>=0.11
holidays
requests
pulp
joblib

For local Jupyter Notebook environments, you may optionally install:

notebook
ipykernel

Quick Environment Setup

Clone the repository, then run:

pip install -r requirements.txt

If using Google Colab, copy-paste the pip install commands into a dedicated notebook cell before running the pipeline.