cover

⚡ EcoGrid AI: Intelligent Grid Dispatch & Carbon Reduction

A HackAZ 2026 Submission TRACK 01: AI for Environmental Sustainability

🎯 Project Goal

EcoGrid AI is a scalable, intelligent optimization pipeline designed to minimize carbon footprints while maintaining grid stability across diverse geographical regions.

The system moves beyond localized solutions by providing a framework that ingest real-time environmental factors—specifically $Temperature$ and Global Horizontal Irradiance ($GHI$)—to calculate the most effective power generation mix for any regional demand. By mathematically maximizing renewable energy usage while adhering to strict $CO_2$ caps, EcoGrid AI offers a blueprint for transition-ready energy management worldwide.

🧠 The 3-Stage AI Architecture

EcoGrid AI doesn't just predict; it decides and communicates. Our architecture chains three distinct branches of AI:

Predictive AI (Forecasting): * Model: Multi-Layer Perceptron (MLP) via scikit-learn.
- Function: Ingests $Temp$ and $GHI$ (Global Horizontal Irradiance) to forecast hourly demand (MW) and the maximum potential of renewable sources.
Prescriptive AI (Optimization): * Model: Linear Programming via PuLP.
- Function: Mathematically guarantees the lowest-cost dispatch across Coal, Gas, Solar, Wind, and Hydro that meets demand while staying below the CO₂ limit.
Generative AI (Reporting): * Model: Meta LLaMA 3 (8B Instruct) via Hugging Face.
- Function: Ingests raw numerical outputs and generates a "Daily Energy Dispatch Report"—providing actionable advice for grid operators in plain language.

📂 Repository Structure

.
├── app.py                # Flask Backend: Houses the MLP models, PuLP logic, and LLM pipeline
├── cleandata.py          # Data Engineering: Merges EIA grid data with NSRDB weather metrics
├── index.html            # Frontend: Interactive UI with Chart.js and particle animations
├── data/
│   ├── NSRDB_Solar.csv   # Raw solar & weather data (Tucson, AZ)
│   ├── EIA930_BALANCE... # Raw grid demand and fuel mix data
│   └── FINAL_ML_DATASET.csv # The cleaned, feature-engineered training set
└── README.md             # Project documentation

🛠️ Tech Stack

Frontend: HTML5, CSS3 (Glassmorphism), JavaScript, Chart.js, Marked.js.
Backend: Python, Flask, Flask-CORS.
Machine Learning: Scikit-Learn (MLPRegressor), Pandas, NumPy.
Operations Research: PuLP (CBC Solver).
Generative AI: PyTorch, Hugging Face Transformers (LLaMA 3 8B Instruct).

🚀 Getting Started

1. Environment Setup

We recommend using a GPU-enabled environment (CUDA) for local LLM inference.

# Clone the repo
git clone https://github.com/your-repo/ecogrid-ai.git
cd ecogrid-ai

# Install dependencies
pip install -r requirements.txt

2. Data Preparation

Run the cleaning script to sync weather patterns with grid demand:

python cleandata.py

3. Launch the Engine

Start the Flask server (this will load the LLaMA model into VRAM):

python app.py

The server runs on http://127.0.0.1:5005 by default.

4. Open the Dashboard

Simply open index.html in any modern browser. Enter your weather parameters and CO₂ target to see the AI in action.

📊 Data Sources

Our models are grounded in real-world data specifically for the Tucson, Arizona region:

EIA (Energy Information Administration): Hourly grid balance and fuel generation.
EPA (eGRID): Carbon intensity factors for fossil fuel plants.
NREL (National Renewable Energy Laboratory): Solar irradiance and local meteorological data.