☀️ PV Power Output Predictor (ML System + Web App)

A production-ready machine learning system for predicting photovoltaic (PV) power output using temperature and irradiance.

This project combines:

• end-to-end ML pipeline design
• experiment tracking with MLflow
• modular architecture using src/ layout
• a deployable Flask web application
• modern dependency management with uv

---

🌍 Live Demo

👉 View App on Render

---

Features

• Predict PV power output from environmental inputs
• Experiment tracking and model versioning (MLflow)
• Clean project structure (src/, scripts/, config/)
• Web interface (Flask) with dark mode UI
• Reproducible environment (uv)

---

How It Works

1. Data is loaded from data/raw/
2. Features (temperature, irradiance) are scaled using StandardScaler
3. A neural network (MLPRegressor) is trained
4. Experiments are logged with MLflow
5. The trained model is saved in outputs/models/
6. Flask app loads the model and serves predictions

In brief:

Input (Temp, Irradiance)
        ↓
Feature Scaling
        ↓
MLPRegressor Model
        ↓
MLflow Tracking
        ↓
Saved Model (outputs/)
        ↓
Flask App → Prediction UI

---

Project Structure

pv-predictor/
├── app/                # Flask web app
├── config/             # YAML configs
├── data/raw/           # input dataset
├── mlruns/             # MLflow experiments (ignored in prod)
├── outputs/models/     # trained model + scaler
├── scripts/            # training pipeline
├── src/pv_predictor/   # core ML logic
├── notebooks/          # exploration
├── pyproject.toml      # dependencies (uv)
└── README.md

---

Setup (with uv)

uv sync

---

Train the Model

python scripts/02_train_model.py

---

Run MLflow UI

mlflow ui

Then open: http://localhost:5000

---

Run the Web App

python app/app.py

Visit: http://127.0.0.1:5001

---

Deployment

This app can be deployed on:

• Render (recommended for Flask apps)
• Streamlit (for dashboard-style apps)

Render Start Command:

gunicorn app.app:app --bind 0.0.0.0:$PORT

---

Tech Stack

• Python (Flask, scikit-learn, pandas, NumPy)
• MLflow (experiment tracking)
• uv (dependency management)
• HTML/CSS (custom UI with dark mode)

---

Future Improvements

• Add PV performance metrics (PR, efficiency)
• Integrate anomaly detection (UMAP, LOF)
• Forecasting (LSTM / sklearn)
• Build interactive dashboard (Streamlit)
• Load models directly from MLflow registry

---

Preview

---

📊 Sample Prediction

Input: Temperature = 25°C Irradiance = 800 W/m²

Output: Predicted Power ≈ 0.82 kW

---

🔄 Project Evolution

• v1 (Prototype): Simple Flask app with static model
• v2 (Current): Full ML system with:
  • modular pipeline (src/)
  • experiment tracking (MLflow)
  • reproducible environment (uv)
  • production-ready deployment

---

Author

Built as part of a machine learning + photovoltaic systems project.

---