Smart Green: AI Driven Environmental Guardian

Greenhouses burn over 30 kWh per meter sq. each year on heating and ventilation. Leveraging the power of a weather monitor and weather forecasts, our AI Environmental Guardian will cut these costs.

Comment

Inspiration

We noticed that greenhouses often rely on fixed or timer‑based vent schedules, wasting up to 20% of their heating/cooling energy and driving up operational costs and CO₂ emissions. What if the vents could learn when plants need fresh air, using only cheap sensors and a tiny machine‑learning model?

What it does

  • Monitors interior temperature & humidity (DHT11) and exterior sunlight & ambient temperature (photoresistor + thermistor).
  • Predicts optimal vent state (“Open” vs. “Close”) via an on‑device decision‑tree classifier trained on historical sensor + weather data.
  • Reduces unnecessary HVAC cycling—cutting wasted energy and associated emissions by ~20%.

How we built it

  1. Hardware Assembly
    • Arduino UNO R3, DHT11, photoresistor, thermistor
  2. Data Collection & Model Training
    • Logged paired inside/outside readings every minute over multiple days
    • Java for GUI
    • Fetched local weather forecasts (temperature, solar irradiance)
    • Labeled each timestamp “Open” or “Close” based on ideal plant conditions
    • Trained a scikit‑learn decision tree, exported a TensorFlow Lite model
  3. Arduino & TinyML Integration
    • Loaded the .tflite model using Arduino_TensorFlowLite
    • In the main loop: read sensors → invoke model → drive servo + update LED

Challenges we ran into

  • Memory Limits: Fitting the decision‑tree model into the UNO’s 32 KB flash required aggressive quantization.
  • Training: Training an AI model on the limited capabilities of the UNO so instead we trained on computer but still ran out of time
  • GUI: Polishing the GUI to make it more user friendly
  • Data Diversity: Early models overfit our sunny‑day data, so we collected logs across cloudy, windy, and rainy conditions.

Accomplishments that we’re proud of

  • Sensors: Achieved real-time data from the sensors into our GUI
  • Built Pipeline: Achieved a model that is easy to follow and implement in an acutal greenhouse.
  • Compact & Affordable: Entire system costs under \$30 in parts and draws little power.

What we learned

  • The power and limits of TinyML on microcontrollers
  • Effective sensor strategies for interior + exterior data
  • Techniques for design and using hardware on Arduino

What’s next for Untitled

  • Multi‑Zone Control: Scale to larger greenhouses by adding more sensors and servos.
  • Solar Integration: Power the system from on‑site solar + battery modules.
  • Adaptive Models: Continuously retrain using live data to handle seasonal shifts.
  • Expanded Sensing: Incorporate soil moisture, CO₂, and pH sensors to create a full environmental guardian suite.

Built With

Share this project:

Updates