EnerGenius

Adaptive fan and lighting control using occupancy, environmental sensing and occupant preferences.

EnerGenius explores how a low-cost room controller can adapt its behaviour as conditions and user preferences change. It combines a logged-data ventilation prior, conservative offline Q-learning, lighting feedback and a residual online-learning layer.

Status: Ongoing development. Hardware integration and closed-loop evaluation remain in progress. The system controls fan and LED demand; HVAC is outside its scope.

Portfolio · Technical reference · Hardware handoff

What is implemented

• A room-observation model covering occupancy, temperature, humidity, CO2, ambient light and previous actions.
• A ventilation warm start from logged VAV positions, combined with supported-action offline Q-values and explicit safety logic.
• Closed-loop lighting feedback using measured lux and occupant targets.
• Preference feedback and an online Q-learning session that requires genuine post-action observations.
• An ESP32 serial adapter, policy persistence, evaluation scripts and unit checks.

The hardware demo currently runs inference only. Persistent deployment learning still needs action/transition logging and a versioned deployment workflow. Logged VAV positions provide a behaviour prior, not a measured calibration between fan PWM and airflow.

Architecture

Occupancy + environment + previous actions
                    |
    Ventilation prior + offline Q + lux feedback
                    |
            Fan PWM + LED PWM
                    |
       New observations + preference feedback
                    |
         Online-learning session

Component	Entry point
Data preparation and whole-day splits	prepare_dataset.py, room_control/dataset.py
Control policies and safety behaviour	room_control/controllers.py
Offline learning	room_control/offline_rl.py
Online transition learning	room_control/session.py
Hardware telemetry and commands	room_control/serial_hardware.py
Held-out audit and plots	evaluate.py, demo.py

Run locally

From a fresh checkout, create a Python environment and run:

python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
python prepare_dataset.py
python main.py
python evaluate.py
python demo.py

Run the existing checks with:

python -m unittest discover -s tests -v

See the technical reference for parameter overrides, dataset details, output files and hardware assumptions.

Data and evaluation

The controller uses a public room-occupancy and environmental-quality dataset: dataset DOI, data paper.

Offline reports are retained in reports/real_room/latest/. They describe logged-data checks and engineering estimates. Real-world comfort and energy performance remain under evaluation; provisional figures are documented in the technical reference rather than presented as deployment results.