Learning Thermostat

This project implements a ML-based thermostat control system using Functional Mock-up Units (FMUs) and co-simulation. The system consists of multiple components including a physical plant model and a room environment from the INTO-CPS's case-study.

The key feature is a neural network-based controller (ThermostatML) that learns optimal heating control strategies by observing a baseline controller, then takes over operation once trained. The system uses model swapping to seamlessly transition from the baseline controller to the learned model during runtime.

Getting Started

To build and run the complete simulation pipeline:

make all

This will:

• Build all FMU components
• Run the simulation
• Generate performance graphs and reports

The trained model weights are saved to /var/tmp/learning_thermostat/thermostat_nn_model.pt and can be reused for inference in subsequent runs.