Skip to content

Provided Examples List

The CosimGym repository comes with a series of pre-configured tests (scenarios) to demonstrate the transition from simple physics simulations to comprehensive multi-federation Reinforcement Learning workloads.

These are defined inside src/scenarios/.

Base Co-Simulation (Physics & Math)

  • simple_test (Case 0): A fundamental single-federation test connecting a spring-mass-damper system to an autonomous input signal generator. Excellent for debugging that your HELICS installation is functional.
  • simple_test_multifederations: Expands Case 0 to split the spring and the damper inputs across a hierarchical multi-broker network.
  • bui_hp_test_base (Case 1): A building thermal zone integrated with a Heatpump module and CSV weather drivers, regulated by a classical explicit PID controller.
  • pv_batt_test_base (Case 4): A micro-grid emulation consisting of Photovoltaic panels, a battery storage model, a static electrical load, and weather inputs controlled by a rule-based (RB) management system.

Reinforcement Learning Training

  • simple_DQN_test / simple_SACsb3_test: Implements discrete and continuous action algorithms over the simple spring system to demonstrate Gym wrapping bare mechanics.
  • bui_hp_DQN / bui_hp_SAC (Cases 2 & 3): Replaces the PID controller from Case 1 with stable-baselines3 agents learning optimal thermal setpoints based on ambient conditions. Includes "rolling reset" variations.
  • pv_batt_DQN / pv_batt_SAC (Cases 5 & 6): Replaces the Rule-Based manager from Case 4 with an algorithm learning deep policies to balance grid stability and storage degradation constraints.

To run any case, set the scenario name inside the entry-point script, then run it:

# src/test_script.py        (base co-simulation cases)
main('simple_test')

# src/test_script_rl.py     (RL training cases)
main('simple_DQN_test')

conda activate cosim_gym
python src/test_script.py        # or: python src/test_script_rl.py

There is no --scenario command-line flag; the scenario is chosen by the main('<name>') call.