Reinforcement learning environments for process control

Quick start ⚡

Setup a CSTR environment with a setpoint change

import pcgym

# Simulation variables
nsteps = 100
T = 25

# Setpoint
SP = {'Ca': [0.85 for i in range(int(nsteps/2))] + [0.9 for i in range(int(nsteps/2))]} 

# Action and observation Space
action_space = {'low': np.array([295]), 'high': np.array([302])}
observation_space = {'low': np.array([0.7,300,0.8]),'high': np.array([1,350,0.9])}

# Construct the environment parameter dictionary
env_params = {
    'N': nsteps, # Number of time steps
    'tsim':T, # Simulation Time
    'SP' :SP, 
    'o_space' : observation_space, 
    'a_space' : action_space, 
    'x0': np.array([0.8, 330, 0.8]), # Initial conditions [Ca, T, Ca_SP]
    'model': 'cstr_ode', # Select the model
}

# Create environment
env = pcgym.make_env(env_params)

# Reset the environment
obs, state = env.reset()

# Sample a random action
action = env.action_space.sample()

# Perform a step in the environment
obs, rew, done, term, info = env.step(action)

Documentation

You can read the full documentation here!

Installation ⏳

The latest pc-gym version can be installed from PyPI:

pip install pcgym

Examples

Example notebooks with training walkthroughs, implementing constraints, disturbances and the policy evaluation tool can be found here.

Implemented Process Control Environments 🎛️

Environment	Reference	Source	Documentation
CSTR	Hedengren, 2022	Source
First Order Sytem	N/A	Source
Multistage Extraction Column	Ingham et al, 2007 (pg 471)	Source
Nonsmooth Control	Lim,1969	Source

Citing pc-gym

If you use pc-gym in your research, please cite using the following

@software{pcgym2024,
  author = {Max Bloor and  Jose Neto and Ilya Sandoval and Max Mowbray and Akhil Ahmed and Mehmet Mercangoz and Calvin Tsay and Antonio Del Rio-Chanona},
  title = {{pc-gym}: Reinforcement Learning Environments for Process Control},
  url = {https://github.com/MaximilianB2/pc-gym},
  version = {0.1.6},
  year = {2024},
}

Other Great Gyms 🔍

• ✨safe-control-gym
• ✨safety-gymnasium
• ✨gymnax