swmmEnv 0.1.0

Multi-agent reinforcement learning environment for SWMM stormwater simulation

SWMMEnv

Multi-agent reinforcement learning environment for SWMM (Storm Water Management Model) simulation.

Overview

SWMMEnv provides a PettingZoo-compatible interface for training multi-agent reinforcement learning (MARL) algorithms on stormwater management simulations. It integrates:

• PySWMM: SWMM simulation engine
• PettingZoo: Multi-agent RL environment interface
• MARLlib: Training framework (MAPPO, QMIX, etc.)

Features

• Read standard SWMM .inp files and rainfall .dat files
• Control pump stations, gates, and weirs
• Global reward for coupled stormwater systems
• Config-driven design for different SWMM models
• Time synchronization between RL steps and SWMM simulation steps
• State normalization for stable training

Installation

pip install -e .

Quick Start

from swmmEnv import SWMMParallelEnv, load_config

# Load configuration
config = load_config("config/example.yaml")

# Create environment
env = SWMMParallelEnv(config)

# Reset and get initial observations
observations, info = env.reset()

# Take a step
actions = {"pump_1": 0.8, "gate_1": 0.5}
observations, rewards, terminations, truncations, infos = env.step(actions)

# Close environment
env.close()

Configuration

See config/default_config.yaml for configuration structure.

Project Structure

swmmEnv/
├── swmmEnv/
│   ├── sim/           # Simulation modules (engine, time_sync, normalizer, mapping)
│   ├── envs/          # RL environments (core MDP + PettingZoo wrapper)
│   ├── reward/        # Reward functions
│   └── config/        # Configuration system
├── tests/             # Unit tests
├── examples/          # Example scripts
└── data/              # Sample SWMM models

License

MIT License