platform
Water System Chlorination Optimization
This project focuses on evolving an agentic system that can reliably control the time-varying dynamics via chlorination injection in a water distribution system/network.
 |
 |
 |
|---|
Decision makers
- Local and national government officials
- Civil/Environmental Engineers
- Public and Federal Health Institutions
Objectives
To optimize the real-time chlorination levels in a water network to, therefore, minimize chlorine (Cl) bound violations, minimize amount of Cl injected, minimize infection risk, and maximize the fairness and smoothness of Cl injections at varying locations in the network.
Deliverables
- A thorough paper on arXiv is in progress. For now we do have a short report on arXiv
- A user interface is also in progress. It'll likely be game-like wherein the user is able to control the Cl injections given timestep-wise data against the evolved agent.
Data attributes
Context
- Pseudo, real-time Cl concentrations at varying points in the network
- Water flow measurements in certain pipes
Actions
Decision makers will control the amount of Cl injected at varying points in the network after assessing the context at each timestep.
Outcomes
Decision makers are evaluated on multiple objectives/outcomes:
- Cl Bound Violations (Minimize)
- Total Cl Injected (Minimize)
- Infection Risk (Minimize)
- Fairness of Cl Injections; i.e. the similarity in magnitudes of Cl injected at different sites (Maximize)
- Smoothness of Cl Injections; i.e. the similarity in magnitudes of Cl injected at the same site across consecutive timesteps (Maximize)
Data
The data was derived as part of the 1st AI for Drinking Water Chlorination Challenge at IJCAI 2025: GitHub repo
The simulation itself is generated by the EPyT-Control library
Code
The in-development code for this project is available here: Code
Needs
List of needs:
- Contacts with decision makers (how to weigh different objectives for use in real-world networks)
- UI/UX experts
- Compute
References
Discussion
(no discussion yet)