Water network optimisation with AI & digital twins
Move beyond static “design only” models. We build operational digital twins of water networks and layer AI optimisation on top, so that pumps, valves and tanks run with lower pressure, lower energy and better service – every day, not just in a report.
What this means in practice: an online copy of your network running in the background, continuously learning from SCADA data and proposing better setpoints and control rules for real operation.
10–30% lower average pressure*
through AI-driven setpoints and valve strategies.
Energy savings at pump stations
by running the right pumps at the right time, not just “all on”.
Fewer burst incidents & complaints
via more stable transients and smoother operation.
*Typical range from international case studies – actual results depend on network configuration and operating policies.
From static model to living digital twin
You already invest in hydraulic models – EPANET, InfoWorks, WaterGEMS and similar tools. Typically they are used a few times a year for planning projects. The rest of the time, your network runs based on rule-of-thumb operating strategies.
We connect your existing model (or build a new one) to real operational data and turn it into a digital twin that can simulate “what if we change this control?” before you touch the real network.
Where AI comes in
On top of the digital twin, we apply reinforcement learning and optimisation algorithms – the same class of methods used in our peer-reviewed journal work – to learn control policies that:
- Respect minimum and maximum pressure limits
- Minimise energy cost and pressure surges
- Maintain storage levels and critical service constraints
The result is a “network autopilot” that can suggest better pump and valve settings for the current demand pattern and forecast.
How a typical optimisation project works
1. Connect your network
We start from your existing EPANET / InfoWorks model, or build a clean EPANET model from drawings and GIS. SCADA, PLC and energy-tariff data are mapped to the same network.
2. Build the digital twin
The model is wrapped in a Python-based simulation layer so that we can run thousands of scenarios automatically. Calibration is refined until the twin behaves like your real system.
3. Train the AI controller
Reinforcement learning agents explore different pump and valve strategies in the twin – never on the real network – and learn policies that minimise pressure and energy while keeping service constraints.
4. Deploy as decision support
The best policies are exposed through a web-based dashboard tailored to your network. Operators can see suggested setpoints for today’s conditions and compare them with the current strategy.
Built on open tools – no new licences
Our optimisation platform is deliberately built on open and widely available tools, so you are not locked into proprietary software or annual licence fees.
- EPANET 2.2 for digital twin.
- Python for automation, data processing and integration
- Open-source RL libraries (e.g. Stable-Baselines, etc.) for the learning engine
- Lightweight web dashboards built with modern front-end frameworks
- Deployable on your own servers, cloud, or kept as an offline planning tool
- All configuration scripts and models can be handed over to your in-house team
A custom web interface for each network
We do not force your network into a generic control screen. Each project includes a tailored web interface that reflects your own zones, pump stations and operating language.
- Map view with key tanks, pump stations and DMAs
- Daily recommended setpoints and operating schedules
- “What if?” sliders to test alternative operating strategies
- Simple indicators: expected energy use, average pressure and risk level
Your team sees one page per network – fast to load, easy to understand, and focused on decisions rather than raw data.
Example layout (conceptual):
- Left panel – network map with live/typical pressures
- Centre panel – today’s recommended strategy and constraints
- Right panel – comparison: “current vs AI” pressure and energy
This interface can be hosted as a secure internal page (e.g. behind VPN) or used offline during planning workshops.
Is this suitable for your utility?
A good fit if you…
- Operate one or more complex transmission/distribution networks with pump stations and storage
- Already use (or plan to use) hydraulic models for planning
- Have SCADA or logged data for key sites
- Want to reduce pressure/energy without compromising service
We start small
Most utilities start with a pilot on one representative network. We demonstrate benefits, fine-tune the workflow and then scale to additional systems when you are ready.
This approach keeps risk low, builds internal confidence and ensures the AI behaves in a way that matches your operational philosophy.