INDUSTRY FIELD GUIDE
Water utility data for testing AI on simulated distribution events
A simulated pressure drop can give a model a clear event to investigate.
NAICS 221310 · Water Supply and Irrigation Systems
A simulated pressure drop can give a model a clear event to investigate. EPA’s EPANET models water distribution hydraulics and water quality. It gives a utility a way to construct a fictitious training environment with known changes. EPA also publishes cybersecurity guidance for water systems. A public modeling tool establishes the method. It does not establish an offer to buy a utility’s records. EPA EPANET · EPA water sector cybersecurity
Build an event whose cause is known
Create a fictitious network. Introduce a demand change or simulated sensor dropout and export permitted time series. Include the event start, expected behavior, observed measurements and verified cause. Add a benign lookalike, such as a normal demand peak. A buyer can test whether an assistant identifies the anomaly, states what it cannot infer, and recommends an appropriate next check within the simulation.
Record units, time zones, sampling interval and missing values. Keep results tied to the version of the simulation. A separate network layout for evaluation can help assess whether the model generalizes.
Keep the sample outside operational systems
Exclude live network diagrams, control screens, plant credentials, precise asset locations and customer-level usage. Obtain authorization from the utility and security team. Use a synthetic or isolated environment for initial scoping, with no connection to operational equipment.
Describe the simulation and intended evaluation. Explain which fields are invented and which outcomes are independently verifiable. Describe a dataset
Original Fiund exercise
Read a detached simulation's missing-sample event
An original simulation packet uses abstract node aliases and demonstration values. It is not a real network and has no operational connection or equipment instructions.
| Record | Source | What it records |
|---|---|---|
| SIM-V1 | Practice run definition | Run V1 states that samples at 09:00 and 09:05 are expected for abstract node N. |
| EVENT-S1 | Simulation event log | For V1 only, the 09:05 measurement is deliberately suppressed as a synthetic dropout. |
| TRACE-S2 | Exported practice trace | N has a value of zero at 09:00 and no observation at 09:05. |
| TRACE-S3 | Different run | Run V2 has a low recorded value at 09:05 under its own definition. |
Question for the model
What does V1 establish about the zero and the absent sample? Can V2's value fill the gap?
Show the reference answer and checks
V1's 09:00 zero is a recorded demonstration value. Its 09:05 observation is absent, and S1 documents the synthetic suppression for that same run. V2's value belongs to a different run and cannot fill V1's gap. The event log explains the invented dropout only within V1; it supplies no real utility diagnosis or operational action.
- Converting the absent observation into a measured zero.
- Combining traces from different simulation versions without an authorized mapping.
- Applying a synthetic event explanation to a real system.
Fields to preserve
- Run version
- Keep event log and trace in the same simulation.
- Sampling expectation
- Document which intervals should contain observations.
- Missingness and value
- Retain presence separately from the numeric reading.
Test another case
Create a new run with a recorded low value and no suppression event. The model must stop calling every low or absent reading the same event.
Use a wholly fictitious or authorized isolated environment. Exclude real network maps, controls, customer usage, access credentials and asset locations.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.