--

Dataset Card for DiTEC-WDN

Dataset Summary

DiTEC-WDN Dataset consists of 36 Water Distribution Networks (WDNs). Each network has unique 1,000 scenarios with distinct characteristics. Scenario represents a timeseries of directed shared-topology graphs, referred to as states or snapshots. In terms of graph-ml, it can be seen as a spatiotemporal graph where nodes and edges are multivariate time series.

A node can represent a reservoir, junction, or tank, while an edge refers to a pipe, pump, or valve. Notably, this dataset specifies concrete valve types, including PSV, PBV, FCV, TCV, and GPV. In addition, pump has two distinct types: head pump and power pump.
Each component has its own parameters, as listed in the below table:

Component	Parameter	Type	Unit
Head pump, Power pump, Pipe, PRV, PSV, FCV, TCV	Initial Status	Static (Category)	-
Head pump, Power pump	Base speed	Static (Float)	-
Head pump, Power pump	Efficiency X	Curve	SIFU¹
Head pump, Power pump	Efficiency Y	Curve	%
Head pump	Pump curve X	Curve	SIFU
Head pump	Pump curve Y	Curve	m
Head pump	Energy pattern	Pattern	kW-hours
Power pump	Power	Static (Float)	kW
Pipe	Diameter	Static (Float)	m
Pipe	Minor loss	Static (Float)	-
Pipe	Roughness	Static (Float)	mm (DW²) / - (Otherwise)
Pipe	Length	Static (Float)	m
PRV	Initial Setting	Static (Float)	m
PSV	Initial Setting	Static (Float)	m
FCV	Initial Setting	Static (Float)	SIFU
TCV	Initial Setting	Static (Float)	-
Tank	Elevation	Static (Float)	m
Tank	Diameter	Static (Float)	m
Tank	Initial level	Static (Float)	m
Tank	Minimum volume	Static (Float)	m³
Junction	Input demand	Pattern	SIFU
Junction	Elevation	Static (Float)	m
Reservoir	Base head	Static (Float)	m
Reservoir	Head pattern	Pattern	m

¹ SIFU stands for SI Flow Units including LPS, LPM, MLD, CMH, and CMD.
² DW refers to Darcy-Weisbach headloss equation.

Parameter-oriented dataset

For general usage, user may want to download a single parameter of a specific component in a concrete network. This can be done in this example code:

from datasets import load_dataset
raw_dataset = load_dataset(path="rugds/ditec-wdn", data_files="EXN_2GB_24H/junction_elevation*")

The downloaded subset is pressure values of all junctions in network EXN. User should check out the file structure to observe available parameters in each WDN folder.

Component-based dataset

For those who download all parameters w.r.t. a generic component type (node or link), we recommend the list of 72 configs in this README.md. The naming format is <network_name>_<storage_size>_<duration>_<general_component>. Please ensure sufficient storage before processing the dataset. Let's see this example:

from datasets import load_dataset
raw_dataset = load_dataset(path="rugds/ditec-wdn", name="EXN_2GB_24H_node")

This will download parquet files containing nodal features of the network EXN. Note that this action merges all parameters regardless of their type, so user should pay attention to the shape.

PyTorch(PyG) dataset

We also support a data interface to provide a parameter-mergable and ready-to-use dataset for PyTorch users. The interface and examples can be found in our official GitHub.

Miscellaneous information

All metadata, such as profiler info, simulation and optimization configurations, and network topology, is included in every Parquet file for a given WDN. To access it, user should retrieve a target WDN folder, pick a (light-weight) parquet file, and read its metadata using pyarrow libary. We provide a snippet supporting this as follows:

import tempfile
from huggingface_hub import hf_hub_download
import pyarrow.parquet as pq
def get_metadata(parquet_path: str, path: str = "rugds/ditec-wdn", key: str = "attrs") -> dict:
    with tempfile.TemporaryDirectory(dir="data", ignore_cleanup_errors=True) as dirpath:
        parquet_path = hf_hub_download(repo_id=path, repo_type="dataset", filename=parquet_path, local_dir=dirpath)
        metadata = pq.read_metadata(parquet_path)
        binary_meta_dict = metadata.metadata
        bkey = key.encode()
        if bkey in binary_meta_dict:
            attrs = binary_meta_dict[bkey].decode()
            return attrs
        else:
            print("Error! Key not found!")
            return {}

Supported Tasks and Leaderboards

• graph-ml: The dataset can be used to train a model for graph-related tasks, including node-level, link-level, and graph-level regressions.
• time-series-forecasting: The dataset can be used to train a model for time series forecasting tasks, including multivariate state estimation and next state forecasting.

Success on both tasks is typically measured by achieving:
(1) low Mean Absolute Error (MAE)
(2) low Mean Absolute Percentage Error (MAPE)
(3) low (Root) Mean Squared Error (RMSE/MSE)
(4) high coefficient of determination (R**2)
(5) high Nash–Sutcliffe Efficiency (NSE)

Languages

en

Dataset Structure

Data Instances

[More Information Needed]

Data Fields

[More Information Needed]

Data Splits

[More Information Needed]

Dataset Creation

Curation Rationale

The dataset is designed to (1) encourage open scientific research in the fundamental field of water, (2) eliminate the risk of exposing sensitive data while addressing strict privacy concerns that hinder advancements in machine learning, and (3) fulfill the need for a large-scale water distribution network benchmark for study comparisons and scenario analysis.

Source Data

The source data consists of simulation metadata files (.INP) collected from public resources. However, we retain only topology and component names, discarding all hydraulic-related values (e.g., elevation, demand time series, diameter, etc). These discarded values are later re-synthesized by our generator. As a result, the dataset contains no real or sensitive information from the original metadata or real-world systems.

Initial Data Collection and Normalization

The list below briefly describes the original water distribution networks:

WDN	Junctions	Pipes	Reservoirs	Tanks	Pumps	Patterns
ky1 [Jolly2014]	856	985	1	2	1	2
ky2 [Jolly2014]	811	1125	1	3	1	3
ky3 [Jolly2014]	269	371	3	3	5	3
ky4 [Jolly2014]	959	1158	1	4	2	3
ky5 [Jolly2014]	420	505	4	3	9	3
ky6 [Jolly2014]	543	647	2	3	2	4
ky7 [Jolly2014]	481	604	1	3	1	4
ky8 [Jolly2014]	1325	1618	2	5	4	4
ky10 [Jolly2014]	920	1061	2	13	13	4
ky13 [Jolly2014]	778	944	2	5	4	3
ky14 [Jolly2014]	377	553	4	3	5	3
ky16 [Jolly2014]	791	915	3	4	7	3
ky18 [Jolly2014]	772	917	4	0	3	9
ky24_v [Jolly2014]	288	292	2	0	0	3
19 Pipe System [Wood1972]	12	21	2	0	0	3
Anytown [Walski1987]	19	41	3	0	1	1
new_york [Schaake1969]	19	42	1	0	0	4
Jilin [Bi2014]	27	34	1	0	0	1
hanoi [Fujiwara1990]	31	34	1	0	0	0
fossolo [Bragalli2008]	36	58	1	0	0	0
FOWM [Walski2005]	44	49	1	0	0	0
EPANET Net 3 [Clark1995]	92	119	2	3	2	5
FFCL-1 [Rossman1996]	111	126	0	1	0	3
Zhi Jiang [Zheng2011]	113	164	1	0	0	0
WA1 [Vasconcelos1997]	121	169	0	2	0	6
OBCL-1 [Vasconcelos1997]	262	289	1	0	1	5
modena [Bragalli2008]	268	317	4	0	0	0
NPCL-1 [Clark1994]	337	399	0	2	0	17
Marchi Rural [Marchi2014]	379	476	2	0	0	0
CTOWN [Ostfeld2012]	388	444	1	7	11	5
d-town [Marchi2014]	399	459	1	7	11	5
balerma [Reca2006]	443	454	4	0	0	0
L-TOWN [Vrachimis2020]	782	909	2	1	1	107
KL [Kang2012]	935	1274	1	0	0	0
Exnet [Farmani2004]	1891	2467	2	0	0	0
Large [Sitzenfrei2023]	3557	4021	1	0	0	0

Their INPs are available here. Please note that the files were employed exclusively for optimization, not included in the dataset, as each scenario corresponds to a distinct INP file theoretically.

Who are the source language producers?

The dataset was machine-generated. The process detail could be found in the co-existence paper (see Citation Information).

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

The known limitations have been discussed in the binding paper. Please see Citation Information.

Additional Information

Dataset Curators

This work is funded by the project DiTEC: Digital Twin for Evolutionary Changes in Water Networks (NWO 19454).

Licensing Information

CC BY 4.0

Citation Information

For the dataset usage, please cite this:

@article{truong2025dwd,
  author    = {Huy Truong and Andr{\'e}s Tello and Alexander Lazovik and Victoria Degeler},
  title     = {DiTEC-WDN: A Large-Scale Dataset of Hydraulic Scenarios across Multiple Water Distribution Networks},
  journal   = {Scientific Data},
  year      = {2025},
  volume    = {12},
  number    = {1},
  pages     = {1733},
  doi       = {10.1038/s41597-025-06026-0},
  url       = {https://doi.org/10.1038/s41597-025-06026-0},
  issn      = {2052-4463}
}

Contributions

We thank the Center for Information Technology of the University of Groningen for their support and for providing access to the Hábrók high performance computing cluster.