Neural Network-Informed Optimal Water Flow Problem: Modeling, Algorithm, and Benchmarking

This dataset accompanies the article “Neural Network-Informed Optimal Water Flow Problem: Modeling, Algorithm, and Benchmarking” and is intended to support full reproducibility of the numerical experiments. It provides all data needed to (i) construct the two benchmark water distribution networks (NET1 and Anytown), (ii) formulate the associated Optimal Water Flow (OWF) problems, and (iii) train the ICNN and IC2N surrogate models used to approximate pipe friction losses and pump energy consumption. The repository is organized so that researchers can directly reuse the network inputs and training datasets to implement, validate, and extend the proposed neural network-informed optimization framework.

The repository is organized into two main folders, NET1 and Anytown, each corresponding to one case study. Within each of these folders, the Water_network_input subdirectory contains all data needed to build the water network and formulate the OWF problem (including topology, hydraulic parameters, demand profiles, and electricity price profiles), while the Training_data subdirectory contains the supervised learning datasets used to train the neural networks: ICNN for pipe friction losses and IC2N for pump energy consumption.

本数据集配套于论文《神经网络辅助最优水流问题：建模、算法与基准测试》，旨在支持数值实验的完整复现。本数据集提供了完成以下三项任务所需的全部数据：(i) 构建两个基准配水网络（NET1与Anytown）；(ii) 建立对应的最优水流（Optimal Water Flow, OWF）问题模型；(iii) 训练用于近似管道摩擦损失与泵组能耗的ICNN与IC2N代理模型。本仓库的架构设计便于研究人员直接复用网络输入与训练数据集，以实现、验证并拓展所提出的神经网络辅助优化框架。

本仓库分为两个主文件夹，分别为NET1与Anytown，各自对应一个案例研究。在每个文件夹内，Water_network_input子目录包含构建配水网络与建立OWF问题所需的全部数据（涵盖拓扑结构、水力参数、需量曲线与电价曲线）；而Training_data子目录则包含用于训练神经网络的监督学习数据集：其中ICNN用于管道摩擦损失的拟合，IC2N用于泵组能耗的拟合。