ANE-MIST: Attributed Network Embedding With Minimum Spanning Tree for Optimizing Water Pipe Network Layouts

Abstract

The development of large-scale network layouts not only requires speed but also a model capable of accommodating all critical elements of the network while preserving the properties of those elements. Existing heuristic methods lack scalability, and static network embedding techniques fail to capture evolving topologies. This paper proposes ANE-MiST, a novel framework integrating Attributed Network Embedding with Minimum Spanning Tree (MiST) optimization to address these challenges. By leveraging graph machine learning, our model optimizes the design speed and pipe usage for constructing network layouts, where the pipes cover all nodes based on node properties such as geographical coordinate, elevation, and demand volume. Through the concept of network representation learning, the model can capture the dynamics of changes in the water pipeline network layout, whether due to obstacles or sudden destruction such as natural disasters. The model can reconstruct layout without starting from scratch when changes occur in the network. This approach offers a promising solution for scalable water infrastructure design and paves the way for future integration with hydraulic simulation and hybrid graph learning models.