Rukmi, Alvida Mustika (2026) Sistem Konstruksi Tata Letak Jaringan Saluran Pipa Air Berbasis Graph Embedding. Doctoral thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 7025221018-Doctoral.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
7025221018-Doctoral.pdf Restricted to Repository staff only Download (10MB) | Request a copy |
Abstract
Pada jaringan distribusi air, pendekatan umum yang digunakan untuk membangun tata letak jaringan masih didominasi oleh pendekatan optimasi seperti algoritma genetika, simulated annealing, maupun minimum spanning tree yang berbasis jarak geometris. Pendekatan-pendekatan tersebut bekerja langsung pada ruang pencarian yang sangat besar sehingga memerlukan waktu komputasi yang tinggi, terutama pada jaringan berskala besar dan dinamis, jika terjadi perubahan secara geografis maupun adanya obstacle di dalam jaringan.
Sebuah metode machine learning berbasis graph embedding mempunyai kemampuan dalam proses mereduksi graf berskala besar ke dalam dimensi yang rendah. Setiap node pada jaringan direpresentasikan berupa vektor ke dalam ruang embedding yang mampu mempertahankan hubungan topologi dan atribut node. Representasi tersebut kemudian dimanfaatkan untuk mengukur kemiripan antar-node.
Penelitian ini memanfaatkan graph embedding sebagai inti proses pembangunan tata letak jaringan pipa air. Penggunaan graph embedding mampu mengurangi kompleksitas ruang pencarian dalam proses konstruksi jaringan. Model mempelajari representasi pengelompokan node-node berdasarkan karakteristik struktural yang mirip dan membentuk kandidat koneksi jaringan yang lebih representatif dibandingkan penggunaan jarak geometris semata. Proses pembangunan tata letak jaringan hanya dilakukan pada pasangan node yang memiliki kedekatan tinggi di ruang embedding, tanpa harus menelusuri seluruh kemungkinan hubungan antar-node yang jumlahnya meningkat secara eksponensial terhadap ukuran jaringan.
Model merepresentasikan informasi atribut, seperti kebutuhan air, elevasi, geospasial koordinat node ke dalam proses pembelajaran embedding. Integrasi informasi tersebut mampu menghasilkan tata letak jaringan yang tidak hanya mempertahankan kedekatan topologi, tetapi juga mencerminkan kesamaan karakteristik antar-node. Didukung oleh Minimum Spanning Tree (MST), model ini menghasilkan tata letak yang lebih optimal dalam memenuhi kebutuhan sistem distribusi air.
Pada perubahan atribut node jaringan, model ANE-MIST berbasis Dynamic Attributed Graph embedding ini mampu merekronstuksi tata letak jaringan dan menunjukkan efisiensi dalam komputasi waktu. Konstruksi jaringan dengan ANE-MIST lebih optimal dalam penggunaan pipa dibandingkan Dijkstra+MST dan Node2Vec+MST. Perbandingan skor Mean Average Precision (MAP), Mean Reciprocal Rank (MRR), dan Area Under the Receiver Operating Characteristic Curve (AUC-ROC) ketiga model tersebut menunjukkan ANE-MIST mempunyai performa kerja yang lebih baik. Tata letak jaringan yang dibangun model, berhasil divalidasi oleh EPANET.
Melalui ANE-MIST sebagai mekanisme pembelajaran representasi untuk rekonstruksi tata letak jaringan pipa air, mampu mempertahankan informasi topologi, atribut node, dan dinamika perubahan jaringan. Model ini tidak hanya mengurangi kompleksitas ruang pencarian selama proses konstruksi jaringan, tetapi juga menghasilkan tata letak yang lebih efisien dan adaptif terhadap perubahan kondisi jaringan, sehingga berpotensi mendukung pengembangan sistem desain jaringan distribusi air yang cerdas.
====================================================================================================================================
In water distribution networks, conventional approaches for constructing network layouts are still predominantly based on optimization techniques, such as Genetic Algorithms, Simulated Annealing, and Minimum Spanning Tree (MST) using geometric distances. These approaches operate directly on an extremely large search space, resulting in high computational costs, particularly for large-scale and dynamic networks where geographical changes or obstacles alter the network topology.
Graph embedding, a machine learning technique, has the capability to transform largescale graphs into low-dimensional representations. Each node in the network is embedded as a vector in a low-dimensional embedding space that preserves both the topological relationships and node attributes. These learned representations are subsequently used to measure the similarity between nodes.
This study employs graph embedding as the core mechanism for water distribution network layout construction. By learning latent representations of nodes with similar structural characteristics, graph embedding effectively reduces the search space complexity during the network construction process. Instead of relying solely on geometric distances, the proposed model generates more representative candidate connections based on similarities in the embedding space. Consequently, the network layout is constructed only among node pairs with high embedding similarity, eliminating the need to evaluate all possible node connections, whose number increases exponentially with network size.
The proposed model incorporates node attributes, including water demand, elevation, and geospatial coordinates, into the embedding learning process. Integrating these heterogeneous attributes enables the learned representations to preserve not only topological proximity but also the similarity of node characteristics. Combined with the Minimum Spanning Tree (MST) algorithm, the proposed framework produces a network layout that is more optimal for satisfying the operational requirements of water distribution systems.
To accommodate changes in node attributes, the proposed Attributed Node Embedding with Minimum Spanning Tree (ANE-MIST) model is extended through Dynamic Attributed Graph Embedding, enabling adaptive reconstruction of network layouts while maintaining computational efficiency. Experimental results demonstrate that ANE-MIST achieves more efficient pipe utilization than both Dijkstra+MST and Node2Vec+MST. Comparative evaluations using Mean Average Precision (MAP), Mean Reciprocal Rank (MRR), and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC) show that ANE-MIST consistently outperforms the baseline methods. Furthermore, the network layouts generated by the proposed model are successfully validated using EPANET, confirming their hydraulic feasibility.
By employing ANE-MIST as a representation learning framework for water distribution network layout reconstruction, the proposed approach simultaneously preserves network topology, node attributes, and dynamic network changes within a unified embedding space. This approach not only reduces the search space complexity during network construction but also generates network layouts that are more efficient and adaptive to changing network conditions. Consequently, it provides a promising solution for the development of smart water distribution network design systems.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | layout jaringan, Dynamic Attributed Graph embedding, saluran pipa air, network layout, Dynamic Attributed Graph embedding, water pipeline |
| Subjects: | Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.76.E95 Expert systems Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55001-(S3) PhD Thesis (Comp Science) |
| Depositing User: | Alvida Mustika Rukmi |
| Date Deposited: | 04 Aug 2026 07:14 |
| Last Modified: | 04 Aug 2026 07:14 |
| URI: | http://repository.its.ac.id/id/eprint/142885 |
Actions (login required)
 |
View Item |