An ANN-Embedded MILP approach for optimizing the Steady-State operation of gas pipeline networks
•An ANN-MILP approach is proposed for energy-efficient gas pipeline network operation.•The proposed approach approximates non-convex hydraulic-thermal constraints to linear forms via ReLU-activated ANN.•In a simple linear network, ANN-MILP achieves the same global optimal schemes as DP with 10x spee...
Uložené v:
| Vydané v: | Chemical engineering science Ročník 320; s. 122619 |
|---|---|
| Hlavní autori: | , , , |
| Médium: | Journal Article |
| Jazyk: | English |
| Vydavateľské údaje: |
Elsevier Ltd
15.01.2026
|
| Predmet: | |
| ISSN: | 0009-2509 |
| On-line prístup: | Získať plný text |
| Tagy: |
Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
|
| Shrnutí: | •An ANN-MILP approach is proposed for energy-efficient gas pipeline network operation.•The proposed approach approximates non-convex hydraulic-thermal constraints to linear forms via ReLU-activated ANN.•In a simple linear network, ANN-MILP achieves the same global optimal schemes as DP with 10x speedup.•24.2% energy saving in 110-node network via gas/electric-driven compressor flow coordination.
Optimization of natural gas pipeline network operations is crucial for enhancing energy efficiency and reducing carbon emissions. However, the complex hydraulic-thermal coupling constraints and mixed-integer decision variables in such models result in a large-scale mixed-integer nonlinear programming problem that is extremely hard to solve. To address this challenge, this paper proposes a novel artificial neural network-embedded mixed-integer linear programming (ANN-MILP) approach. Leveraging the piecewise linear property of ReLU activation functions, the approach approximates the non-convex hydraulic and thermal constraints of pipelines, as well as compressor station operational constraints, into MILP-compatible forms via artificial neural network surrogate models. A steady-state optimization model is then formulated to minimize energy consumption, subject to node flow balance, pressure and temperature bounds, and other operational constraints. In a simple linear network case, the ANN-MILP approach achieves a nearly identical operational scheme to that obtained by dynamic programming, while improving solution efficiency by nearly 10 times. In a real-world 110-node network, by optimally allocating the flow rates of gas turbine-driven and electric-driven compressors, the energy consumption is reduced by 24.2% compared with the actual industrial operation, whereas a state-of-the-art MINLP solver fails to converge within 1 h. Both case studies are validated using commercial simulation software, with maximum deviations in pressure and temperature remain below 5%. These results show that the ANN-MILP approach achieves both computational efficiency and engineering reliability, offering a scalable solution for energy saving and consumption reduction in complex energy systems. |
|---|---|
| ISSN: | 0009-2509 |
| DOI: | 10.1016/j.ces.2025.122619 |