Home > Research > Publications & Outputs > Robust Network Capacity Expansion with Non-line...

Associated organisational unit

Electronic data

  • ATMOS-v2019-article

    Accepted author manuscript, 463 KB, PDF document

    Available under license: CC BY: Creative Commons Attribution 4.0 International License

Links

Text available via DOI:

View graph of relations

Robust Network Capacity Expansion with Non-linear Costs

Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSNConference contribution/Paper

Published
NullPointerException

Abstract

The network capacity expansion problem is a key network optimization problem practitioners regularly face. There is an uncertainty associated with the future traffic demand, which we address using a scenario-based robust optimization approach. In most literature on network design, the costs are assumed to be linear functions of the added capacity, which is not true in practice. To address this, two non-linear cost functions are investigated: (i) a linear cost with a fixed charge that is triggered if any arc capacity is modified, and (ii) its generalization to piecewise-linear costs. The resulting mixed-integer programming model is developed with the objective of minimizing the costs. Numerical experiments were carried out for networks taken from the SNDlib database. We show that networks of realistic sizes can be designed using non-linear cost functions on a standard computer in a practical amount of time within negligible suboptimality. Although solution times increase in comparison to a linear-cost or to a non-robust model, we find solutions to be beneficial in practice. We further illustrate that including additional scenarios follows the law of diminishing returns, indicating that little is gained by considering more than a handful of scenarios. Finally, we show that the results of a robust optimization model compare favourably to the traditional deterministic model optimized for the best-case, expected, or worst-case traffic demand, suggesting that it should be used whenever computationally feasible.