Capital renewal optimisation for large-scale infrastructure networks: genetic algorithms versus advanced mathematical tools

Abstract

Civil infrastructure assets require continuous renewal (repair, rehabilitation or replacement) actions to modernise the inventory and sustain its operability. Allocating limited renewal funds among numerous asset components, however, represents a complex optimisation problem. Earlier efforts using genetic algorithms (GAs) could optimise small size problems yet exhibiting steep degradation in solution quality as problem size increases. Even by applying sophisticated mechanisms such as ‘segmentation’ to improve the performance of GAs, large processing time hinders the practicality of the algorithm for large-scale problems. This article, therefore, aims at improving both processing speed and solution quality for very large-scale problems (up to 50,000 assets). The article develops optimisation models using an advanced modelling tool (GAMS/CPLEX), and compares its results with GAs on three different model formulations. Both approaches proved to be beneficial, yet the advanced mathematical approach showed superior performance.