Multiyear Maintenance and Rehabilitation Optimization for Large-Scale Infrastructure Networks: An Enhanced Genetic Algorithm Approach

Abstract

Multiyear network maintenance and rehabilitation optimization is a key, longstanding challenge for infrastructure asset management. Although genetic algorithms (GAs) have been widely used as the default optimization tool, successes were limited to small-scale networks. As the network size increases, the performance of conventional GAs quickly deteriorates because the traditional crossover and mutation operations disrupt promising solution compositions and drastically reduce the likelihood of obtaining a feasible solution. To address this gap, this paper introduced an enhanced GA that pivots on two innovations: a new crossover technique that swaps annual plans as a block of genes; and a novel mutation technique that incorporates linear programming (LP) to solve annual plans with a randomly perturbed budget profile. Both operations preserved the integrity of individual annual plans throughout the evolutionary process and enhanced local search capabilities. The hybrid LP-GA was tested with two practical case studies, one with a small-scale sewer network flushing program, and the other involving 13,610 pavement segments. Both case studies showed that the proposed algorithm quickly converged with 100% feasible solutions to optimum or near-to-optimum solutions. Through this work, we offered a sophisticated algorithmic tool for infrastructure planning, setting a stage for further advances in the domain.