Abstract:Maintenance planning for railway networks is an exacting task, impacting numerous stakeholders with diverse objectives. While a multi-objective optimization model addresses the problem, it often results in numerous solutions on a Pareto front. Past researchers have introduced various Pareto pruning methods to autonomously create a shortlist of promising solutions. Nonetheless, solely relying on data-driven approaches to finalize solutions can lead to undesirable outcomes for the decision maker (DM). To strike a balance between a human-involved and a data-centric decision, this article proposes a two-step approach. In the first step, Latent Profile Analysis (LPA) is utilized to group similar solutions on the Pareto front together, and a representative solution from each group is presented to the DM. After the DM identifies a focus region, Data Envelopment Analysis (DEA), including CCR and BCC models, is employed in the second step to rank the solutions in a selected group by their relative efficiency. Applying the proposed approach to a real-world case study demonstrates its capability in the railway asset management context. Key advantages of LPA lie in its ability to provide the appropriate number of solutions in the pruned set and its flexibility in grouping solutions. DEA enhances decision-making by ranking solutions within the focus region. Ultimately, by integrating automated algorithms with human insight, the two-step approach successfully identifies an efficient solution and mitigates the risk of obtaining undesirable outcomes on the Pareto front.

A two-step post-optimality approach for a multi-objective railway maintenance planning problem