Comparative Study of Two Genetic Algorithms for Steel Production Planning Under Different Order Backlog Circumstances

This chapter aims at comparing two genetic algorithms (GA) designed for steel production planning. While the first GA attempts to minimize production costs globally by scheduling multiple production turns in parallel, the second algorithm sequentially schedules production turns and, thus, optimizes costs locally. Although it is widely accepted by the academic community that the parallel approach beats the sequential approach, steel companies still appear to rely on local optimization schemes in practice (Özgür, et al. Comput. Ind. Eng. 106606:2021). We suspect that this scepticism towards the parallel approaches stems from two key drawbacks present in the pertinent literature: Previous articles (i) have not provided performance experiments that reveal by how much the parallel and sequential scheduling strategies actually differ and they have (ii) failed to explicate the production data utilized in their experiments. We address these issues through a comprehensive comparison of the two approaches under different production data circumstances—accompanied by a detailed description of the underlying synthetic backlog data generator. Our experiments will show that the parallel approach works better in general; however, it might be outperformed occasionally if the order backlog situation is particularly fragmented or uncertain.