Диссертации, представленные на защиту и подготовленные в НИУ ВШЭ

This PhD dissertation proposes two new metaheuristic frameworks that extend the Genetic Algorithm (GA) to address premature convergence and limited exploration in combinatorial optimization. The first, the Genetic Engineering Algorithm (GEA), is inspired by genetic engineering principles and introduces three biologically motivated operators, gene mining, purification, and insertion, which reuse elite genetic material to improve diversity while still guiding the population toward convergence. The second, the Genetic Speciation Algorithm with Interplay Among Species (GSAIS), takes a different approach, applying cluster-based speciation combined with predator-prey dynamics so that individuals are grouped into distinct behavioral roles that shape their fitness updates, search behavior, and evolutionary operations. Within-species operations and short-term memory mechanisms are used to preserve diversity further and prevent the population from stagnating. Both algorithms are evaluated on NP-hard optimization problems drawn from supply chain and healthcare applications. To be more specific the Generalized Quadratic Assignment Problem (GQAP) and Operating Room (OR) Planning and Scheduling Problem is used to study patient scheduling in healthcare systems. Standard continuous benchmark functions are also included to test general optimization performance beyond these two optimization problems. Both algorithms model evolution and speciation more closely than standard GA, and this closer alignment with natural processes is what allows them to overcome the classical limitations of premature convergence and diversity loss. The results on these NP-hard problems confirm that this more biologically grounded design translates into stronger and more reliable optimization performance.