Multi-objective search inside Sim::OPT
The current OPTcue distribution includes an NSGA-II implementation for multi-objective evolutionary search. The algorithm works on the discrete Sim::OPT genotype: candidate designs are represented by the same level-based coordinates used elsewhere in the framework.
Within a block search, NSGA-II varies the active block while retaining the surrounding coordinates as context. This allows evolutionary multi-objective exploration to coexist with the broader Sim::OPT decomposition rather than requiring a separate representation of the problem.
Algorithm
The readable Perl implementation contains the expected NSGA-II components: dominance tests, non-dominated sorting and rank assignment, crowding distance, tournament selection, crossover, mutation and environmental selection over the parent-plus-offspring population.
The implementation is duplicate-objective aware and maintains an authoritative archive of evaluated designs from which the reported Pareto set is obtained.
Configuration
The main controls include population size, evolutionary step, crossover probability, mutation probability, number of mutations per child and the selected objective columns. Objective directions can independently specify minimization or maximization.
The current distribution supplies NSGA-II as the optional evolutionary search module. References in older OPTcue documentation to other experimental algorithms should not be read as a list of algorithms included in the present package.
Outputs
NSGA-II writes a final-population CSV and a Pareto CSV. The Pareto output is based on every evaluated design retained in the archive, rather than only the final evolutionary population.
When NSGA-II is embedded in a block-coordinate sequence, Sim::OPT still needs a scalar transition state for continuing to the next block. That transition state should not be confused with the complete Pareto archive.