Guided Genetic Algorithm

Maintain a population and a shared set of nonnegative feature-penalty counters, initially zero. Define each feature by an exhibited variable assignment or group of assignments and an application-specific undesirability cost. When the best current population member’s fitness remains unchanged for a configured number of generations, increment all maximal-utility present-feature counters, with utility cost/(1+counter). Use the augmented objective f+lambda*sum(counter*indicator) for the minimizing formulation and recompute each chromosome’s template: a gene weight sums counters for its exhibited features that involve that gene. Thus weights depend on the chromosome under shared penalties; offspring templates are recalculated, not inherited. At a crossover position with positive total parental weight, choose parent1’s value with probability weight2/(weight1+weight2), favoring the lighter gene. For mutation with positive total template weight, sample a gene in proportion to its weight and choose an objective-best domain value, with random tie handling in the displayed operator. The manuscript decrements a changed gene’s weight only if positive and allows repeated gene selection. Population replacement, mutation/crossover rates, stagnation threshold and stopping are configurable; the application uses elitist parent-plus-offspring replacement. These statements describe the documented framework and positive-weight operators, not a complete executable specification for all boundaries.