Production scheduling with genetic algorithms and simulation

A real-world application which develops daily production plans for a large manufacturing company is presented. It is a hybrid system, which combines a genetic algorithm with simulation. Because of the time constraints involved when generating daily schedules, a number of modifications to the standard genetic algorithm were required. A real-valued chromosome representation stored in a hierarchical, dynamic data structure is proposed. Steady-state, rank-based selection, a two-point order crossover and a simple, order-based mutation were implemented. An adaptive feedback controller was introduced to vary the mutation rate as a function of population convergence. Integration of a tabu list minimizes time wasted reevaluating known solutions. A rank-based fitness function is proposed to handle multiple, competing objectives.