Genetic Algorithm Parameter Control: Application to Scheduling with Sequence-Dependent Setups

Abstract

Genetic algorithms, and other forms of evolutionary computation, are controlled by numerous parameters, such as crossover and mutation rates, population size, among others depending upon the specific form of evolutionary computation as well as which operators are employed. Setting the values for these parameters is no simple task. In this paper, we develop a genetic algorithm with adaptive control parameters for an NP-Hard scheduling problem known as weighted tardiness scheduling with sequence-dependent setups. Our genetic algorithm uses the permutation representation along with the non-wrapping order crossover and insertion mutation operators. We encode the control parameters within the members of the population and evolve these during search using Gaussian mutation. We demonstrate this approach out-performs a manually tuned genetic algorithm for the problem, and that it converges upon effective parameter values very early in the run.

Variable Annealing Length and Parallelism in Simulated Annealing

“…Therefore, metaheuristics are a more practical approach. A variety of algorithms have been proposed for the problem, such as dynamic programming (Tanaka and Araki 2013), neighborhood search (Liao, Tsou, and Huang 2012), iterated local search (Xu, Lü, and Cheng 2014), value-biased stochastic sampling (Cicirello and Smith 2005), ACO (Liao and Juan 2007), GA (Cicirello 2015;2006), SA (Cicirello 2007), etc. We preprocess the instances transforming them as suggested by others (Tanaka and Araki 2013;Cicirello 2015).…”

“…To minimize the impact of setup times, we increase the processing time of each job, j k , by its minimum setup time, and reduce all setup times accordingly (Cicirello 2015):…”

Variable Annealing Length and Parallelism in Simulated Annealing

“…Therefore, metaheuristics are a more practical approach. A variety of algorithms have been proposed for the problem, such as dynamic programming (Tanaka and Araki 2013), neighborhood search (Liao, Tsou, and Huang 2012), iterated local search (Xu, Lü, and Cheng 2014), value-biased stochastic sampling (Cicirello and Smith 2005), ACO (Liao and Juan 2007), GA (Cicirello 2015;Cicirello 2006), SA (Cicirello 2007), etc. We preprocess the instances transforming them as suggested by others (Tanaka and Araki 2013; Cicirello 2015).…”

“…This operator removes a random element, and reinserts it at a different random position. Several existing metaheuristics for this scheduling problem use this operator, and it has been shown effective for permutation optimization problems characterized by asymmetric edges and general position within the permutation (Cicirello 2016). This problem has both: sequence-dependent setups (asymmetric edges), and due dates influence general position within permutation.…”