Investigation of Linear Genetic Programming for Dynamic Job Shop Scheduling

Abstract

Using genetic programming-based hyper-heuristic methods to automatically design dispatching rules has become one of the most effective methods to solve dynamic job shop scheduling. However, most genetic programming-based hyper-heuristic methods are developed based on tree-like structures. On the other hand, linear genetic programming variants, whose individuals are designed in a linear fashion, also have been successfully applied to some classification and symbolic regression problems and achieved promising results. But the studies of linear genetic programming as a hyper-heuristic for evolving dispatching rules for job shop scheduling are still in the infancy. To apply linear genetic programming to dynamic job shop scheduling (DJSS), this paper makes a comprehensive investigation on the design issues of linear genetic programming (e.g., the number of registers and the variation step size) and validates that linear genetic programming has a competitive performance with standard genetic programming and can produce compact dispatching rules.