Core Concepts
The study explores the optimization of evolutionary solver parameters, including population size, mutation rate, and convergence rate, to enhance the efficiency of solving the NP-hard problem of minimizing total tardiness in single machine scheduling.
Abstract
The study investigates the optimization of evolutionary solver parameters to address the NP-hard problem of minimizing total tardiness in single machine scheduling. It examines various parameter combinations, including population sizes (100, 50, 25, 10), mutation rates (0.75, 0.075, 0.0075), and convergence rates (0.0001, 0.1), to identify an optimal generic set of parameters that can effectively solve this complex scheduling challenge.
The key findings are:
- Using a population size of 50, mutation rate of 0.075, and convergence rate of 0.0001 yielded the best solutions within a runtime of 1-3 seconds.
- Reducing the population size further to 25 or 10 resulted in the solver failing to find optimal solutions, even with adjustments to the mutation rate and convergence.
- The study demonstrates the capability of the evolutionary solver in rapidly converging on different total tardiness values for various 10-job problems, with optimal solutions achieved in all 20 test cases when run for up to 42 seconds.
- Future research will focus on extending the study to parallel machine scheduling scenarios with non-zero ready times and minimizing total weighted tardiness.
Stats
The processing times for the 10 jobs range from 10 to 20 units.
The due dates for the 10 jobs range from 50 to 150 units.
Quotes
"The impact of manufacturing scheduling on a company's operational performance, and consequently its sustainability, is significant, particularly when considering the correlation between tardiness and operational expenses."
"Evolutionary algorithms, a cornerstone of our approach, incorporate principles of natural evolution into the process of finding optimal solutions for Solver problems."