Solving the Traveling Salesperson Problem Using Frequency Fitness Assignment

Tianyu Liang,Zhize Wu,Jrg Laessig,Daan van den Berg,Thomas Weise

DOI: https://doi.org/10.1109/ssci51031.2022.10022296

2022-01-01

Abstract:Metaheuristic optimization is based on the idea that better solutions are preferable to worse ones. Frequency Fitness Assignment (FFA) is a module that can be plugged into most optimization algorithms. It replaces the objective values in the selection step with their encounter frequency during the search so far. The search effort is therefore distributed evenly over the whole range of the objective values. Recently, it was shown that using FFA can significantly improve the performance of algorithms on hard problems such as Trap and Jump functions and the NP-hard MaxSat problem. However, the objective functions of all of these problems have relatively small ranges. This work is the first to explore the impact of FFA on metaheuristics for solving Traveling Salesperson Problem (TSP) instances, whose objective values tend to cover a wider range. We plug FFA into the (1+1) EA to obtain the (1+1) FEA. We perform extensive experiments on 18 instances from TSPLIB using two different unary search operators. We find that the (1+1) FEA does not get stuck in local optima and can solve many more instances to optimality than the (1+1) EA. However, it tends to be slower in reaching good intermediate solutions. Its performance decreases with the problem scale and the number of different possible tour lengths.