Researches On Multiobjective Traveling Salesman Problems Based On Hybrid Multiobjective Evolutionary Algorithm

In logistic and supply chain system,a lot of application problems can often translate into traveling salesman problem(TSP),which is well known as an important OR problem.However,it is noticeable that many real-world optimization problems always have to involve multiple objectives functions.These multiobjective TSPs have gained more and more concerns from a lot of researchers in the community of OR recently.Therefore,this thesis investigates and studies the applications of a hybrid multiobjective evolutionary algorithm(EA)in solving a general multiobjective TSP and a special multiobjective TSP,which can be termed as pollution-routing problem(PRP),based on the new findings in the field of evolutionary multiobjective optimization.A series of simulation experiments on the test instances from TSPLIB are carried out to examine the validity of the proposed algorithm in addressing multiobjective TPPs through making the comparisons with several state-of-the-art multiobjective evolutionary algorithms.The detailed research works can be described as follows.1.Design of a hybrid multiobjective evolutionary algorithm based on ant colony optimization and differential evolution.When solving a multiobjective optimization problem,the purpose of EA is to achieve a set of Pareto optimal solutions with a well uniform distribution.This means that two different search capacities should be involved during the running course of a multiojbective EA,that is,one is the exploitation search for the better non-dominated solutions and the other is the exploration search for the better evenly-distributed set of non-dominated solutions.Here,a two-phase search strategy is used in our proposed hybrid multiobjective EA.In detail,the operators of ant colony optimization are employed to make an exploitation search for a set of Pareto optimal solutions based on a decomposition mechanism,while the operators of differential evolution are employed to make an exploration search to obtain a better Pareto front.2.Experiment of algorithmic comparison in the general multi objective TSP.The test instances are firstly generated from the benchmark instances in TSPLIB,and then a series of simulation experiments are carried out in order to examine the performance of the proposed algorithm through a comparison with several state-of-the-art multiobjective EAs.Experimental results show that our proposed hybrid multiobjective EA outperforms the peer algorithms on the most test instances.3.Experiment of algorithmic comparison in the multiobjective PRP.A mathematical model with minimizing the cost of carbon emission and minimizing the cost of distribution is used to represent the multiobjective PRP.Based on the experiments on multiobjective simulation test instances,our proposed hybrid multiobjective EA always performs better than the peer algorithms on all test instances.