Research On Multi-objective Optimization Evolutionary Algorithm Based On Preference Information

While traditional MOEAs have shown excellent balance between convergence and diversity on a range of MOPs with two or three objectives in real applications,the decision makes are not interested in a unique set of solutions rather than the whole population on PF.In addition,Pareto-based MOEAs have some shortcomings in dealing with many-objective optimization problems because of insufficient selection pressure toward trade-off solutions.Due to above,it is crucial to incorporate DM preference information into MOEAs and seek a representative subset of Pareto optimal solutions with an increase in the number of objectives.This kind of MOEAs called preference-based MOEAs.According to the ways of the DM's elicit preference information to guide the population to converge toward the ROI,existing preference-based MOEAs could be classified into two different methods.The first one modifies the traditional Pareto dominance relation by introducing preference information into the dominance relation.The second strategy is based on the algorithms with decomposition,and use predefined reference vectors to guarantee distribution.For the first one,this paper proposes a new dominance relation,called Radominance,which can improve diversity among the Pareto-equivalent solutions and increase the selection pressure in evolutionary process.It has the ability to guide the population toward areas more responsive to the needs of the DM according to a reference point and preference angle.In addition,the two kind of algorithms have their own advantages for different problems.In real life problem,it is necessary to adopt the most suitable MOEA for the problems characteristics,yet the characteristics of problems are not known beforehand.We develop a new preference-based method that can incorporate preference information into any MOEA and handle preference-based MOPs.We convert the DM's preferences into a penalty function and revise solutions' fitness based on penalty values.With this method,the preference-based MOP is converted to ordinary MOP,and the preferred solutions are given more selection pressure and approximate to the PF.In the experiment,four state-of-the-art preference-based MOEAs are chosen as comparison algorithms,and the test problems suites are ZDT,DTLZ and WFG.The results show that the algorithm and strategy presented in this paper are highly competitive.