A Multi-objective Optimization Algorithm Based On Decision Preference

As a method for solving complex problems,the multi-objective optimization algorithms have been widely used in many practical fields such as the industry and commerce.The traditional multi-objective optimization algorithms,such as NSGA-II and MOEA/D,have achieved very good results in the distribution and convergence of solution sets.However,in real life and production activities,the decision makers will be inclined to different preferences for different optimization objectives according to their professional knowledge and business conditions.Therefore,among the optimal solution sets,only a part of the solutions that the decision makers are interested in.At the same time,as the market environment changes,the preferences of decision makers will also change.Therefore,we prefer to the preferred solutions that the decision maker is interested in rather than finding all the optimal solution sets.This strategy not only improves the convergence performance of the algorithm,but also reduces the time cost of the algorithm.Therefore,this thesis conducts an in-depth study on multi-objective optimization algorithm based on the decision preferences.Aiming at how to integrate the importance relationship among optimization objectives into multi-objective optimization algorithm,a preference three-way decomposition model is proposed in this thesis.The model integrates the weight values given by the decision makers into the fitness function to calculate the fitness values of individuals in the population.Then,the individuals are selected to enter the next iteration process according to the fitness values.At the same time,with the weight value of each objective as the standard,these objectives can be divided into three different parts,and then these three parts are optimized separately.Finally,the overall optimal set is obtained,which can reduce the scale and difficulty of the optimization problem.The experimental results show that when the decision maker gives different preference information,the proposed algorithm can adjust the search range of the population according to the preference information.In addition,the proposed algorithm is compared with the traditional NSGA-II algorithm on the DTLZ1 test of six targets.The results show that the proposed algorithm has better convergence.In view of the dynamic change of decision maker's preference,a multi-objective optimization model based on dynamic preference is proposed in this thesis.The model finds that no matter how the preferences of decision makers change,there are only four kinds of location relationships between the old and new preference regions: there is no overlap between the old and the new preference regions,there is overlap between the old and the new preference regions but they do not contain each other,the new preference regions contain the old preference regions,and the old preference regions contain the new preference regions.For these four cases,the proposed model provides different optimization strategies.It mainly solves the shortcoming that the traditional model can not respond quickly to the change of decision makers' preference.The experimental results on the two-dimensional and the three-dimensional test problem,prove that the proposed model has a great advantage in convergence compared with the traditional MOEA/D-PRE model.