Preference-based Evolutionary Multi-objective Optimization And Its Applications In Satellite Mission Planning

Multi-objective optimization problems(MOPs)exist widely in real life.Unlike the single objective optimization,the solution of the multi-objective optimization problem consists of a series of compromised solution called Pareto optimal solutions."Optimizing" and "decision making" are two key factors for solving multi-objective optimization problems.Evolutionary multi-objective optimization(EMO)focuses on "optimizing",aiming to use population-based mataheuristic algorithms to find all the Pareto optimal solutions in one run;Multi-criteria decision-making(MCDM)focuses on "decision",helping decision makers to find the most satisfactory Pareto optimal solution through preference modeling.Preference-based Multi-Objective Evolutionary Algorithms(PMOEAs)is an elaborate combination of evolutionary multi-objective optimization and multi-criteria decision making.Using preference information provided by the decision maker to guide the evolutionary process,PMOEAs can converge to the Pareto optimal solution that accords with the preference of decision makers.Nowadays,PMOEA is a research hotspot in the field of Evolutionary Computing(EC).In this thesis,the PMOEA is studied,focusing on the shortcomings of the current theoretical research,the main research contents are algorithm overview and ontology modeling,new algorithm design(for low-dimensional and high-dimensional optimization problems respectively),and algorithm application in the field of satellite mission planning.The specific works are as follows:(1)The overview and ontology modeling of PMOEAs.At present,a large number of PMOEAs have been proposed.Firstly,this thesis reviews the existing algorithms,analyzes the preference informayion used,preference integration,test problems,evaluation factors of each algorithm,and constructs a preference-based meta-heuristic algorithm ontology based on the review.Ontology modeling is an important tool to describe domain knowledge.It helps to understand concepts and the relationship between concepts in the domain through detailed annotations and visualization tools.The ontology can effectively analyze the relationship between two algorithms,in addition,it can also be used for information query,bibliometric analysis,algorithm automatic classification,and discovery of research directions.(2)Design of target-region-based multi-objective evolutionary algorithms for low dimensional optimization problems.Target region is a simple,flexible and easy-to-understand preference model,but the existing research is not perfect.Aiming at two-objective and three-objective optimization problems,multi-objective evolutionary algorithms(i.e.T-NSGA-II,T-SMS-EMOA and T-R2-EMOA)are designed.The algorithms can converge quickly to the target region provided by the decision maker and support multiple target regions simultaneously.The new algorithms support both the target area and the reference point.Experiments on benchmark problems show the effectiveness of the new algorithms,and the advancement of the proposed algorithms is proved through the comparison with other similar algorithms.(3)Design of target-region-based multi-objective evolutionary algorithms for high dimensional optimization problemsTwo many-objective evolutionary algorithms(i.e.T-MOEA/D and T-NSGA-III)are designed for high-dimensional optimization problems(the number of objective functions is more than or equal to four).The algorithms support both a-priori optimization and interactive optimization.The effectiveness of the algorithms is shown through benchmark problem results.The performance of the algorithms in a-prior optimization is compared,and the advantages of the algorithms in interactive optimization are shown.Compared with a-prior optimization,in interactive optimization the algorithms can quickly respond to the changes of preferences and help them find the most satisfactory solution.(4)The application of preference-based evolutionary multi-objective optimization in satellite mission planningSatellite mission planning is a complex,constrained optimization problem,which includes many optimization objectives,such as total observation profit,the balance of resource utilization,observation timeliness and so on.The traditional method of transforming multi-objective optimization into single-objective optimization by linear weighting has many limitations.Aiming at the problem of agile satellite earth observation mission planning,this thesis proposes a solution framework of preference-based multi-objective optimization.A suitable coding/decoding strategy and a crossover mutation operator are designed to solve the problem.The proposed algorithms are validated by two preference articulation methods of reference point and target region.Simulation results show that the proposed algorithms can find the preferred Pareto optimal solution in two-objective,three-objective and five-objective mission planning problems.