Ensemble Of Multiple Operators In Multiobiective Evolutionarv Algorithms

Evolutionary Algorithm (EA) is a global parallel random algorithm that imitate the procedure of animals evolve in the nature. EA is simple to implement, flexible, and robust which is generally used in kinds of fields in our world. There are plenty of different reproduction operators for single or multiple objective optimization problems in EA, but they are usually designed for some certain types of problems. It has been proved by theory and experiments that it is impossible to design one single reproduction operator that can fit all kinds of problems. Naturally, we can consider combining different reproduction operators to make use of their specific advantages to solve complicated optimization problems. Considering the above idea, the paper tries to explore different combining strategies and design suitable multi-operator algorithm for multiobjective optimization problems (MOPs).This paper works firstly on classifying those multi-operator strategies in the literature and then exemplify the core different between those strategies by giving the "The Jury System". We tries our best to show each strategy work using as few words as possible and then goes into them by giving classic algorithms.The given classic multi-operator algorithms are all designed for single objective optimization problems and there is little work focusing on multi-operator algorithms for multiobjective optimization problems currently. This situation is caused by MOP’s complexity. Based on the MOEA/D framework, the paper presents two multi-operator algorithms for MOP:fixed multi-operator algorithm and cheap surrogate model based multi-operator algorithm. In each chapter, we introduce them and prove their efficiency by experiments.Fixed multi-operator algorithm ensembles several differential evolution (DE) operators to solve continuous MOP. Firstly, we pick out three DE mutation operators, and then tune them to obtain a parameter pool for those DE operators. At last, the algorithm produces new candidate solutions using these DE operators and randomly selected parameters in the parameter pool.It is not easy to build and update a suitable probability model. Moreover, an inaccurate model would misguide the search. To avoid this problem, this paper proposes a cheap surrogate model based multi-operator algorithm for MOP. Firstly, we generate N candidate solutions using N reproduction operators for each parent solution. Then we select one best solution out of those candidate solutions using the designed cheap surrogate model. Last, we use the selected solution to update parent solutions.