Multi-Objective Differential Evolution with Multi-Noisy Random Vectors (mnv-MODE) for the Solution of Many-Objective Optimization Problems

The ubiquity of multi-objective optimization problems (MOOPs) in real life attracted the attention of many scientists during the last two decades and motivated them to do a large amount of research in multi-objective evolutionary algorithms (MOEAs) which are broadly used in solving MOOPs. However, no algorithm can be considered as the universal optimizer for MOOPs. In this dissertation, multi-objective differential evolution (MODE) is used to develop a new approach called mnv-MODE which aims to solve ZDT1-ZDT4, ZDT6, UF1-UF10 and MaOP1-MaOP10 benchmark problems with 2, 3 and 5 objectives. Four different versions of the proposed algorithm are introduced by modifying MODE and using a local search. Compared to other MOEAs, the results show that our proposed mnv-MODE versions (especially version 4) have the best IGD values on the majority of test instances. This means that mnv-MODE achieved better performance than some efficient algorithms such as SPEA2, MOEA/D and NSGA- II for the solved test Problems. Keywords: Multi-objective optimization problems, multi-objective evolutionary algorithms, multi-objective differential evolution.