Research On Evolutionary Operators And Algorithms Of Multi-ojbective Optimization

Multi-objective problems, requiring the optimization of two or more conflicting criteria, abound in the real world. Multi-objective optimisers produce solution sets that represent the trade-offs between problem criteria. In general, there are essentially three fundamental evolutionary operators in Evolutionary Multiobjective Optimization Algorithms, i.e., selection, crossover and mutation, which are employed to gradually approach the optimal solution. Considering the crossover operator, many research works have been conducted on improving the performance due to its importance to evolutionary procedure. This thesis designs three novel evolutionary operators considering the following two aspects. Firstly, different algorithmic frameworks need different crossover operators to compensate its dificiency. Secondly, the quality of the offspring produced by the crossover operator is highly dependent on the characteristics of target problems.Firstly, we proposed an adaptive differential evolutionary operator(ADE) for immune algorithm fromework. We found that immune algorithm lacks of diversity due to its clone operator, so we applied DE with global search ability to the framework of immune algorithm. We have two modifications when cooperating DE to multiobjectie optimization. Firstly, we select the parents for differential vector from non-dominated and dominated archive to provided correct direction for evolution. Secondly, we design an adaptive parameter control method. The effictiveness of ADE was validated by experiments.Secondly, we proposed a hybrid evolutionary crossover operator(AHX) based on the gene level. Different from the existing hybrid operators that are commonly operated on chromosome level, the proposed operator is executed on gene level to combine the advantages of simulated binary crossover(SBX) with local search ability and DE with strong global search capability. More opportunities are assigned to DE in the early evolutionary stage for global search; whereas, with the generation grows, more chances are gradually allocated to SBX for local search. The balance between global and local search is well maintained by an adaptive control approach.At last, we proposed an archive-guided velocity update meothod for multiobjective particle swarm optimization(MOPSO). The global best and personal best particles are important for multi-objective optimization problems, but they are difficult to select. How to properly select leaders is key challenge in MOPSO. We designed an elite archive based on decomposition and select three leader particles(global best, local best and personal best) from the archive to guide the whole particle swarm to evolve. The experimental studies validate that the designed velocity update method can improve the performance of MOPSO algorithm.