多目标进化算法中多样性策略的研究

Diversity Strategies on Multiobjective Evolutionary Algorithms

进化多目标优化中由于进化算子固有的随机误差以及进化过程中选择压力和选择噪音的影响使得进化群体容易丧失多样性,而保持进化群体的多样性不仅有利于进化群体搜索,而且也是多目标优化的重要目标。对多目标进化算法的多样性策略进行了分类,在统一的框架下描述了各种策略的机制,并分析了各自的特性。随后,分析并比较了多样性保持算子的复杂度。最后,证明了一般意义下多目标进化算法的收敛性,指出在设计新的多样性策略中需要保证进化世代间的单调性,避免出现退化现象。

The intrinsic random errors on the evolutionary operators and the pressure of selection and the noise of selec- tion in the evolutionary process easily make the loss of diversity on the evolutionary population. But the maintenance of diversity on the population is very important because it is not only benificial to the search process but also becomes the essential objective in multiobjective optimization. With the unified framework, this paper first classifiesd the diversity strategy on the MOEAs and described the principles and mechanisms on different types of diversity strategies and ana- lyzed their characteristics. Then this paper analyzed the complexity of these diversity operators. At last, this paper proved the convergence of MOEAs in the general sense and pointed out that it is necessary to keep the monotonicity in the evolutionary population and avoid the degradation of population as the design of new diversity strategy.