Baldwinian Evolutionary Multi-objective Optimization Algorithms Based On Distribution Model Learning Of Population

Nowadays, the research on EMO(Evolutionary Multi-objective Optimization) algorithms has been playing an important role in the area of evolutionary computation. With the advantage of producing a set of Pareto optimal solutions in a single run, EMO algorithms have gradually become the mainstream method in solving the MOPs(Multi-objective Optimization Problems). Therefore, in most of existing EMO algorithms, selection mechanism in the objective space has attracted a lot of research effort and the search mechanism in the decision space just simply inherits the conventional reproduction operators designed for single-objective optimization algorithms. However, few works has been done to improve the searching efficiency of EMO algorithms according to the characteristic of MOPs.Recently, Baldwinian learning has been commonly introduced into evolutionary computation which provides a new path for solving the MOPs. Based on the regularity of continuous MOPs, we proposed a Baldwinian learning strategy. By combining the Baldwinian learning operator with two different evolutionary multi-objective optimization algorithms, two new Baldwinian evolutionary multi-objective optimization algorithms were proposed. So, this thsis contains mainly as follows:(1)A Baldwinian evolutionary operator was proposed based on the distribution model learning of population. Based on the regularity of the continuous MOPs, this operator obtaines the real-time evolutionary information through the establishment of the probability distribution model of current population. Meanwhile, with the historical information of parent population, the individuals in the population could be searched along the forecasted evolutionary direction.(2)The designed Baldwinian evolutionary operator was introduced into two different algorithms, the Multi-objective Immune Algorithm with Nondominated Neighbor-based Selection and the Evolutionary Multi-objective Optimization Algorithm Based on Decomposition, and then two multi-objective optimization algorithms MIAB and MOEA/D/BL were proposed, which were Baldwinian Evolutionary Multi-objective Optimization Algorithm Based on Immune Clone and Baldwinian Evolutionary Multi-objective Optimization Algorithm Based on Decomposition respectively. Through the analysis of the advantages and disadvantages of the two original algorithms, individual was searched by introducing Baldwinian evolution operator at an appropriate place in an appropriate way in MIAB and MOEA/D/BL. Experimental results have showed that this Baldwinian evolutionary operator designed for the characteristics of the problem was efficient. Not only the convergence speed of the original algorithm has been accelerated, but also a good diversity of population has been maintained.