Researches On A Class Of Species-Based Evolutionary Algorithm For Multi-Objective Optimization Problems

A lot of application problems in the science and engineering applications usually involve multiple objectives which often conflict with each other. These multi-objective optimization problems (MOOPs) are rather different from the traditional optimization problems with single objective due to inexistence of the global optimal solution which can optimize all objective simultaneously. Obviously, such this complex feature would pose a great challenge to the solution algorithm for MOOPs. Evolutionary algorithms (EAs) have been widely applied into addressing MOOPs since the population-based mechanism enables an EA achieve multiple Pareto optimums in its single running course, which is just the major reason that evolutionary multi-objective optimization has become a newly-generated focus from the community of evolutionary computation in recent years.It is noticeable that searching multiple optimal solutions in parallel is not such a challenge only when an EA is used to address MOOPs. When addressing multi-modal optimization problems, EA is also required to achieve as many as possible global optimum even some local optimal solutions. Therefore, it becomes a very interesting research issue to introduce the algorithm mechanism, which was used in EAs for multi-modal problems, into the design of EAs for MOOPs.Based on the mechanism of system engineering, this thesis will investigate and study a class of new multi-objective EAs that are inspired from the Species-based scheme, which are firstly proposed for EAs in multi-modal optimization. The main contents of this thesis can be summarized as follows.Firstly, two typical EAs, that is, genetic algorithm (GA) and particle swarm optimization (PSO), are introduced their algorithm principles in brief and the relevant research works on GA and PSO for MOOPs are reviewed.Secondly, a new species-based multi-objective GA (speMOGA), which combines the species-based scheme and the mechanism of NSGA-? that was a well-known multi-objective GA, is proposed for MOOPs. In the proposed algorithm, a species seed determination method based on two new features of individuals and a species construction method where a species is adaptively constructed by a species seed and a predefined number of individuals that are the closest to the species seed in the current population are specially designed in multi-objective optimization respectively.Thirdly, the species mechanism is extended into PSO for MOOPs and then a new species-based PSO algorithm is also designed. In the proposed algorithm, a species seed determination method based on an extra archive of the achieved Pareto solutions and an adaptive species construction method that is similar to the species construction method in speMOGA are designed respectively. In addition, the update methods of each individual's Pbest and Gbest are also specially designed in order to the proposed algorithm adapt well in multi-objective optimization.Finally, all proposed algorithms are tested their performances through the comparisons with their peer algorithms based on a series of benchmark multi-objective functions. In addition, several key parameters and operators are also examined their influence degree upon the performance of proposed algorithms.