Preference Algorithms Based On Many-objective Optimization

Artificial immune system is a kind of highly evolved and parallel distributed adaptive system. It is equipped with stronger learning, memory, recognition and feature extraction ability which the biological immune system has. Based on these principles and mechanisms, artificial immune system has been developed. Though information processing the artificial immune system can solve the problems related to the field of evolutionary computation and be applied to the practical engineering successfully. Artificial immune algorithm is just a kind of learning algorithm based on this system and it is also an important part of research of artificial immune system. In order to improve the search capability of artifical immune algorithms and save the computing resources, in this paper we research artifical immune system based preference multi-objective optimization algorithm. The specific work is arranged as follows:1. A new reference direction based immune clonal algorithm is proposed. It uses the reference direction based preference method to assign the preference rank value to every antibody, and preferentially choose the elites near the preference region to fill the active antibody population. Then according to the proportionally cloning, the preference rank the antibody has is higher and the opportunity of its cloning is bigger. Moreover, we use the intelligent combination operator to do crossing which can solve the multi-objective optimization problems with a lot of decision variables. Finally, this algorithm integrates the light beam search to select the external population. The experiments prove that this proposed algorithm can solve the many-objective optimization problems successfully, even for the problem with one hundred objectives.2. A new angle based preference selection machanism is proposed. The majority of the existing preference selection methods has a complicated series of steps and some methods need to use.the large amount of calculation of the scaling function. So we except to create a kind of simple preference model. In fact our proposed preference selection mechanism just uses the angle between the solution vector and preference direction as the selection criteria of solutions. It is easy to understand and operate. Moreover, we combine it and popular NNIA and mainly use it to select the active antibody population and external population. Experiments prove that Angle-NNIA can solve the many-objective optimization problems successfully. Compared to other preference algorithms and operators, it shows the absolute advantage.3. In this chapter, we improve the algorithm Angle-NNIA proposed in the last chapter. On the one hand we change the selection method of external population into light beam search. But its constraint condition is so harsh that the obtained solutions is too rare. So on the other we implement the adaptive processing in the process of selecting active antibodies from the external population. Experiments prove that this modified algorithm performs the stable convergence on DTLZ1, DTLZ2and DTLZ3problems.