This chapter aims to review the state of the art in algorithms of multiobjective optimization with artificial immune systems (MOAIS). As it will be focused in the chapter, Artificial Immune Systems (AIS) have some intrinsic characteristics which make them well suited as multiobjective optimization algorithms. Following this basic idea, different implementations have been proposed in the literature. This chapter aims to provide a thorough review of the literature on multiobjective optimization algorithms based on the emulation of the immune system.
Many real world problems involve the simultaneous optimization of various and often conflicting objectives. Evolutionary algorithms seem to be the most attractive approaches for this class of problems, because they are usually population based techniques that can find multiple compromise solution in a single run, and they do not require any hypotheses on the objective functions (e.g. unimodality and convexity). Among other techniques, in the last decade a new paradigm based on the emulation of the immune system behaviour has been proposed. Since the pioneer works, many different implementations have been proposed in literatures. The aim of this chapter is to review the most significant works in this field, giving a common framework for classification and showing strengths and weaknesses of artificial immune systems metaphor in multiobjective optimization with respect to other bio-inspired algorithms.
The chapter is structured as follows. Section 3 gives a background on the immune system and multiobjective optimization terminology used in the chapter. Section 4 explains the methodology used to select the reference list used for the review, while in section 5 the papers are reviewed according to their research field. Finally in Section 6 future and emerging trends in multiobjective optimization with artificial immune systems are drawn.
Key Terms in this Chapter
Antibody: Antibodies are the candidate solutions of the problem to be optimized.
Memory: Memory is an offline repertoire of optimal solutions found during the evolution of the algorithm. Memory has a key role for proof of convergence of a multiobjective algorithm because it ensures the survival of the best configurations (elitism).
Avidity: See Antibody-Antigen affinity.
Suppression: In order to preserve diversity in the solutions at each iteration, antibodies which are very affine to each other (see antibody-antibody affinity) are deleted and eventually randomly replaced. Suppression can be applied either to online population or to offline memory
Antibody-Antibody Affinity: Affinity among antibodies is defined as a measure of the distance among candidate solutions. In accordance with the antibody representation, it is possible to define different distances (e.g. Euclidean for continuous representation, Hamming for binary representation, etc.), both in variable and in objective spaces
Antibody-Antigen Affinity: Scalar index adopted as measure for the goodness of a solution with respect to the objective to be optimized. It is usually related to the objective value to be minimized/maximized, with or without the use of scaling or correcting factors. In multiobjective optimization this index (also referred to as avidity) is usually obtained by ranking solutions in accordance to Pareto optimality conditions
Antigen: In optimization problems, antigens are the optimal configurations of the problem