An improved artificial immune network for multimodal function optimization

Multimodal optimization problems are typical those problems where both a global optimum and one or more local optima are included, which can be tackled with by evolutionary algorithms. This paper proposes an improved artificial immune network for multimodal function optimization. In the antibody population, antibodies are allocated to three spaces, i.e., the elitist space, the common space and the poor space. To be specific, the antibodies in the elitist space undergo self-learning mutation, those in the common space undergo elitist-learning mutation and those in the poor space undergo the random hypermutation. At each generation, the mutated antibodies with higher affinity are regrouped into a new elitist space while those with medium affinity are regrouped into a new common space. On the other hand, the old poor space is removed; instead, a group of randomly generated antibodies are grouped into a new poor space. Six benchmark functions in both 10D and 30D are used to evaluate the optimization performance of the proposed artificial immune network and the other existing artificial immune networks, such as opt-aiNet, IA-AIS and AINet-SL. The simulation results indicate that the proposed artificial immune network is an effective and relative efficient method in optimizing both the unimodal and multimodal functions.