Use of self-organizing suppression and q-Gaussian mutation in artificial immune systems

摘要

Purpose - The purpose of this paper is to propose two operators for diversity and mutation control in artificial immune systems (AISs). Design/methodology/approach - The proposed operators are applied in substitution to the suppression and mutation operators used in AISs. The proposed mechanisms were tested in the opt-aiNet, a continuous optimization algorithm inspired in the theories of immunology. The traditional opt-aiNet uses a suppression operator based on the immune network principles to remove similar cells and add random ones to control the diversity of the population. This procedure is computationally expensive, as the Euclidean distances between every possible pair of candidate solutions must be computed. This work proposes a self-organizing suppression mechanism inspired by the self-organizing criticality (SOC) phenomenon, which is less dependent on parameter selection. This work also proposes the use of the q-Gaussian mutation, which allows controlling the form of the mutation distribution during the optimization process. The algorithms were tested in a well-known benchmark for continuous optimization and in a bioinformatics problem: the rigid docking of proteins. Findings - The proposed suppression operator presented some limitations in unimodal functions, but some interesting results were found in some highly multimodal functions. The proposed q-Gaussian mutation presented good performance in most of the test cases of the benchmark, and also in the docking problem. Originality/value - First, the self-organizing suppression operator was able to reduce the complexity of the suppression stage in the opt-aiNet. Second, the use of q-Gaussian mutation in AISs presented better compromise between exploitation and exploration of the search space and, as a consequence, a better performance when compared to the traditional Gaussian mutation.