Multi-Objective Structure and Parameter Evolution of Neuro-Fuzzy Systems
Amiram Moshaiov, Adham Salih · 2021 IEEE Symposium Series on Computational Intelligence (SSCI) · 2021
This paper presents an evolutionary algorithm that allows the adaptation of neuro-fuzzy systems according to conflicting objectives. The adaptation includes both the structures and the parameters of the evolved solutions. The suggested algorithm, which is termed Fuzzy-Evolution of Membership and Structures by Decomposition (FEMS/D). As suggested by its name, the algorithm is based on the decomposition approach, which is a divide-and-conquer technique that transforms the original multi-objective problem into a set of single-objective subproblems. In addition, this paper provides initial demonstrations of the applicability of FEMS/D to the development of nondominated robot motion controllers, which exhibit a range of behaviors between the safest and the fastest ones. Finally, the adaptation of the controllers to an abrupt change in the environment is also demonstrated.