Automatic synthesis of fuzzy systems: An evolutionary overview with a genetic programming perspective

Abstract

Studies in Evolutionary Fuzzy Systems (EFSs) began in the 90s and have experienced a fast development since then, with applications to areas such as pattern recognition, curve‐fitting and regression, forecasting and control. An EFS results from the combination of a Fuzzy Inference System (FIS) with an Evolutionary Algorithm (EA). This relationship can be established for multiple purposes: fine‐tuning of FIS's parameters, selection of fuzzy rules, learning a rule base or membership functions from scratch, and so forth. Each facet of this relationship creates a strand in the literature, as membership function fine‐tuning, fuzzy rule‐based learning, and so forth and the purpose here is to outline some of what has been done in each aspect. Special focus is given to Genetic Programming‐based EFSs by providing a taxonomy of the main architectures available, as well as by pointing out the gaps that still prevail in the literature. The concluding remarks address some further topics of current research and trends, such as interpretability analysis, multiobjective optimization, and synthesis of a FIS through Evolving methods. This article is categorized under: Technologies > Computational Intelligence Technologies > Machine Learning

Modeling and adaptive neuro-fuzzy inference system control of quarter electric vehicle

“…Nevertheless, the fundamental core of a FIS lies in the establishment of the rule base, which presents two major obstacles: the establishment of a rule base requires expert knowledge of the field of study, and the refinement of the MF base is a key element of the process to reduce the output error. The present study utilizes an ANFIS to attain the identification and fine-tuning of a FIS [28]. The global structure of the ANFIS shares the same elements as the FIS, except for the neural network component which Includes numerous nodes that resemble human neurons [29].…”

Sistemas Fuzzy evolucionários para controle de semáforos

“…Na Figura 1, adaptada de Cordon et al ( 2004), tem-se um esquema básico da automatização de um FIS, onde fica evidente o uso de uma meta-heurística (MH) com a incumbência de reduzir a intervenção humana. Costumase dividir um FIS em dois componentes que formam a Base de Conhecimento: Base de Regras (Rule Base, RB, em inglês), consistindo no conjunto de regras Fuzzy, e Base de Parâmetros (Parameter Base, PB, em inglês), que envolve os parâmetros remanescentes do FIS, como funções de pertinência, operadores de agregação e métodos de defuzzificação (Koshiyama et al, 2019). PSO.…”

“…Há duas abordagens básicas na construção da Base de Regras (Koshiyama et al, 2019), a saber: Classificador Michigan e Classificador Pittsburgh. Na abordagem Michigan cada regra do FIS é considerada um cromossomo, ou indivíduo.…”