Adaptive neuro fuzzy Inference systems in identification, modeling and control: The state-of-the-art

Abstract

Adaptive Neural Fuzzy Inference Systems ANFIS have an increasing tendency to be used in scientific research and practical applications. The digitization of production and the emergence of Industry 4.0 enabled the development of this trend, primarily due to the ability to adapt to the task by integrating artificial neural networks and fuzzy logic, which can potentially use the advantages of both techniques in unique frameworks. This approach facilitated the modeling, data analysis, classification and control processes. The advantage of the ANFIS, compared to conventional methods, is reflected in the ability to predict the output based on a set of inputs and on the rule base. Also, these systems are suitable, because they provide the possibility to adjust the parameters of the control system. This paper presents the structure of the ANFIS system and gives a detailed review of the achievements so far, through a comparative analysis, where some possible spheres of interdisciplinary application are highlighted. Possibilities for variations, improvements and innovations of the algorithm, as well as reducing the complexity of the network architecture itself, are discussed. Proposals for some new, as yet unused combinations with metaheuristic optimization methods are presented. Finally, important guidelines are provided on when and where it is useful to apply ANFIS systems.

Heat Flow Process Identification Using ANFIS-GA Model

“…One optimization algorithm can be used to set all parameters, or the parameters in the premise of ANFIS are set by one algorithm, and the parameters of the consequence by another algorithm. When using one of the gradient algorithms, there is a risk of getting stuck in a local minimum, and this is exactly what paved the way for metaheuristic algorithms [11]. An extensive review of the recent literature shows that metaheuristic algorithms are far more common than gradient algorithms and that their number is still growing Figure 3 [11].…”

“…When using one of the gradient algorithms, there is a risk of getting stuck in a local minimum, and this is exactly what paved the way for metaheuristic algorithms [11]. An extensive review of the recent literature shows that metaheuristic algorithms are far more common than gradient algorithms and that their number is still growing Figure 3 [11].…”

“…Figure 3 -The ratio of representation of metaheuristic algorithms in hybrid ANFIS structures (right).Percentage of papers on KOBSON when ANFIS and metaheuristic algorithms are keywords (left)[11].…”