Neuro-Fuzzy Control
Performance improvement of fuzzy logic controllers (FLC) can be achieved by adjusting the membership functions (MF). Neuro-fuzzy approaches are mostly used in such adjustment procedure, which involves several parameters of the MFs to be adjusted. In many
- PDF / 1,616,004 Bytes
- 38 Pages / 439.37 x 666.142 pts Page_size
- 85 Downloads / 205 Views
Neuro-Fuzzy Control
7.1 Introduction The performance of any fuzzy system or fuzzy controller mainly depends on the input–output membership functions, the If–Then rules, and tuning of both (Nauck and Kruse 1993, 1996). The choice of defuzzification method is another factor, which also influences the performance (Yager and Filev 1994). Unfortunately, there are no formal methods to define the membership functions or to construct the rule-base for fuzzy systems or controllers. The issues have been prominent in Chap. 6 and evolutionary learning1 approaches were explored to address some of the issues. In Chap. 6, evolutionary learning is seen as an optimisation or search problem requiring a simple scalar performance index. The performance of the fuzzy system is aggregated into a scalar performance index on which basis evolutionary algorithms select outperforming rule-base, MFs or scaling parameters or their combinations. Evolutionary learning algorithms are the suitable choices where no a priori information about the MFs and the rule-base is available. There have been many successful applications of evolutionary fuzzy systems reported in the literature. Due to the nature of evolutionary algorithms, evolutionary fuzzy systems are presumably slow processes and the performance of the system inherently depends on the size of the population and the number of generations required for a solution to be robust for specific problems. The most striking features of neural networks are their flexible structures, available learning algorithms and capability of learning from experiential data. Due to these inherent advantages, neural networks found applications in many engineering applications such as pattern recognition, signal processing, modelling and control of complex systems (Akesson and Toivonen 2006; Narendra and Parthasarathy 1990; Narendra and Mukhopadhyay 1997; Sarangapani 2006). Consequently, the combination of neural networks with fuzzy systems has been recognised as a powerful alternative approach to learning fuzzy systems. Such a combination should be able to learn linguistic rules, membership functions, or to optimise existing ones. Learning 1
An influential paper by Hinton and Nowlan (1987) showed that learning can guide evolution and learning evolution can work synergistically together. N. Siddique, Intelligent Control, Studies in Computational Intelligence 517, DOI: 10.1007/978-3-319-02135-5_7, Springer International Publishing Switzerland 2014
179
180
7 Neuro-Fuzzy Control
in this case means creating a rule-base or membership functions from scratch based on training data presented as a fixed or free learning problem (Nauck and Kruse 1996). The learning procedure operates on local information, and causes only local modifications in the underlying fuzzy system. Some neuro-fuzzy systems are capable of learning and providing fuzzy rules in linguistic or explicit form. However, most of the current neuro-fuzzy approaches address parametric identification or learning only. In general, the designer chooses the shap
Data Loading...
