Suitability of fuzzy systems and neural networks for industrial applications

Abstract

The presentation provides a comparison of fuzzy and neural systems for industrial applications. Both neural networks and fuzzy systems perform nonlinear mapping and both systems internally operate within a limited signal range between zero and one. Neural networks can handle basically an unlimited number of inputs and outputs while fuzzy systems have one output and number of inputs is practically limited to 2 or 3. The resulted nonlinear function produced by neural networks is smooth while functions produced by fuzzy systems are relatively rough. At the same time the design of fuzzy systems transparent and easy to follow while the development of neural networks is much more labor intensive. It is shows that most commonly used neural network architecture of MLP - Multi Layer Perceptron is also one of the least efficient ones. Also most commonly used EBP - Error Back Propagation algorithm is not only very slow, but also it is not able to find solutions for optimal neural network architectures. EBP can solve problems only when large number of neurons is used, but this way neural network loses its generalization property. Performances of both fuzzy systems and neural networks are compared leading to the conclusion that neural networks can produce much more accurate nonlinear mapping and they may require less hardware.