Plenary lecture 1: identification of non-linear systems using neural networks, with application at fuzzy systems
Constantin Voloşencu · International Conference on Information and Automation · 2009
The paper present a short review of the ways of using artificial neural networks with continuous values for non-linear identification, with application at the fuzzy systems. Feedforward neural networks with continuous values may be seen as general approximation functions. Using adequate training methods and according Kolmogorov's theorem, we may approximate any kind of multivariable non-linear functions. Fuzzy systems, developed based on different membership functions, inference methods, rule bases and defuzzification methods are non-linear systems. The paper presents some study cases of using feedforward neural networks with hidden layer and neurons with continuous values to approximate fuzzy systems with two input and one output variables. Some quality criteria for the training of the neural networks are introduced. Based on usage of different training methods, training parameter of neural networks a comparison is presented, emphasising the quality of training.