Effects of User Selected Conditions on Modelling of Dynamic Systems Using Adaptive Fuzzy Model

Abstract

Kertas kerja ini membincangkan sifat–sifat utama satu pengenalpasti sistem iaitu model kabur suai yang dilatih dengan algoritma perambatan balik. Model kabur piawai berasaskan aturan kabur telah digunakan untuk mengenal pasti sistem dinamik diskret tak lelurus. Kaedah pemilihan pemboleh ubah masukan, bilangan aturan dan kadar latihan ada dibincangkan dengan ringkas. Tiga kaedah pemilihan parameter awal telah dipertimbangkan, iaitu kaedah pemilihan dalam talian, luar talian dan rawak. Aspek pelaksanaan dan pengiraan algoritma ini turut diketengahkan. Tiga contoh sistem dinamik tak lelurus telah digunakan untuk menunjukkan kesan–kesan kaedah latihan yang dipilih oleh pengguna dalam proses pengenalpastian ini. Keputusan daripada proses pengenalpasti model ini menunjukkan ia boleh menganggarkan sistem dinamik dengan baik. Ujian pengesahan model secara sekaitan telah digunakan untuk mengesahkan kecukupan model berkenaan. Kata kunci: Pengenalpasti sistem; model kabur; algoritma perambatan balik; sistem dinamik. In this paper, major properties of an adaptive fuzzy model as a system identifier when trained by the back–propagation algorithm are discussed. The standard rule–based fuzzy models were used to identify discrete–time nonlinear dynamic systems. The method of selection of the input variables, the number of rules, and the learning rate are briefly discussed. Three methods for choosing the initial parameter of the fuzzy model are considered, namely the on–line, the off–line and the random initial parameters. The implementation and the computational aspects of the training algorithm are also highlighted. Three examples of discrete–time nonlinear systems are used in the simulation study to show the effects of user selected conditions on the identification process. The results of the identification procedure show that they approximate the dynamic plants quite well. The correlation based model validity tests are used to validate the identified fuzzy model. Key words:System identification; modeling; fuzzy system; back-propagation algorithm; dynamic systems.