Neural Networks in System Identification

System identification is an important way of investigating and understanding the world around. Identification is a process of deriving a mathematical model of a predefined part of the world, using observations. There are several different approaches of system identification, and these approaches utilize different forms of knowledge about the system. When only input-output observations are used behavioral or black box model can be constructed. In black box modeling neural networks play an important role. The purpose of this paper is to give an overview of the application of neural networks in system identification. It defines the task of system identification, shows the basic questions and introduces the different approaches can be applied. It deals with the basic neural network architectures, the capability of neural networks and shows the motivations why neural networks are applied in system identification. The paper presents the main steps of neural identification and details the most important special problems, which must be solved when neural networks are used in system modeling. The general statements are illustrated by a real world complex industrial application example, where important practical questions and the strength and weakness of neural identification are also discussed.