scientific article; zbMATH DE number 18410
[For the entire collection see Zbl 0722.00052.] Input-output system identification is here taken to have the following meaning: from prior knowledge and observation of outputs produced by known inputs, it is desired to construct a mathematical model for the unknown physical system that will allow one to predict to within a specified tolerance the system output (or perhaps only certain attributes of the output) resulting from an arbitrarily y chosen input belonging to a specified class. The mathematical models themselves are referred to simply as systems most of the time. These systems are necessarily comprised of two sets, the elements of which represent inputs and outputs, respectively, and a relation between the input and output sets. However, we restrict the generality by requiring the relation to be a function from inputs to outputs; thus an output is determined by an input --- a situation already implied in the definition of identification. A system is denoted by \((Y,F,U)\), or some modification thereof, where \(U\) is a set, the input space; \(Y\) is a set, the output space, and \(F\) is a mapping from \(U\) into \(Y\). Although we believe there is only a trivial identification ``theory possible at the level of generality indicated above with \(U\) and \(Y\) described only as sets, this is where we start. Mathematical structure is then imposed in two steps. First, \(Y\) is required to be a linear space, which allows consideration of linear spaces of mappings \(F\). Second, \(U\) and \(Y\) is to be a normed linear space. This much structure allows consideration of normed linear space. This much structure allows consideration of normed linear spaces of mappings from \(U\) into \(Y\), and of metric subspaces of these. A next logical step would be to introduce time and causality structure, i.e. to require \(U\) and \(Y\) to be spaces of time-functions and \(F: U\to Y\) to be causal (nonanticipative). This is not properly done here; there is only a brief concluding remark to indicate how the material in Sections 2 and 3 applies to time-systems. There has been no mention of probability or of statistical estimation, nor will there be, since we are not considering the effects of noisy observations and we are not considering inherently random systems in this note. With no structure at all or with the single condition that the output space \(Y\) be linear, only exact identification can be considered, and this is the first topic to follow. The concept of exact identification appears to be of little direct practical use, but it is of some interest, nevertheless. With \(U\) a metric space, \(Y\) a normed linear space, and \({\mathcal H}\) a subset of a normed linear space, one can consider approximate identification and also some notions of difficulty of identification. These topics follow in Section 3.
- On metric dimensions of discrete-time systems
- Identifiability of classes of input-output systems
- Complexity of identification of input-output systems
- scientific article; zbMATH DE number 3968483 (Why is no real title available?)
- scientific article; zbMATH DE number 12053 (Why is no real title available?)
- A DESCRIPTION OF A SET OF HYPOTHESES FOR SYSTEM IDENTIFICATION PROBLEM*
- State space consistency and differentiability
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3975802)
