ABSTRACT
This chapter considers parameter estimation techniques. These techniques are essential in system identification, as they provide the means for determining (off-line) or adjusting (on-line) the parameters of a chosen model structure, using sampled data {measurements). The least squares method can be applied when the system output is linear with respect to its parameters. This is true for linear static mappings {linear regression models), as well as for some linear dynamic structures (such as ARX structures), and some non-linear systems (e.g., power series and multi-linear systems). However, usually the least squares method can not be directly applied in non-linear or dynamic systems, since these types of systems are, in general, non-linear with respect to their parameters. In the previous chapters we have pointed out structures such as the OE {Chapter 3), the sigmoid neural networks {Chapter 4), or the Wiener structure {Chapter 5), with which the least squares method can not be directly applied.
