A least-squares fitting technique for use with large nonlinear plant models

The fitting of large steady state plant models to measured plant data is very often hindered by a surfeit of data. It becomes necessary to accept the reading from one instrument measuring one quantity while disregarding that from another measuring another. An approach to the fitting process which could utilise all the available data, with the known uncertainties in these data items being additionally used to perform the necessary weighting is required. Linear regression provides a least-squares solution when fitting a linear set of equations to observations, and it has been found possible to extend this principle to the nonlinear situation. By reference to an example of a gas-cooled reactor, the method is outlined and explained