Identification and self-adaptive control.

Identification requires assumptions about the unknown disturbance. For an open loop identification experiment, the disturbance and the input are physically independent. Therefore the most reasonable assumption is one of statistical independence. The estimation technique presented in the first part of the thesis exploits the statistical independence to reduce parameter estimation errors. The resulting algorithm is identical to that of the two stage least squares method [30].;Nevertheless it is believed that the original aspect of the approach is the treatment of the disturbance. In practice it is often desirable to avoid open loop experimentations due to economic and safety restrictions. In the second part of the thesis, the identifiability problem for the Box and Jenkins feedback control system is re-examined to extend the previous work of Turtle [52]. Based on a new estimation error equation, a self-adaptive optimization procedure is proposed. However due to doubts concerning stability and convergence, the procedure is not, at the moment, sufficiently well-developed for practical applications.;The final part of the thesis investigates the possibility of estimating the process and disturbance dynamics by use of an external perturbation signal while the process is operating under the Box and Jenkins control system. It is shown that a correlation analysis can produce consistent estimates. However for a limited amount of data these estimates are, in general, biased and hence do not always produce an optimum control system. Nevertheless further improvement can be achieved by applying the predictor updating procedure of Turtle [52] after the correlation analysis.