Real Time Block Recursive Parameter Estimation of Second Order Systems

PhD Thesis, 1997

C. Goodwin

Abstract

Many real world dynamic systems can be approximated well using second order systems. It is often required, therefore, in engineering and other situations to determine the characterizing parameters of observed data, with the assumption that the data represents a second order system.

This study investigates the parameter estimation problem encompassing a wide range of techniques and algorithms. Conventional approaches are tested and in some cases combined to produce hybrid algorithms. Two novel methods are also applied, and compared with the other techniques. These novel methods are neural networks and genetic algorithms.

Further, a new algorithm is proposed which is applied to all techniques tested. This new algorithm adaptively adjusts the sampling frequency at which observed data is read, based on previous estimates of the parameters. It is shown that this improves the accuracy of the parameter estimation process.

A complete simulation environment is devised enabling parameter estimation to be tested under a range of situations. Firstly, when the system parameters are constant with time. Then secondly, when the parameters vary through the time period of the observed data. The simulation enables the parameters to be estimated in blocks of data. Further enhancement of the algorithms enable them to perform recursively, taking account of previous block’s estimates. Finally, all algorithms are tested on their tolerance to two types of noise. The complete simulation allows recursive block parameter estimation which adaptively varies the sampling frequency to increase the accuracy of the estimation, under a range of noise conditions.