Identification For Non-uniformly Sampled-data Systems

The traditional sampled-data systems are with equal input updating period and output sampling period, the operation frequency of which are same and synchronous, such systems with single operation frequency are called single-rate systems. However, in the chemical processes, many variables that indicate the product quality can not be online measured, which results in slower output sampling frequency than input updating frequency. Such systems with two or more operating frequencies are called multirate systems. Although this kind of systems with many different frequencies, the sampling frequency for every input or output channel always remains the same, in other words, the sampling instants of which are constant. If the updating and sampling instants of the systems are non-uniformly spaced, then we get non-uniformly sampled-data systems. These systems exist widely in process industries due to hardware device constraints. Seaching for effective parameter identification methods for non-uniformly sampled-data systems is worth with theoretical and engineering values. Therefor, the research mainly focuses on identification for non-uniformly sampled-data systems. Basing on plentiful references, the subject is studied in this thesis and the main contributions are as follows.1. With white noise interference, the CAR model for non-uniformly sampled-data systems is derived. To speed up the convergence rate and improve the parameter estimation accuracy, by introducing the forgetting factor and the innovation length respectively, the forgetting gradint algorithm and the multi-innovation stochastic gradient algorithm are presented. The convergence of the two improved algorithms are analyzed detailedly by using the theory of stochastic processes, and the effectiveness of the proposed algorithms is verified by the simulation examples.2. By applying auxiliary model method, the auxiliary model based stochastic gradint algorithm is proposed to estimate the parameters of output error model for non-uniformly sampled-data systems, and its convergence property is proved by using the convergence theorem of martingale. The auxiliary model based forgetting gradint algorithm and the auxiliary model based multi-innovation stochastic gradient algorithm are presented too. The simulation examples show that the identification results are satisfactoried by choosing appropriate forgetting factor and innovation length.3. When the disturbed noises are depicted by MA models and ARMA models, respectively, the gradient type identification methods of CARMA model, CARARMA model, OEMA model and Box-Jenkins model for non-uniformly sampled-data systems are studied. The basic idea is to replace the unknown noise terms with their estimates and the unknown true outputs with which of the auxiliary model. The simulation results indicate that these algorithms are effective.