Font Size: a A A

Identification For Nonlinear Systems With Non-Uniformly Sampled-data

Posted on:2015-09-06Degree:DoctorType:Dissertation
Country:ChinaCandidate:R R LiuFull Text:PDF
GTID:1228330467975935Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Due to hardware limitation, economic constraint and environmental requirement, there are many non-uniformly sampled-data (NUSD) systems in process industry (such as petroleum, chemical, food, medicine and so on). However, the parameters of NUSD systems cannot be estimated using the traditional identification algorithms designed for signal-rate sampled-data systems. In order to overcome this issure, under the susports of National Natural Science Foundation (61273142), and Foundation project of Technology Innovation for Graduate in Jiansu Province (CXLX120648), several identification algorithms have been proposed to estimate the parameters of NUSD nonlinear system. Here, we only focus on three kinds of NUSD nonlinear systems, i.e. polynomials with known orders, hard nonlinearities and normal nonlinearity with unknown structure, and then investigates the application in soft sensor. The contributions are as follows:· A hierarchical multi-innovation stochastic gradient identification algorithm (HMISG) is proposed for Hammerstein-Wiener (H-W) nonlinear systems with non-uniformly sampling data. Firstly, the corresponding state space models of H-W system are derived using the lifting technique. Considering the causality constraints, the H-W system is decomposed into two subsystems. Then, the parameters are estimated using the multi-innovation based stochastic gradient algorithm with forgetting factors (VFF-HMISG). Furthermore, a novel tuning function is presented to adjust the forgetting factor online. As a result, the convergent rate of VFF-HMISG is improved and the disturbance is also rejected. Simulation examples validate the performance of proposed algorithm. Finally, this VFF-HMISG is extened to identify NUSD Hammerstein/Wiener systems.· An iterative recursive least squares identification algorithm is proposed to identify the NUSD Wiener system with dead-zone nonlinearities. Firstly, the corresponding state space models of Wiener system are derived using the lifting technique. Then, the analytic form of the system is obtained by using the switching function. However, the information vector contains unknown inner variables and unknown parameters together, and it is difficult to estimate those parameters simultaneously. In order to deal with this issue, an auxiliary model-based iterative recursive least squares algorithm with variable forgetting factor (VFF-AM-IRLS) is presented in this paper. Finally, the numerical simulation demonstrate that the proposed VFF-AM-IRLAS algorihm has high accuracy and low computational burden.· Aimed to a generalized NUSD nonlinear system with unknown structure, a multi-model recursive least squares algorithm based on fuzzy c means cluster (FCM) is proposed in this paper. Firstly, based on divide-and-conquer strategy, the system is described as a kind of switching system. Secondly, the switching strategy and sub-dataset of each local model are obtained by using the fuzzy c means cluster. As a result, the nonlinear identification problem is transformed into the identificantion of multiple linear local models, and a least squares algorithm is employd to achieve this target. Finally, the proposed algorithm is validated by usinga mathematic model and pH neutralization process.· For a practical NUSD nonlinear system, i.e, the p-xylene oxidative side-reaction, a soft sensor based on FCM is developed to predict the content of carben dioxide in reaction off gas. First, after normalized and correlation analysised, the input and output variables are obtained form off-line lab analyses datas. Then, to obtain high-performance soft sensor, the system constructure is determined using FCM cluster algorithm, and parameters of local models are estimated based on multi-model recursive least squares algorithm. Furthermore, simulation results show better performance of the soft sensor.
Keywords/Search Tags:non-uniformly sampling, nonlinear systems, stochastic gradient, recursive algorithm, least squares algorithm, local model networks, fuzzy c-meancluster, soft sensor
PDF Full Text Request
Related items