Font Size: a A A

Theories And Methods For Offset-Free Model Predictive Control

Posted on:2016-02-29Degree:DoctorType:Dissertation
Country:ChinaCandidate:H K WangFull Text:PDF
GTID:1228330461452650Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
The rapid industrialization has been provided great opportunities for process indus-tries. At the same time, process industries have to face the daily challenge of squeezing on cost, throughput, quality or environmental compliance. Advanced process control s-trategies, e.g. model predictive control (MPC), are powerful tools to tackle the challenges. Offset-free control means system abilities that tracking the given reference signals and mak-ing sure the closed loop control system has no steady-state offsets. Offset-free control is a fundamental reqirement in the design of a control system. Existing MPCs can achieve offset-free control in general, however, they perform differently in disturbance rejection because of the disturbance models and the feedback policies. Unmeasured disturbances in existing offset-free MPC approaches are assumed to be integrated white noises with un-known statistics, this assumption may lead to degradation of system performance. On one hand, such integrated white noise models are rarely adequate to describe the dynamic be-havior of the disturbances, because the nature of disturbances is often unknown. Optimal filtering performance can not be guranteed if the filter is designed based on such integrated white noise models, and this will affect the disturbance rejection performance. On the other hand, steady-state data is required to obtain the statistics of the unmeasured disturbances in existing approaches, and this will increase the computational complexity more or less. If the controlled plant works within a wide range or the load changes frequently, steady-state data is difficult to obtain, and the corresponding control performance will deteriorate over time. This paper deals with disturbance modeling and the design of filters to improve the control performance, especially the disturbance rejection performance. The main contributions are summerized as follows.(1) The problem of minimum-variance unbiased (MVU) filtering for linear systems with disturbances that have unknown statistics is studied. To extend the MVU filters to a general case, a recursive filter for linear systems with a rank deficient matrix is pro-posed to estimate both the state and the disturbance. The aysmtotic stability and global optimality of the proposed filter are discussed.(2) An MPC framework with MVU filters is proposed to deal with unmeasured distur-bances with unknown statistics. This approach allows unmeasured disturbances with arbitrary statistics, and the work required by disturbance modeling can be alleviated, and the computational complexity can be reduced.(3) A new control strategy for Hammerstein systems with unknown nonlinearities is pro-posed to overcome the shortcomes of existing approaches, e.g. computational com-plexity, inversion of nonlinear functions, etc. This approach enables the MPC to better handle, model mismatch and time-vary ing parameters, and the disturbance rejection performance can be improved.At last, a brief conclusion is given, and some directions can be studied in the future are provided.
Keywords/Search Tags:Offset-Free Control, Model Predictive Control, Disturbance Rejection MVU Filters, Hammerstein Systems
PDF Full Text Request
Related items