The Modeling And Controller Design Based On The PLS Method

With the development of information and computer technology, the data-driven methods and theories have achieved great breakthroughs, which arouse significant attention in both academic and industrial field. As a multivariable regression technique, Partial Least Squares (PLS) has been successfully applied in process industry areas like process modeling, fault diagnosis soft sensor and so on. However, there are still some problems in controller design based on the PLS method. With the sophisticated mechanism inherent in the process as well as the strong nonlinearity and multivariable, it is extremely difficult to establish the rigorous mechanism model; this also affects the application of control law. So by incorporation the excellent merits of PLS method such as the self-decoupling, dimension reduction and collinear elimination, we propose an adaptive model updating method and its control strategy, we also develop a Dynamic Fuzzy PLS (DFPLS) modeling method, and the controller design strategy based on the DFPLS model. The main topic of this thesis shows as:(1) There might be mismatch between the model and the process which may be caused by the environment changes, operation mode shift or aging equipments. The mismatch may have negative effect on the controller, which even may deteriorate the stability of the whole system. To solve this problem, we propose an adaptive model updating algorithm by combining the Recursive PLS (RPLS) method and the Recursive Least Squares (RLS) method, and based on the adaptive model, we develop an Internal Model Controller (IMC) strategy. From the simulation result, it shows that, with the adaptive PLS model updating algorithm, the mismatch can be minimized under the circumstances of gain changes or pole changes. And with the adaptive model, the IMC we design shows a good performance in the test.(2) Nonlinearity inherently is always a difficult problem in industry modeling and controlling, and it exists in nearly every practical process. The nonlinearity in multivariable system makes the problem even harder to solve. The Fuzzy modeling method is an effective approach to model the nonlinearity, by incorporating the fuzzy model into PLS method, we can establish the dynamic fuzzy PLS model to represent the dynamic and nonlinearity of process by turning the multivariable dynamic nonlinear modeling problem into several single-input single-output dynamic nonlinear modeling problem.(3)Based on the DFPLS model, we propose fuzzy PID control method under the control framework of PLS which is brought by Kasper/Ray. And the controller shows a good performance in the pH neutralization process. Meanwhile, according to the Internal Model Control structure, we propose a dynamic fuzzy PLS Internal Model Controller strategy, which based on the DFPLS prediction and the instant linear model. The control strategy shows a good performance on the pH neutralization process and the Aspen Dynamic Model Methylcyclohexane (MCH) distillation column.